Spaces:
Sleeping
Sleeping
Update streamlit_app.py
Browse files- streamlit_app.py +695 -438
streamlit_app.py
CHANGED
@@ -14,7 +14,8 @@ from plotly.subplots import make_subplots
|
|
14 |
import warnings
|
15 |
import datasets
|
16 |
import logging
|
17 |
-
|
|
|
18 |
|
19 |
# Configure logging
|
20 |
logging.basicConfig(level=logging.INFO)
|
@@ -23,41 +24,42 @@ logger = logging.getLogger(__name__)
|
|
23 |
# Constants
|
24 |
PLOT_PALETTE = {
|
25 |
"jailbreak": "#D000D8", # Purple
|
26 |
-
"benign": "#008393",
|
27 |
-
"control": "#EF0000",
|
28 |
}
|
29 |
|
|
|
30 |
# Utility functions
|
31 |
def load_and_prepare_dataset(dataset_config):
|
32 |
"""Load the risky conversations dataset and prepare it for analysis."""
|
33 |
logger.info("Loading dataset...")
|
34 |
-
|
35 |
dataset_name = dataset_config["dataset_name"]
|
36 |
logger.info(f"Loading dataset: {dataset_name}")
|
37 |
-
|
38 |
# Load the dataset
|
39 |
dataset = datasets.load_dataset(dataset_name, split="train")
|
40 |
logger.info(f"Dataset loaded with {len(dataset)} conversations")
|
41 |
-
|
42 |
# Convert to pandas
|
43 |
pandas_dataset = dataset.to_pandas()
|
44 |
-
|
45 |
# Explode the conversation column
|
46 |
pandas_dataset_exploded = pandas_dataset.explode("conversation")
|
47 |
pandas_dataset_exploded = pandas_dataset_exploded.reset_index(drop=True)
|
48 |
-
|
49 |
# Normalize conversation data
|
50 |
conversations_unfolded = pd.json_normalize(
|
51 |
pandas_dataset_exploded["conversation"],
|
52 |
)
|
53 |
conversations_unfolded = conversations_unfolded.add_prefix("turn.")
|
54 |
-
|
55 |
# Ensure there's a 'conversation_metrics' column, even if empty
|
56 |
if "conversation_metrics" not in pandas_dataset_exploded.columns:
|
57 |
pandas_dataset_exploded["conversation_metrics"] = [{}] * len(
|
58 |
pandas_dataset_exploded
|
59 |
)
|
60 |
-
|
61 |
# Normalize conversation metrics
|
62 |
conversations_metrics_unfolded = pd.json_normalize(
|
63 |
pandas_dataset_exploded["conversation_metrics"]
|
@@ -65,7 +67,7 @@ def load_and_prepare_dataset(dataset_config):
|
|
65 |
conversations_metrics_unfolded = conversations_metrics_unfolded.add_prefix(
|
66 |
"conversation_metrics."
|
67 |
)
|
68 |
-
|
69 |
# Concatenate all dataframes
|
70 |
pandas_dataset_exploded = pd.concat(
|
71 |
[
|
@@ -77,42 +79,41 @@ def load_and_prepare_dataset(dataset_config):
|
|
77 |
],
|
78 |
axis=1,
|
79 |
)
|
80 |
-
|
81 |
logger.info(f"Dataset prepared with {len(pandas_dataset_exploded)} turns")
|
82 |
return pandas_dataset, pandas_dataset_exploded
|
83 |
|
|
|
84 |
def get_available_turn_metrics(dataset_exploded):
|
85 |
"""Dynamically discover all available turn metrics from the dataset."""
|
86 |
# Find all columns that contain turn metrics
|
87 |
turn_metric_columns = [
|
88 |
col for col in dataset_exploded.columns if col.startswith("turn.turn_metrics.")
|
89 |
]
|
90 |
-
|
91 |
# Extract the metric names by removing the prefix
|
92 |
metrics = [col.replace("turn.turn_metrics.", "") for col in turn_metric_columns]
|
93 |
-
|
94 |
# Sort metrics for consistent ordering
|
95 |
metrics = sorted(metrics)
|
96 |
-
|
97 |
logger.info(f"Discovered {len(metrics)} available turn metrics")
|
98 |
return metrics
|
99 |
|
|
|
100 |
def get_human_friendly_metric_name(metric_name):
|
101 |
"""Convert a metric name to a human-friendly display name."""
|
102 |
# Mapping of metric names to human-friendly display names
|
103 |
metric_display_names = {
|
104 |
# Length metrics
|
105 |
"conversational_length_tokens": "Length in Tokens",
|
106 |
-
"conversational_length_words": "Length in Words",
|
107 |
"conversational_length_bytes": "Length in Bytes",
|
108 |
-
|
109 |
# Readability metrics
|
110 |
"readability_flesch_score": "Flesch Reading Ease",
|
111 |
"readability_standard_score": "Reading Grade Level",
|
112 |
-
|
113 |
# Compression metrics
|
114 |
"lzw_compression": "LZW Compression Ratio",
|
115 |
-
|
116 |
# Language model metrics
|
117 |
"ll_mean_logprobs": "Mean Log Probability",
|
118 |
"ll_max_logprobs": "Maximum Log Probability",
|
@@ -120,21 +121,17 @@ def get_human_friendly_metric_name(metric_name):
|
|
120 |
"rll_mean_logprobs": "Reverse Mean Log Probability",
|
121 |
"rll_max_logprobs": "Reverse Maximum Log Probability",
|
122 |
"rll_min_logprobs": "Reverse Minimum Log Probability",
|
123 |
-
|
124 |
# Lexical diversity
|
125 |
"type_token_ratio": "Type-Token Ratio",
|
126 |
-
|
127 |
# Language distribution
|
128 |
"zipf_score": "Zipf Distribution Score",
|
129 |
-
|
130 |
# Evaluation metrics
|
131 |
"rubric_evaluation.Stealth": "Stealth Score",
|
132 |
-
|
133 |
# Working memory metrics
|
134 |
"wm_total_entity_references": "Total Entity References",
|
135 |
"wm_unique_entities_count": "Unique Entities Count",
|
136 |
"wm_named_entity_mentions": "Named Entity Mentions",
|
137 |
-
"wm_concept_noun_mentions": "Concept Noun Mentions",
|
138 |
"wm_pronoun_references": "Pronoun References",
|
139 |
"wm_entity_density_per_word": "Entity Density per Word",
|
140 |
"wm_entity_density_per_100_words": "Entity Density per 100 Words",
|
@@ -143,7 +140,6 @@ def get_human_friendly_metric_name(metric_name):
|
|
143 |
"wm_entity_repetition_ratio": "Entity Repetition Ratio",
|
144 |
"wm_cognitive_load_score": "Cognitive Load Score",
|
145 |
"wm_high_cognitive_load": "High Cognitive Load",
|
146 |
-
|
147 |
# Discourse coherence metrics
|
148 |
"discourse_coherence_to_next_user": "Coherence to Next User Turn",
|
149 |
"discourse_coherence_to_next_turn": "Coherence to Next Turn",
|
@@ -152,164 +148,241 @@ def get_human_friendly_metric_name(metric_name):
|
|
152 |
"discourse_user_topic_drift": "User Topic Drift",
|
153 |
"discourse_user_entity_continuity": "User Entity Continuity",
|
154 |
"discourse_num_user_turns": "Number of User Turns",
|
155 |
-
|
156 |
# Tokens per byte
|
157 |
"tokens_per_byte": "Tokens per Byte",
|
158 |
}
|
159 |
-
|
160 |
# Check exact match first
|
161 |
if metric_name in metric_display_names:
|
162 |
return metric_display_names[metric_name]
|
163 |
-
|
164 |
# Handle conversation-level aggregations
|
165 |
-
for suffix in [
|
|
|
|
|
|
|
|
|
|
|
|
|
166 |
if metric_name.endswith(suffix):
|
167 |
-
base_metric = metric_name[
|
168 |
if base_metric in metric_display_names:
|
169 |
agg_type = suffix.split("_")[-1].title()
|
170 |
return f"{metric_display_names[base_metric]} ({agg_type})"
|
171 |
-
|
172 |
# Handle turn-level metrics with "turn.turn_metrics." prefix
|
173 |
if metric_name.startswith("turn.turn_metrics."):
|
174 |
-
base_metric = metric_name[len("turn.turn_metrics."):]
|
175 |
if base_metric in metric_display_names:
|
176 |
return metric_display_names[base_metric]
|
177 |
-
|
178 |
# Fallback: convert underscores to spaces and title case
|
179 |
clean_name = metric_name
|
180 |
for prefix in ["turn.turn_metrics.", "conversation_metrics.", "turn_metrics."]:
|
181 |
if clean_name.startswith(prefix):
|
182 |
-
clean_name = clean_name[len(prefix):]
|
183 |
break
|
184 |
-
|
185 |
# Convert to human-readable format
|
186 |
clean_name = clean_name.replace("_", " ").title()
|
187 |
return clean_name
|
188 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
189 |
# Setup page config
|
190 |
st.set_page_config(
|
191 |
page_title="Complexity Metrics Explorer",
|
192 |
page_icon="π",
|
193 |
layout="wide",
|
194 |
-
initial_sidebar_state="expanded"
|
195 |
)
|
196 |
|
|
|
197 |
# Cache data loading
|
198 |
@st.cache_data
|
199 |
def load_data(dataset_name):
|
200 |
"""Load and cache the dataset"""
|
201 |
-
df, df_exploded = load_and_prepare_dataset({
|
202 |
-
'dataset_name': dataset_name
|
203 |
-
})
|
204 |
return df, df_exploded
|
205 |
|
|
|
206 |
@st.cache_data
|
207 |
def get_metrics(df_exploded):
|
208 |
"""Get available metrics from the dataset"""
|
209 |
return get_available_turn_metrics(df_exploded)
|
210 |
|
|
|
211 |
def main():
|
212 |
st.title("π Complexity Metrics Explorer")
|
213 |
-
st.markdown(
|
214 |
-
|
|
|
|
|
215 |
# Dataset selection
|
216 |
st.sidebar.header("ποΈ Dataset Selection")
|
217 |
-
|
218 |
# Available datasets
|
219 |
available_datasets = [
|
220 |
"risky-conversations/jailbreaks_dataset_with_results_reduced",
|
221 |
"risky-conversations/jailbreaks_dataset_with_results",
|
222 |
"risky-conversations/jailbreaks_dataset_with_results_filtered_successful_jailbreak",
|
223 |
-
"Custom..."
|
224 |
]
|
225 |
-
|
226 |
selected_option = st.sidebar.selectbox(
|
227 |
"Select Dataset",
|
228 |
options=available_datasets,
|
229 |
index=0, # Default to reduced dataset
|
230 |
-
help="Choose which dataset to analyze"
|
231 |
)
|
232 |
-
|
233 |
# Handle custom dataset input
|
234 |
if selected_option == "Custom...":
|
235 |
selected_dataset = st.sidebar.text_input(
|
236 |
"Custom Dataset Name",
|
237 |
value="risky-conversations/jailbreaks_dataset_with_results_reduced",
|
238 |
-
help="Enter the full dataset name (e.g., 'risky-conversations/jailbreaks_dataset_with_results_reduced')"
|
239 |
)
|
240 |
if not selected_dataset.strip():
|
241 |
st.sidebar.warning("Please enter a dataset name")
|
242 |
st.stop()
|
243 |
else:
|
244 |
selected_dataset = selected_option
|
245 |
-
|
246 |
# Add refresh button
|
247 |
if st.sidebar.button("π Refresh Data", help="Clear cache and reload dataset"):
|
248 |
st.cache_data.clear()
|
249 |
st.rerun()
|
250 |
-
|
251 |
# Load data
|
252 |
with st.spinner(f"Loading dataset: {selected_dataset}..."):
|
253 |
try:
|
254 |
df, df_exploded = load_data(selected_dataset)
|
255 |
available_metrics = get_metrics(df_exploded)
|
256 |
-
|
257 |
# Display dataset info
|
258 |
col1, col2, col3, col4 = st.columns(4)
|
259 |
with col1:
|
260 |
-
st.metric("Dataset", selected_dataset.split(
|
261 |
with col2:
|
262 |
st.metric("Conversations", f"{len(df):,}")
|
263 |
with col3:
|
264 |
st.metric("Turns", f"{len(df_exploded):,}")
|
265 |
with col4:
|
266 |
st.metric("Metrics", len(available_metrics))
|
267 |
-
|
268 |
data_loaded = True
|
269 |
except Exception as e:
|
270 |
st.error(f"Error loading dataset: {e}")
|
271 |
st.info("Please check if the dataset exists and is accessible.")
|
272 |
-
st.info(
|
|
|
|
|
273 |
data_loaded = False
|
274 |
-
|
275 |
if not data_loaded:
|
276 |
st.stop()
|
277 |
-
|
278 |
# Sidebar controls
|
279 |
st.sidebar.header("ποΈ Controls")
|
280 |
-
|
281 |
# Dataset type filter
|
282 |
-
dataset_types = df[
|
283 |
selected_types = st.sidebar.multiselect(
|
284 |
"Select Dataset Types",
|
285 |
options=dataset_types,
|
286 |
default=dataset_types,
|
287 |
-
help="Filter by conversation type"
|
288 |
)
|
289 |
-
|
290 |
# Role filter
|
291 |
-
if
|
292 |
-
roles = df_exploded[
|
293 |
# Assert only user and assistant roles exist
|
294 |
-
expected_roles = {
|
295 |
actual_roles = set(roles)
|
296 |
-
assert actual_roles.issubset(
|
297 |
-
|
|
|
|
|
298 |
st.sidebar.subheader("π₯ Role Filter")
|
299 |
col1, col2 = st.sidebar.columns(2)
|
300 |
-
|
301 |
with col1:
|
302 |
include_user = st.checkbox("User", value=True, help="Include user turns")
|
303 |
with col2:
|
304 |
-
include_assistant = st.checkbox(
|
305 |
-
|
|
|
|
|
306 |
# Build selected roles list
|
307 |
selected_roles = []
|
308 |
-
if include_user and
|
309 |
-
selected_roles.append(
|
310 |
-
if include_assistant and
|
311 |
-
selected_roles.append(
|
312 |
-
|
313 |
# Show selection info
|
314 |
if selected_roles:
|
315 |
st.sidebar.success(f"Including: {', '.join(selected_roles)}")
|
@@ -317,82 +390,98 @@ def main():
|
|
317 |
st.sidebar.warning("No roles selected")
|
318 |
else:
|
319 |
selected_roles = None
|
320 |
-
|
321 |
# Filter data based on selections
|
322 |
-
filtered_df = df[df[
|
323 |
-
filtered_df_exploded =
|
324 |
-
|
325 |
-
|
326 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
327 |
elif selected_roles is not None and len(selected_roles) == 0:
|
328 |
# If roles exist but none are selected, show empty dataset
|
329 |
-
filtered_df_exploded = filtered_df_exploded.iloc[
|
330 |
-
|
|
|
|
|
331 |
# Check if we have data after filtering
|
332 |
if len(filtered_df_exploded) == 0:
|
333 |
-
st.error(
|
|
|
|
|
334 |
st.stop()
|
335 |
-
|
336 |
# Metric selection
|
337 |
st.sidebar.header("π Metrics")
|
338 |
-
|
339 |
# Dynamic metric categorization based on common patterns
|
340 |
def categorize_metrics(metrics):
|
341 |
"""Dynamically categorize metrics based on naming patterns"""
|
342 |
categories = {"All": metrics} # Always include all metrics
|
343 |
-
|
344 |
# Common patterns to look for
|
345 |
patterns = {
|
346 |
-
"Length": [
|
347 |
-
"Readability": [
|
348 |
-
"Compression": [
|
349 |
-
"Language Model": [
|
350 |
-
"Working Memory": [
|
351 |
-
"Discourse": [
|
352 |
-
"Evaluation": [
|
353 |
-
"Distribution": [
|
354 |
-
"Coherence": [
|
355 |
-
"Entity": [
|
356 |
-
"Cognitive": [
|
357 |
}
|
358 |
-
|
359 |
# Categorize metrics
|
360 |
for category, keywords in patterns.items():
|
361 |
-
matching_metrics = [
|
|
|
|
|
362 |
if matching_metrics:
|
363 |
categories[category] = matching_metrics
|
364 |
-
|
365 |
# Find uncategorized metrics
|
366 |
categorized = set()
|
367 |
for cat_metrics in categories.values():
|
368 |
if cat_metrics != metrics: # Skip "All" category
|
369 |
categorized.update(cat_metrics)
|
370 |
-
|
371 |
uncategorized = [m for m in metrics if m not in categorized]
|
372 |
if uncategorized:
|
373 |
categories["Other"] = uncategorized
|
374 |
-
|
375 |
return categories
|
376 |
-
|
377 |
metric_categories = categorize_metrics(available_metrics)
|
378 |
-
|
379 |
# Metric selection interface
|
380 |
selection_mode = st.sidebar.radio(
|
381 |
"Selection Mode",
|
382 |
["By Category", "Search/Filter", "Select All"],
|
383 |
-
help="Choose how to select metrics"
|
384 |
)
|
385 |
-
|
386 |
if selection_mode == "By Category":
|
387 |
selected_category = st.sidebar.selectbox(
|
388 |
-
"Metric Category",
|
389 |
options=list(metric_categories.keys()),
|
390 |
-
help=f"Found {len(metric_categories)} categories"
|
391 |
)
|
392 |
-
|
393 |
available_in_category = metric_categories[selected_category]
|
394 |
-
default_selection =
|
395 |
-
|
|
|
|
|
|
|
|
|
396 |
# Add select all button for category
|
397 |
col1, col2 = st.sidebar.columns(2)
|
398 |
with col1:
|
@@ -401,33 +490,35 @@ def main():
|
|
401 |
with col2:
|
402 |
if st.button("Clear All", key="clear_all_category"):
|
403 |
st.session_state.selected_metrics_category = []
|
404 |
-
|
405 |
# Use session state for persistence
|
406 |
if "selected_metrics_category" not in st.session_state:
|
407 |
st.session_state.selected_metrics_category = default_selection
|
408 |
-
|
409 |
selected_metrics = st.sidebar.multiselect(
|
410 |
f"Select Metrics ({len(available_in_category)} available)",
|
411 |
options=available_in_category,
|
412 |
default=st.session_state.selected_metrics_category,
|
413 |
key="metrics_multiselect_category",
|
414 |
-
help="Choose metrics to visualize"
|
415 |
)
|
416 |
-
|
417 |
elif selection_mode == "Search/Filter":
|
418 |
search_term = st.sidebar.text_input(
|
419 |
"Search Metrics",
|
420 |
placeholder="Enter keywords to filter metrics...",
|
421 |
-
help="Search for metrics containing specific terms"
|
422 |
)
|
423 |
-
|
424 |
if search_term:
|
425 |
-
filtered_metrics = [
|
|
|
|
|
426 |
else:
|
427 |
filtered_metrics = available_metrics
|
428 |
-
|
429 |
st.sidebar.write(f"Found {len(filtered_metrics)} metrics")
|
430 |
-
|
431 |
# Add select all button for search results
|
432 |
col1, col2 = st.sidebar.columns(2)
|
433 |
with col1:
|
@@ -436,19 +527,23 @@ def main():
|
|
436 |
with col2:
|
437 |
if st.button("Clear All", key="clear_all_search"):
|
438 |
st.session_state.selected_metrics_search = []
|
439 |
-
|
440 |
# Use session state for persistence
|
441 |
if "selected_metrics_search" not in st.session_state:
|
442 |
-
st.session_state.selected_metrics_search =
|
443 |
-
|
|
|
|
|
|
|
|
|
444 |
selected_metrics = st.sidebar.multiselect(
|
445 |
"Select Metrics",
|
446 |
options=filtered_metrics,
|
447 |
default=st.session_state.selected_metrics_search,
|
448 |
key="metrics_multiselect_search",
|
449 |
-
help="Choose metrics to visualize"
|
450 |
)
|
451 |
-
|
452 |
else: # Select All
|
453 |
# Add select all button for all metrics
|
454 |
col1, col2 = st.sidebar.columns(2)
|
@@ -458,262 +553,309 @@ def main():
|
|
458 |
with col2:
|
459 |
if st.button("Clear All", key="clear_all_all"):
|
460 |
st.session_state.selected_metrics_all = []
|
461 |
-
|
462 |
# Use session state for persistence
|
463 |
if "selected_metrics_all" not in st.session_state:
|
464 |
-
st.session_state.selected_metrics_all = available_metrics[
|
465 |
-
|
|
|
|
|
466 |
selected_metrics = st.sidebar.multiselect(
|
467 |
f"All Metrics ({len(available_metrics)} total)",
|
468 |
options=available_metrics,
|
469 |
default=st.session_state.selected_metrics_all,
|
470 |
key="metrics_multiselect_all",
|
471 |
-
help="All available metrics - be careful with performance for large selections"
|
472 |
)
|
473 |
-
|
474 |
# Show selection summary
|
475 |
if selected_metrics:
|
476 |
st.sidebar.success(f"Selected {len(selected_metrics)} metrics")
|
477 |
-
|
478 |
# Performance warning for large selections
|
479 |
if len(selected_metrics) > 20:
|
480 |
-
st.sidebar.warning(
|
|
|
|
|
481 |
elif len(selected_metrics) > 50:
|
482 |
-
st.sidebar.error(
|
|
|
|
|
483 |
else:
|
484 |
st.sidebar.warning("No metrics selected")
|
485 |
-
|
486 |
# Metric info expander
|
487 |
with st.sidebar.expander("βΉοΈ Metric Information", expanded=False):
|
488 |
st.write(f"**Total Available Metrics:** {len(available_metrics)}")
|
489 |
st.write(f"**Categories Found:** {len(metric_categories)}")
|
490 |
-
|
491 |
if st.checkbox("Show all metric names", key="show_all_metrics"):
|
492 |
st.write("**All Available Metrics:**")
|
493 |
for i, metric in enumerate(available_metrics, 1):
|
494 |
st.write(f"{i}. `{metric}`")
|
495 |
-
|
496 |
# Main content tabs
|
497 |
-
tab1, tab2, tab3, tab4, tab5 = st.tabs(
|
498 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
499 |
with tab1:
|
500 |
st.header("Distribution Analysis")
|
501 |
-
|
502 |
if not selected_metrics:
|
503 |
st.warning("Please select at least one metric to visualize.")
|
504 |
return
|
505 |
-
|
506 |
-
# Create
|
507 |
-
|
508 |
-
|
509 |
-
|
510 |
-
|
511 |
-
|
512 |
-
|
513 |
-
|
514 |
-
st.
|
515 |
-
|
516 |
-
|
517 |
-
|
518 |
-
|
519 |
-
|
520 |
-
|
521 |
-
|
522 |
-
|
523 |
-
|
524 |
-
|
525 |
-
|
526 |
-
|
527 |
-
|
528 |
-
|
529 |
-
|
530 |
-
|
531 |
-
|
532 |
-
|
533 |
-
|
534 |
-
)
|
535 |
-
|
536 |
-
fig.update_layout(
|
537 |
-
xaxis_title=get_human_friendly_metric_name(metric),
|
538 |
-
yaxis_title="Count",
|
539 |
-
height=400
|
540 |
-
)
|
541 |
-
|
542 |
-
st.plotly_chart(fig, use_container_width=True)
|
543 |
-
|
544 |
-
# Summary statistics
|
545 |
-
col1, col2 = st.columns(2)
|
546 |
-
|
547 |
-
with col1:
|
548 |
-
st.write("**Summary Statistics**")
|
549 |
-
summary_stats = metric_data.groupby('type')[full_metric_name].agg(['count', 'mean', 'std', 'min', 'max']).round(3)
|
550 |
-
st.dataframe(summary_stats)
|
551 |
-
|
552 |
-
with col2:
|
553 |
-
st.write("**Percentiles**")
|
554 |
-
percentiles = metric_data.groupby('type')[full_metric_name].quantile([0.25, 0.5, 0.75]).unstack().round(3)
|
555 |
-
percentiles.columns = ['25%', '50%', '75%']
|
556 |
-
st.dataframe(percentiles)
|
557 |
-
|
558 |
with tab2:
|
559 |
st.header("Correlation Analysis")
|
560 |
-
|
561 |
if len(selected_metrics) < 2:
|
562 |
st.warning("Please select at least 2 metrics for correlation analysis.")
|
563 |
else:
|
564 |
-
#
|
565 |
-
|
566 |
-
|
567 |
-
|
568 |
-
# Clean column names for display
|
569 |
-
corr_data.columns = [get_human_friendly_metric_name(col.replace('turn.turn_metrics.', '')) if col.startswith('turn.turn_metrics.') else col for col in corr_data.columns]
|
570 |
-
|
571 |
-
# Calculate correlation matrix
|
572 |
-
corr_matrix = corr_data.select_dtypes(include=[np.number]).corr()
|
573 |
-
|
574 |
-
# Create correlation heatmap
|
575 |
-
fig = px.imshow(
|
576 |
-
corr_matrix,
|
577 |
-
text_auto=True,
|
578 |
-
aspect="auto",
|
579 |
-
title="Correlation Matrix",
|
580 |
-
color_continuous_scale='RdBu_r',
|
581 |
-
zmin=-1, zmax=1
|
582 |
)
|
583 |
-
|
584 |
-
|
585 |
-
|
586 |
-
|
587 |
-
|
588 |
-
|
589 |
-
|
590 |
-
|
591 |
-
|
592 |
-
|
593 |
-
|
594 |
-
corr_val = corr_matrix.iloc[i, j]
|
595 |
-
if abs(corr_val) > 0.7:
|
596 |
-
strong_corrs.append((corr_matrix.columns[i], corr_matrix.columns[j], corr_val))
|
597 |
-
|
598 |
-
if strong_corrs:
|
599 |
-
for metric1, metric2, corr_val in strong_corrs[:3]: # Show top 3
|
600 |
-
fig = px.scatter(
|
601 |
-
corr_data,
|
602 |
-
x=metric1,
|
603 |
-
y=metric2,
|
604 |
-
color='type',
|
605 |
-
title=f"{metric1} vs {metric2} (r={corr_val:.3f})",
|
606 |
-
color_discrete_map=PLOT_PALETTE if len(selected_types) <= 3 else None,
|
607 |
-
opacity=0.6
|
608 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
609 |
st.plotly_chart(fig, use_container_width=True)
|
610 |
-
|
611 |
-
|
612 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
613 |
with tab3:
|
614 |
st.header("Type Comparisons")
|
615 |
-
|
616 |
if not selected_metrics:
|
617 |
st.warning("Please select at least one metric to compare.")
|
618 |
else:
|
619 |
# Box plots for each metric
|
620 |
for metric in selected_metrics:
|
621 |
full_metric_name = f"turn.turn_metrics.{metric}"
|
622 |
-
|
623 |
if full_metric_name not in filtered_df_exploded.columns:
|
624 |
continue
|
625 |
-
|
626 |
st.subheader(f"π¦ {get_human_friendly_metric_name(metric)} by Type")
|
627 |
-
|
628 |
# Create box plot
|
629 |
fig = px.box(
|
630 |
filtered_df_exploded.dropna(subset=[full_metric_name]),
|
631 |
-
x=
|
632 |
y=full_metric_name,
|
633 |
title=f"Distribution of {get_human_friendly_metric_name(metric)} by Type",
|
634 |
-
color=
|
635 |
-
color_discrete_map=
|
|
|
|
|
636 |
)
|
637 |
-
|
638 |
fig.update_layout(
|
639 |
xaxis_title="Dataset Type",
|
640 |
yaxis_title=get_human_friendly_metric_name(metric),
|
641 |
-
height=400
|
642 |
)
|
643 |
-
|
644 |
st.plotly_chart(fig, use_container_width=True)
|
645 |
-
|
646 |
with tab4:
|
647 |
st.header("Individual Conversation Analysis")
|
648 |
-
|
649 |
# Conversation selector
|
650 |
st.subheader("π Select Conversation")
|
651 |
-
|
652 |
# Get unique conversations with some metadata
|
653 |
conversation_info = []
|
654 |
for idx, row in filtered_df.iterrows():
|
655 |
-
conv_type = row[
|
656 |
# Get basic info about the conversation
|
657 |
-
conv_turns = len(row.get(
|
658 |
-
conversation_info.append(
|
659 |
-
|
660 |
-
|
661 |
-
|
662 |
-
|
663 |
-
|
664 |
-
|
|
|
|
|
665 |
# Sort by type and number of turns for better organization
|
666 |
-
conversation_info = sorted(
|
667 |
-
|
|
|
|
|
668 |
# Conversation selection
|
669 |
col1, col2 = st.columns([3, 1])
|
670 |
-
|
671 |
with col1:
|
672 |
selected_conv_display = st.selectbox(
|
673 |
"Choose a conversation to analyze",
|
674 |
-
options=[conv[
|
675 |
-
help="Select a conversation to view detailed metrics and content"
|
676 |
)
|
677 |
-
|
678 |
with col2:
|
679 |
if st.button("π² Random", help="Select a random conversation"):
|
680 |
import random
|
681 |
-
|
|
|
|
|
|
|
682 |
st.rerun()
|
683 |
-
|
684 |
# Get the selected conversation data
|
685 |
-
selected_conv_info = next(
|
686 |
-
|
|
|
|
|
|
|
|
|
687 |
selected_conversation = filtered_df.iloc[selected_idx]
|
688 |
-
|
689 |
# Display conversation metadata
|
690 |
st.subheader("π Conversation Overview")
|
691 |
-
|
692 |
# First row - basic info
|
693 |
col1, col2, col3, col4 = st.columns(4)
|
694 |
with col1:
|
695 |
-
st.metric("Type", selected_conversation[
|
696 |
with col2:
|
697 |
st.metric("Index", selected_idx)
|
698 |
with col3:
|
699 |
-
st.metric("Total Turns", len(selected_conversation.get(
|
700 |
with col4:
|
701 |
# Count user vs assistant turns
|
702 |
-
roles = [
|
703 |
-
|
704 |
-
|
|
|
|
|
|
|
705 |
st.metric("User/Assistant", f"{user_turns}/{assistant_turns}")
|
706 |
-
|
707 |
# Second row - additional metadata
|
708 |
col1, col2, col3 = st.columns(3)
|
709 |
with col1:
|
710 |
-
provenance = selected_conversation.get(
|
711 |
st.metric("Dataset Source", provenance)
|
712 |
with col2:
|
713 |
-
language = selected_conversation.get(
|
714 |
-
st.metric("Language", language.upper() if language else
|
715 |
with col3:
|
716 |
-
timestamp = selected_conversation.get(
|
717 |
if timestamp:
|
718 |
# Handle different timestamp formats
|
719 |
if isinstance(timestamp, str):
|
@@ -722,139 +864,184 @@ def main():
|
|
722 |
st.metric("Timestamp", str(timestamp))
|
723 |
else:
|
724 |
st.metric("Timestamp", "Not Available")
|
725 |
-
|
726 |
# Add toxicity summary
|
727 |
-
conversation_turns_temp = selected_conversation.get(
|
728 |
-
if hasattr(conversation_turns_temp,
|
729 |
conversation_turns_temp = conversation_turns_temp.tolist()
|
730 |
elif conversation_turns_temp is None:
|
731 |
conversation_turns_temp = []
|
732 |
-
|
733 |
if len(conversation_turns_temp) > 0:
|
734 |
# Calculate overall toxicity statistics
|
735 |
all_toxicities = []
|
736 |
for turn in conversation_turns_temp:
|
737 |
-
toxicities = turn.get(
|
738 |
-
if toxicities and
|
739 |
-
all_toxicities.append(toxicities[
|
740 |
-
|
741 |
if all_toxicities:
|
742 |
avg_toxicity = sum(all_toxicities) / len(all_toxicities)
|
743 |
max_toxicity = max(all_toxicities)
|
744 |
-
|
745 |
st.markdown("**π Toxicity Summary:**")
|
746 |
col1, col2, col3 = st.columns(3)
|
747 |
with col1:
|
748 |
# Color code average toxicity
|
749 |
if avg_toxicity > 0.5:
|
750 |
-
st.metric(
|
|
|
|
|
|
|
|
|
|
|
751 |
elif avg_toxicity > 0.1:
|
752 |
-
st.metric(
|
|
|
|
|
|
|
|
|
|
|
753 |
else:
|
754 |
-
st.metric(
|
755 |
-
|
|
|
|
|
|
|
|
|
|
|
756 |
with col2:
|
757 |
# Color code max toxicity
|
758 |
if max_toxicity > 0.5:
|
759 |
-
st.metric(
|
|
|
|
|
|
|
|
|
|
|
760 |
elif max_toxicity > 0.1:
|
761 |
-
st.metric(
|
|
|
|
|
|
|
|
|
|
|
762 |
else:
|
763 |
-
st.metric(
|
764 |
-
|
|
|
|
|
|
|
|
|
|
|
765 |
with col3:
|
766 |
high_tox_turns = sum(1 for t in all_toxicities if t > 0.5)
|
767 |
st.metric("High Toxicity Turns", high_tox_turns)
|
768 |
-
|
769 |
# Get conversation turns with metrics
|
770 |
-
conv_turns_data = filtered_df_exploded[
|
771 |
-
filtered_df_exploded
|
772 |
-
|
773 |
-
|
774 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
775 |
# Alternative approach: filter by matching all conversation data
|
776 |
# This is more reliable but less efficient
|
777 |
conv_turns_data = []
|
778 |
start_idx = None
|
779 |
for idx, row in filtered_df_exploded.iterrows():
|
780 |
# Check if this row belongs to our selected conversation
|
781 |
-
if (
|
782 |
-
|
783 |
-
row
|
|
|
|
|
784 |
# This is a simplified approach - in reality you'd need better conversation matching
|
785 |
pass
|
786 |
-
|
787 |
# Simpler approach: get all turns from the conversation directly
|
788 |
-
conversation_turns = selected_conversation.get(
|
789 |
-
|
790 |
# Ensure conversation_turns is a list and handle different data types
|
791 |
-
if hasattr(conversation_turns,
|
792 |
conversation_turns = conversation_turns.tolist()
|
793 |
elif conversation_turns is None:
|
794 |
conversation_turns = []
|
795 |
-
|
796 |
if len(conversation_turns) > 0:
|
797 |
# Display conversation content with metrics
|
798 |
st.subheader("π¬ Conversation with Metrics")
|
799 |
-
|
800 |
# Get actual turn-level data for this conversation
|
801 |
turn_metric_columns = [f"turn.turn_metrics.{m}" for m in selected_metrics]
|
802 |
-
available_columns = [
|
803 |
-
|
|
|
|
|
|
|
|
|
804 |
# Get sample metrics for this conversation type (since exact matching is complex)
|
805 |
sample_metrics = None
|
806 |
if available_columns:
|
807 |
-
type_turns = filtered_df_exploded[
|
|
|
|
|
808 |
sample_size = min(len(conversation_turns), len(type_turns))
|
809 |
if sample_size > 0:
|
810 |
sample_metrics = type_turns.head(sample_size)
|
811 |
-
|
812 |
# Display each turn with its metrics
|
813 |
for i, turn in enumerate(conversation_turns):
|
814 |
-
role = turn.get(
|
815 |
-
content = turn.get(
|
816 |
-
|
817 |
# Display turn content with role styling
|
818 |
-
if role ==
|
819 |
st.markdown(f"**π€ User (Turn {i+1}):**")
|
820 |
st.info(content)
|
821 |
-
elif role ==
|
822 |
st.markdown(f"**π€ Assistant (Turn {i+1}):**")
|
823 |
st.success(content)
|
824 |
else:
|
825 |
st.markdown(f"**β {role.title()} (Turn {i+1}):**")
|
826 |
st.warning(content)
|
827 |
-
|
828 |
# Display metrics for this turn
|
829 |
if sample_metrics is not None and i < len(sample_metrics):
|
830 |
turn_row = sample_metrics.iloc[i]
|
831 |
-
|
832 |
# Create metrics display
|
833 |
metrics_for_turn = {}
|
834 |
for col in available_columns:
|
835 |
-
metric_name = col.replace(
|
836 |
friendly_name = get_human_friendly_metric_name(metric_name)
|
837 |
-
value = turn_row.get(col,
|
838 |
if pd.notna(value) and isinstance(value, (int, float)):
|
839 |
metrics_for_turn[friendly_name] = round(value, 3)
|
840 |
else:
|
841 |
-
metrics_for_turn[friendly_name] =
|
842 |
-
|
843 |
# Add toxicity metrics if available
|
844 |
-
toxicities = turn.get(
|
845 |
if toxicities:
|
846 |
st.markdown("**π Toxicity Scores:**")
|
847 |
tox_cols = st.columns(4)
|
848 |
tox_metrics = [
|
849 |
-
(
|
850 |
-
(
|
851 |
-
(
|
852 |
-
(
|
853 |
-
(
|
854 |
-
(
|
855 |
-
(
|
856 |
]
|
857 |
-
|
858 |
for idx, (tox_key, tox_name) in enumerate(tox_metrics):
|
859 |
if tox_key in toxicities:
|
860 |
col_idx = idx % 4
|
@@ -863,14 +1050,29 @@ def main():
|
|
863 |
if isinstance(tox_value, (int, float)):
|
864 |
# Color code based on toxicity level
|
865 |
if tox_value > 0.5:
|
866 |
-
st.metric(
|
|
|
|
|
|
|
|
|
|
|
867 |
elif tox_value > 0.1:
|
868 |
-
st.metric(
|
|
|
|
|
|
|
|
|
|
|
869 |
else:
|
870 |
-
st.metric(
|
|
|
|
|
|
|
|
|
|
|
871 |
else:
|
872 |
st.metric(tox_name, str(tox_value))
|
873 |
-
|
874 |
# Display complexity metrics
|
875 |
if metrics_for_turn:
|
876 |
st.markdown("**π Complexity Metrics:**")
|
@@ -878,29 +1080,34 @@ def main():
|
|
878 |
num_cols = min(4, len(metrics_for_turn))
|
879 |
if num_cols > 0:
|
880 |
cols = st.columns(num_cols)
|
881 |
-
for idx, (metric_name, value) in enumerate(
|
|
|
|
|
882 |
col_idx = idx % num_cols
|
883 |
with cols[col_idx]:
|
884 |
-
if
|
|
|
|
|
|
|
885 |
st.metric(metric_name, value)
|
886 |
else:
|
887 |
st.metric(metric_name, str(value))
|
888 |
else:
|
889 |
# Show toxicity even when no complexity metrics available
|
890 |
-
toxicities = turn.get(
|
891 |
if toxicities:
|
892 |
st.markdown("**π Toxicity Scores:**")
|
893 |
tox_cols = st.columns(4)
|
894 |
tox_metrics = [
|
895 |
-
(
|
896 |
-
(
|
897 |
-
(
|
898 |
-
(
|
899 |
-
(
|
900 |
-
(
|
901 |
-
(
|
902 |
]
|
903 |
-
|
904 |
for idx, (tox_key, tox_name) in enumerate(tox_metrics):
|
905 |
if tox_key in toxicities:
|
906 |
col_idx = idx % 4
|
@@ -909,14 +1116,29 @@ def main():
|
|
909 |
if isinstance(tox_value, (int, float)):
|
910 |
# Color code based on toxicity level
|
911 |
if tox_value > 0.5:
|
912 |
-
st.metric(
|
|
|
|
|
|
|
|
|
|
|
913 |
elif tox_value > 0.1:
|
914 |
-
st.metric(
|
|
|
|
|
|
|
|
|
|
|
915 |
else:
|
916 |
-
st.metric(
|
|
|
|
|
|
|
|
|
|
|
917 |
else:
|
918 |
st.metric(tox_name, str(tox_value))
|
919 |
-
|
920 |
# Show basic turn statistics when no complexity metrics available
|
921 |
st.markdown("**π Basic Statistics:**")
|
922 |
col1, col2, col3 = st.columns(3)
|
@@ -926,21 +1148,21 @@ def main():
|
|
926 |
st.metric("Words", len(content.split()))
|
927 |
with col3:
|
928 |
st.metric("Role", role.title())
|
929 |
-
|
930 |
# Add separator between turns
|
931 |
st.divider()
|
932 |
-
|
933 |
# Plot metrics over turns with real data if available
|
934 |
if available_columns and sample_metrics is not None:
|
935 |
st.subheader("π Metrics Over Turns")
|
936 |
-
|
937 |
fig = go.Figure()
|
938 |
-
|
939 |
# Add traces for each selected metric (real data)
|
940 |
for col in available_columns[:5]: # Limit to first 5 for readability
|
941 |
-
metric_name = col.replace(
|
942 |
friendly_name = get_human_friendly_metric_name(metric_name)
|
943 |
-
|
944 |
# Get values for this metric
|
945 |
y_values = []
|
946 |
for _, turn_row in sample_metrics.iterrows():
|
@@ -949,101 +1171,136 @@ def main():
|
|
949 |
y_values.append(value)
|
950 |
else:
|
951 |
y_values.append(None)
|
952 |
-
|
953 |
if any(v is not None for v in y_values):
|
954 |
-
fig.add_trace(
|
955 |
-
|
956 |
-
|
957 |
-
|
958 |
-
|
959 |
-
|
960 |
-
|
961 |
-
|
962 |
-
|
963 |
-
|
|
|
|
|
964 |
if fig.data: # Only show if we have data
|
965 |
fig.update_layout(
|
966 |
title="Complexity Metrics Across Conversation Turns",
|
967 |
xaxis_title="Turn Number",
|
968 |
yaxis_title="Metric Value",
|
969 |
height=400,
|
970 |
-
hovermode=
|
971 |
)
|
972 |
-
|
973 |
st.plotly_chart(fig, use_container_width=True)
|
974 |
else:
|
975 |
-
st.info(
|
976 |
-
|
|
|
|
|
977 |
elif selected_metrics:
|
978 |
-
st.info(
|
|
|
|
|
979 |
else:
|
980 |
st.warning("Select some metrics to see detailed turn-level analysis.")
|
981 |
-
|
982 |
else:
|
983 |
st.warning("No conversation data available for the selected conversation.")
|
984 |
-
|
985 |
with tab5:
|
986 |
st.header("Detailed View")
|
987 |
-
|
988 |
-
#
|
989 |
-
st.
|
990 |
-
|
991 |
-
st.
|
992 |
-
|
993 |
-
col1, col2, col3 = st.columns(3)
|
994 |
-
|
995 |
with col1:
|
996 |
-
st.
|
997 |
-
|
998 |
-
|
999 |
-
st.metric("Total Turns", len(filtered_df_exploded))
|
1000 |
-
|
1001 |
-
with col3:
|
1002 |
-
st.metric("Available Metrics", len(available_metrics))
|
1003 |
-
|
1004 |
-
# Type distribution
|
1005 |
-
st.subheader("π Type Distribution")
|
1006 |
-
type_counts = filtered_df['type'].value_counts()
|
1007 |
-
|
1008 |
-
fig = px.pie(
|
1009 |
-
values=type_counts.values,
|
1010 |
-
names=type_counts.index,
|
1011 |
-
title="Distribution of Conversation Types",
|
1012 |
-
color_discrete_map=PLOT_PALETTE if len(type_counts) <= 3 else None
|
1013 |
-
)
|
1014 |
-
|
1015 |
-
st.plotly_chart(fig, use_container_width=True)
|
1016 |
-
|
1017 |
-
# Sample data
|
1018 |
-
st.subheader("π Sample Data")
|
1019 |
-
|
1020 |
-
if st.checkbox("Show raw data sample"):
|
1021 |
-
sample_cols = ['type'] + [f"turn.turn_metrics.{m}" for m in selected_metrics if f"turn.turn_metrics.{m}" in filtered_df_exploded.columns]
|
1022 |
-
sample_data = filtered_df_exploded[sample_cols].head(100)
|
1023 |
-
st.dataframe(sample_data)
|
1024 |
-
|
1025 |
-
# Metric availability
|
1026 |
-
st.subheader("π Metric Availability")
|
1027 |
-
|
1028 |
-
metric_completeness = {}
|
1029 |
-
for metric in selected_metrics:
|
1030 |
-
full_metric_name = f"turn.turn_metrics.{metric}"
|
1031 |
-
if full_metric_name in filtered_df_exploded.columns:
|
1032 |
-
completeness = (1 - filtered_df_exploded[full_metric_name].isna().sum() / len(filtered_df_exploded)) * 100
|
1033 |
-
metric_completeness[get_human_friendly_metric_name(metric)] = completeness
|
1034 |
-
|
1035 |
-
if metric_completeness:
|
1036 |
-
completeness_df = pd.DataFrame(list(metric_completeness.items()), columns=['Metric', 'Completeness (%)'])
|
1037 |
-
fig = px.bar(
|
1038 |
-
completeness_df,
|
1039 |
-
x='Metric',
|
1040 |
-
y='Completeness (%)',
|
1041 |
-
title="Data Completeness by Metric",
|
1042 |
-
color='Completeness (%)',
|
1043 |
-
color_continuous_scale='Viridis'
|
1044 |
)
|
1045 |
-
|
1046 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1047 |
|
1048 |
if __name__ == "__main__":
|
1049 |
main()
|
|
|
14 |
import warnings
|
15 |
import datasets
|
16 |
import logging
|
17 |
+
|
18 |
+
warnings.filterwarnings("ignore")
|
19 |
|
20 |
# Configure logging
|
21 |
logging.basicConfig(level=logging.INFO)
|
|
|
24 |
# Constants
|
25 |
PLOT_PALETTE = {
|
26 |
"jailbreak": "#D000D8", # Purple
|
27 |
+
"benign": "#008393", # Cyan
|
28 |
+
"control": "#EF0000", # Red
|
29 |
}
|
30 |
|
31 |
+
|
32 |
# Utility functions
|
33 |
def load_and_prepare_dataset(dataset_config):
|
34 |
"""Load the risky conversations dataset and prepare it for analysis."""
|
35 |
logger.info("Loading dataset...")
|
36 |
+
|
37 |
dataset_name = dataset_config["dataset_name"]
|
38 |
logger.info(f"Loading dataset: {dataset_name}")
|
39 |
+
|
40 |
# Load the dataset
|
41 |
dataset = datasets.load_dataset(dataset_name, split="train")
|
42 |
logger.info(f"Dataset loaded with {len(dataset)} conversations")
|
43 |
+
|
44 |
# Convert to pandas
|
45 |
pandas_dataset = dataset.to_pandas()
|
46 |
+
|
47 |
# Explode the conversation column
|
48 |
pandas_dataset_exploded = pandas_dataset.explode("conversation")
|
49 |
pandas_dataset_exploded = pandas_dataset_exploded.reset_index(drop=True)
|
50 |
+
|
51 |
# Normalize conversation data
|
52 |
conversations_unfolded = pd.json_normalize(
|
53 |
pandas_dataset_exploded["conversation"],
|
54 |
)
|
55 |
conversations_unfolded = conversations_unfolded.add_prefix("turn.")
|
56 |
+
|
57 |
# Ensure there's a 'conversation_metrics' column, even if empty
|
58 |
if "conversation_metrics" not in pandas_dataset_exploded.columns:
|
59 |
pandas_dataset_exploded["conversation_metrics"] = [{}] * len(
|
60 |
pandas_dataset_exploded
|
61 |
)
|
62 |
+
|
63 |
# Normalize conversation metrics
|
64 |
conversations_metrics_unfolded = pd.json_normalize(
|
65 |
pandas_dataset_exploded["conversation_metrics"]
|
|
|
67 |
conversations_metrics_unfolded = conversations_metrics_unfolded.add_prefix(
|
68 |
"conversation_metrics."
|
69 |
)
|
70 |
+
|
71 |
# Concatenate all dataframes
|
72 |
pandas_dataset_exploded = pd.concat(
|
73 |
[
|
|
|
79 |
],
|
80 |
axis=1,
|
81 |
)
|
82 |
+
|
83 |
logger.info(f"Dataset prepared with {len(pandas_dataset_exploded)} turns")
|
84 |
return pandas_dataset, pandas_dataset_exploded
|
85 |
|
86 |
+
|
87 |
def get_available_turn_metrics(dataset_exploded):
|
88 |
"""Dynamically discover all available turn metrics from the dataset."""
|
89 |
# Find all columns that contain turn metrics
|
90 |
turn_metric_columns = [
|
91 |
col for col in dataset_exploded.columns if col.startswith("turn.turn_metrics.")
|
92 |
]
|
93 |
+
|
94 |
# Extract the metric names by removing the prefix
|
95 |
metrics = [col.replace("turn.turn_metrics.", "") for col in turn_metric_columns]
|
96 |
+
|
97 |
# Sort metrics for consistent ordering
|
98 |
metrics = sorted(metrics)
|
99 |
+
|
100 |
logger.info(f"Discovered {len(metrics)} available turn metrics")
|
101 |
return metrics
|
102 |
|
103 |
+
|
104 |
def get_human_friendly_metric_name(metric_name):
|
105 |
"""Convert a metric name to a human-friendly display name."""
|
106 |
# Mapping of metric names to human-friendly display names
|
107 |
metric_display_names = {
|
108 |
# Length metrics
|
109 |
"conversational_length_tokens": "Length in Tokens",
|
110 |
+
"conversational_length_words": "Length in Words",
|
111 |
"conversational_length_bytes": "Length in Bytes",
|
|
|
112 |
# Readability metrics
|
113 |
"readability_flesch_score": "Flesch Reading Ease",
|
114 |
"readability_standard_score": "Reading Grade Level",
|
|
|
115 |
# Compression metrics
|
116 |
"lzw_compression": "LZW Compression Ratio",
|
|
|
117 |
# Language model metrics
|
118 |
"ll_mean_logprobs": "Mean Log Probability",
|
119 |
"ll_max_logprobs": "Maximum Log Probability",
|
|
|
121 |
"rll_mean_logprobs": "Reverse Mean Log Probability",
|
122 |
"rll_max_logprobs": "Reverse Maximum Log Probability",
|
123 |
"rll_min_logprobs": "Reverse Minimum Log Probability",
|
|
|
124 |
# Lexical diversity
|
125 |
"type_token_ratio": "Type-Token Ratio",
|
|
|
126 |
# Language distribution
|
127 |
"zipf_score": "Zipf Distribution Score",
|
|
|
128 |
# Evaluation metrics
|
129 |
"rubric_evaluation.Stealth": "Stealth Score",
|
|
|
130 |
# Working memory metrics
|
131 |
"wm_total_entity_references": "Total Entity References",
|
132 |
"wm_unique_entities_count": "Unique Entities Count",
|
133 |
"wm_named_entity_mentions": "Named Entity Mentions",
|
134 |
+
"wm_concept_noun_mentions": "Concept Noun Mentions",
|
135 |
"wm_pronoun_references": "Pronoun References",
|
136 |
"wm_entity_density_per_word": "Entity Density per Word",
|
137 |
"wm_entity_density_per_100_words": "Entity Density per 100 Words",
|
|
|
140 |
"wm_entity_repetition_ratio": "Entity Repetition Ratio",
|
141 |
"wm_cognitive_load_score": "Cognitive Load Score",
|
142 |
"wm_high_cognitive_load": "High Cognitive Load",
|
|
|
143 |
# Discourse coherence metrics
|
144 |
"discourse_coherence_to_next_user": "Coherence to Next User Turn",
|
145 |
"discourse_coherence_to_next_turn": "Coherence to Next Turn",
|
|
|
148 |
"discourse_user_topic_drift": "User Topic Drift",
|
149 |
"discourse_user_entity_continuity": "User Entity Continuity",
|
150 |
"discourse_num_user_turns": "Number of User Turns",
|
|
|
151 |
# Tokens per byte
|
152 |
"tokens_per_byte": "Tokens per Byte",
|
153 |
}
|
154 |
+
|
155 |
# Check exact match first
|
156 |
if metric_name in metric_display_names:
|
157 |
return metric_display_names[metric_name]
|
158 |
+
|
159 |
# Handle conversation-level aggregations
|
160 |
+
for suffix in [
|
161 |
+
"_conversation_mean",
|
162 |
+
"_conversation_min",
|
163 |
+
"_conversation_max",
|
164 |
+
"_conversation_std",
|
165 |
+
"_conversation_count",
|
166 |
+
]:
|
167 |
if metric_name.endswith(suffix):
|
168 |
+
base_metric = metric_name[: -len(suffix)]
|
169 |
if base_metric in metric_display_names:
|
170 |
agg_type = suffix.split("_")[-1].title()
|
171 |
return f"{metric_display_names[base_metric]} ({agg_type})"
|
172 |
+
|
173 |
# Handle turn-level metrics with "turn.turn_metrics." prefix
|
174 |
if metric_name.startswith("turn.turn_metrics."):
|
175 |
+
base_metric = metric_name[len("turn.turn_metrics.") :]
|
176 |
if base_metric in metric_display_names:
|
177 |
return metric_display_names[base_metric]
|
178 |
+
|
179 |
# Fallback: convert underscores to spaces and title case
|
180 |
clean_name = metric_name
|
181 |
for prefix in ["turn.turn_metrics.", "conversation_metrics.", "turn_metrics."]:
|
182 |
if clean_name.startswith(prefix):
|
183 |
+
clean_name = clean_name[len(prefix) :]
|
184 |
break
|
185 |
+
|
186 |
# Convert to human-readable format
|
187 |
clean_name = clean_name.replace("_", " ").title()
|
188 |
return clean_name
|
189 |
|
190 |
+
|
191 |
+
def render_metric_distribution(metric, filtered_df_exploded, selected_types):
|
192 |
+
"""Render distribution plot for a single metric."""
|
193 |
+
full_metric_name = f"turn.turn_metrics.{metric}"
|
194 |
+
|
195 |
+
if full_metric_name not in filtered_df_exploded.columns:
|
196 |
+
st.warning(f"Metric {metric} not found in dataset")
|
197 |
+
return
|
198 |
+
|
199 |
+
st.subheader(f"π {get_human_friendly_metric_name(metric)}")
|
200 |
+
|
201 |
+
# Clean the data
|
202 |
+
metric_data = filtered_df_exploded[["type", full_metric_name]].copy()
|
203 |
+
metric_data = metric_data.dropna()
|
204 |
+
|
205 |
+
if len(metric_data) == 0:
|
206 |
+
st.warning(f"No data available for {metric}")
|
207 |
+
return
|
208 |
+
|
209 |
+
# Create plotly histogram
|
210 |
+
fig = px.histogram(
|
211 |
+
metric_data,
|
212 |
+
x=full_metric_name,
|
213 |
+
color="type",
|
214 |
+
marginal="box",
|
215 |
+
title=f"Distribution of {get_human_friendly_metric_name(metric)}",
|
216 |
+
color_discrete_map=PLOT_PALETTE if len(selected_types) <= 3 else None,
|
217 |
+
opacity=0.7,
|
218 |
+
nbins=50,
|
219 |
+
)
|
220 |
+
|
221 |
+
fig.update_layout(
|
222 |
+
xaxis_title=get_human_friendly_metric_name(metric),
|
223 |
+
yaxis_title="Count",
|
224 |
+
height=400,
|
225 |
+
)
|
226 |
+
|
227 |
+
st.plotly_chart(fig, use_container_width=True)
|
228 |
+
|
229 |
+
# Summary statistics
|
230 |
+
col1, col2 = st.columns(2)
|
231 |
+
|
232 |
+
with col1:
|
233 |
+
st.write("**Summary Statistics**")
|
234 |
+
summary_stats = (
|
235 |
+
metric_data.groupby("type")[full_metric_name]
|
236 |
+
.agg(["count", "mean", "std", "min", "max"])
|
237 |
+
.round(3)
|
238 |
+
)
|
239 |
+
st.dataframe(summary_stats)
|
240 |
+
|
241 |
+
with col2:
|
242 |
+
st.write("**Percentiles**")
|
243 |
+
percentiles = (
|
244 |
+
metric_data.groupby("type")[full_metric_name]
|
245 |
+
.quantile([0.25, 0.5, 0.75])
|
246 |
+
.unstack()
|
247 |
+
.round(3)
|
248 |
+
)
|
249 |
+
percentiles.columns = ["25%", "50%", "75%"]
|
250 |
+
st.dataframe(percentiles)
|
251 |
+
|
252 |
+
|
253 |
# Setup page config
|
254 |
st.set_page_config(
|
255 |
page_title="Complexity Metrics Explorer",
|
256 |
page_icon="π",
|
257 |
layout="wide",
|
258 |
+
initial_sidebar_state="expanded",
|
259 |
)
|
260 |
|
261 |
+
|
262 |
# Cache data loading
|
263 |
@st.cache_data
|
264 |
def load_data(dataset_name):
|
265 |
"""Load and cache the dataset"""
|
266 |
+
df, df_exploded = load_and_prepare_dataset({"dataset_name": dataset_name})
|
|
|
|
|
267 |
return df, df_exploded
|
268 |
|
269 |
+
|
270 |
@st.cache_data
|
271 |
def get_metrics(df_exploded):
|
272 |
"""Get available metrics from the dataset"""
|
273 |
return get_available_turn_metrics(df_exploded)
|
274 |
|
275 |
+
|
276 |
def main():
|
277 |
st.title("π Complexity Metrics Explorer")
|
278 |
+
st.markdown(
|
279 |
+
"Interactive visualization of conversation complexity metrics across different dataset types."
|
280 |
+
)
|
281 |
+
|
282 |
# Dataset selection
|
283 |
st.sidebar.header("ποΈ Dataset Selection")
|
284 |
+
|
285 |
# Available datasets
|
286 |
available_datasets = [
|
287 |
"risky-conversations/jailbreaks_dataset_with_results_reduced",
|
288 |
"risky-conversations/jailbreaks_dataset_with_results",
|
289 |
"risky-conversations/jailbreaks_dataset_with_results_filtered_successful_jailbreak",
|
290 |
+
"Custom...",
|
291 |
]
|
292 |
+
|
293 |
selected_option = st.sidebar.selectbox(
|
294 |
"Select Dataset",
|
295 |
options=available_datasets,
|
296 |
index=0, # Default to reduced dataset
|
297 |
+
help="Choose which dataset to analyze",
|
298 |
)
|
299 |
+
|
300 |
# Handle custom dataset input
|
301 |
if selected_option == "Custom...":
|
302 |
selected_dataset = st.sidebar.text_input(
|
303 |
"Custom Dataset Name",
|
304 |
value="risky-conversations/jailbreaks_dataset_with_results_reduced",
|
305 |
+
help="Enter the full dataset name (e.g., 'risky-conversations/jailbreaks_dataset_with_results_reduced')",
|
306 |
)
|
307 |
if not selected_dataset.strip():
|
308 |
st.sidebar.warning("Please enter a dataset name")
|
309 |
st.stop()
|
310 |
else:
|
311 |
selected_dataset = selected_option
|
312 |
+
|
313 |
# Add refresh button
|
314 |
if st.sidebar.button("π Refresh Data", help="Clear cache and reload dataset"):
|
315 |
st.cache_data.clear()
|
316 |
st.rerun()
|
317 |
+
|
318 |
# Load data
|
319 |
with st.spinner(f"Loading dataset: {selected_dataset}..."):
|
320 |
try:
|
321 |
df, df_exploded = load_data(selected_dataset)
|
322 |
available_metrics = get_metrics(df_exploded)
|
323 |
+
|
324 |
# Display dataset info
|
325 |
col1, col2, col3, col4 = st.columns(4)
|
326 |
with col1:
|
327 |
+
st.metric("Dataset", selected_dataset.split("_")[-1].title())
|
328 |
with col2:
|
329 |
st.metric("Conversations", f"{len(df):,}")
|
330 |
with col3:
|
331 |
st.metric("Turns", f"{len(df_exploded):,}")
|
332 |
with col4:
|
333 |
st.metric("Metrics", len(available_metrics))
|
334 |
+
|
335 |
data_loaded = True
|
336 |
except Exception as e:
|
337 |
st.error(f"Error loading dataset: {e}")
|
338 |
st.info("Please check if the dataset exists and is accessible.")
|
339 |
+
st.info(
|
340 |
+
"π‘ Try using one of the predefined dataset options instead of custom input."
|
341 |
+
)
|
342 |
data_loaded = False
|
343 |
+
|
344 |
if not data_loaded:
|
345 |
st.stop()
|
346 |
+
|
347 |
# Sidebar controls
|
348 |
st.sidebar.header("ποΈ Controls")
|
349 |
+
|
350 |
# Dataset type filter
|
351 |
+
dataset_types = df["type"].unique()
|
352 |
selected_types = st.sidebar.multiselect(
|
353 |
"Select Dataset Types",
|
354 |
options=dataset_types,
|
355 |
default=dataset_types,
|
356 |
+
help="Filter by conversation type",
|
357 |
)
|
358 |
+
|
359 |
# Role filter
|
360 |
+
if "turn.role" in df_exploded.columns:
|
361 |
+
roles = df_exploded["turn.role"].dropna().unique()
|
362 |
# Assert only user and assistant roles exist
|
363 |
+
expected_roles = {"user", "assistant"}
|
364 |
actual_roles = set(roles)
|
365 |
+
assert actual_roles.issubset(
|
366 |
+
expected_roles
|
367 |
+
), f"Unexpected roles found: {actual_roles - expected_roles}. Expected only 'user' and 'assistant'"
|
368 |
+
|
369 |
st.sidebar.subheader("π₯ Role Filter")
|
370 |
col1, col2 = st.sidebar.columns(2)
|
371 |
+
|
372 |
with col1:
|
373 |
include_user = st.checkbox("User", value=True, help="Include user turns")
|
374 |
with col2:
|
375 |
+
include_assistant = st.checkbox(
|
376 |
+
"Assistant", value=True, help="Include assistant turns"
|
377 |
+
)
|
378 |
+
|
379 |
# Build selected roles list
|
380 |
selected_roles = []
|
381 |
+
if include_user and "user" in roles:
|
382 |
+
selected_roles.append("user")
|
383 |
+
if include_assistant and "assistant" in roles:
|
384 |
+
selected_roles.append("assistant")
|
385 |
+
|
386 |
# Show selection info
|
387 |
if selected_roles:
|
388 |
st.sidebar.success(f"Including: {', '.join(selected_roles)}")
|
|
|
390 |
st.sidebar.warning("No roles selected")
|
391 |
else:
|
392 |
selected_roles = None
|
393 |
+
|
394 |
# Filter data based on selections
|
395 |
+
filtered_df = df[df["type"].isin(selected_types)] if selected_types else df
|
396 |
+
filtered_df_exploded = (
|
397 |
+
df_exploded[df_exploded["type"].isin(selected_types)]
|
398 |
+
if selected_types
|
399 |
+
else df_exploded
|
400 |
+
)
|
401 |
+
|
402 |
+
if selected_roles and "turn.role" in filtered_df_exploded.columns:
|
403 |
+
filtered_df_exploded = filtered_df_exploded[
|
404 |
+
filtered_df_exploded["turn.role"].isin(selected_roles)
|
405 |
+
]
|
406 |
elif selected_roles is not None and len(selected_roles) == 0:
|
407 |
# If roles exist but none are selected, show empty dataset
|
408 |
+
filtered_df_exploded = filtered_df_exploded.iloc[
|
409 |
+
0:0
|
410 |
+
] # Empty dataframe with same structure
|
411 |
+
|
412 |
# Check if we have data after filtering
|
413 |
if len(filtered_df_exploded) == 0:
|
414 |
+
st.error(
|
415 |
+
"No data available with current filters. Please adjust your selection."
|
416 |
+
)
|
417 |
st.stop()
|
418 |
+
|
419 |
# Metric selection
|
420 |
st.sidebar.header("π Metrics")
|
421 |
+
|
422 |
# Dynamic metric categorization based on common patterns
|
423 |
def categorize_metrics(metrics):
|
424 |
"""Dynamically categorize metrics based on naming patterns"""
|
425 |
categories = {"All": metrics} # Always include all metrics
|
426 |
+
|
427 |
# Common patterns to look for
|
428 |
patterns = {
|
429 |
+
"Length": ["length", "byte", "word", "token", "char"],
|
430 |
+
"Readability": ["readability", "flesch", "standard"],
|
431 |
+
"Compression": ["lzw", "compression"],
|
432 |
+
"Language Model": ["ll_", "rll_", "logprob"],
|
433 |
+
"Working Memory": ["wm_"],
|
434 |
+
"Discourse": ["discourse"],
|
435 |
+
"Evaluation": ["rubric", "evaluation", "stealth"],
|
436 |
+
"Distribution": ["zipf", "type_token"],
|
437 |
+
"Coherence": ["coherence"],
|
438 |
+
"Entity": ["entity", "entities"],
|
439 |
+
"Cognitive": ["cognitive", "load"],
|
440 |
}
|
441 |
+
|
442 |
# Categorize metrics
|
443 |
for category, keywords in patterns.items():
|
444 |
+
matching_metrics = [
|
445 |
+
m for m in metrics if any(keyword in m.lower() for keyword in keywords)
|
446 |
+
]
|
447 |
if matching_metrics:
|
448 |
categories[category] = matching_metrics
|
449 |
+
|
450 |
# Find uncategorized metrics
|
451 |
categorized = set()
|
452 |
for cat_metrics in categories.values():
|
453 |
if cat_metrics != metrics: # Skip "All" category
|
454 |
categorized.update(cat_metrics)
|
455 |
+
|
456 |
uncategorized = [m for m in metrics if m not in categorized]
|
457 |
if uncategorized:
|
458 |
categories["Other"] = uncategorized
|
459 |
+
|
460 |
return categories
|
461 |
+
|
462 |
metric_categories = categorize_metrics(available_metrics)
|
463 |
+
|
464 |
# Metric selection interface
|
465 |
selection_mode = st.sidebar.radio(
|
466 |
"Selection Mode",
|
467 |
["By Category", "Search/Filter", "Select All"],
|
468 |
+
help="Choose how to select metrics",
|
469 |
)
|
470 |
+
|
471 |
if selection_mode == "By Category":
|
472 |
selected_category = st.sidebar.selectbox(
|
473 |
+
"Metric Category",
|
474 |
options=list(metric_categories.keys()),
|
475 |
+
help=f"Found {len(metric_categories)} categories",
|
476 |
)
|
477 |
+
|
478 |
available_in_category = metric_categories[selected_category]
|
479 |
+
default_selection = (
|
480 |
+
available_in_category[:5]
|
481 |
+
if len(available_in_category) > 5
|
482 |
+
else available_in_category
|
483 |
+
)
|
484 |
+
|
485 |
# Add select all button for category
|
486 |
col1, col2 = st.sidebar.columns(2)
|
487 |
with col1:
|
|
|
490 |
with col2:
|
491 |
if st.button("Clear All", key="clear_all_category"):
|
492 |
st.session_state.selected_metrics_category = []
|
493 |
+
|
494 |
# Use session state for persistence
|
495 |
if "selected_metrics_category" not in st.session_state:
|
496 |
st.session_state.selected_metrics_category = default_selection
|
497 |
+
|
498 |
selected_metrics = st.sidebar.multiselect(
|
499 |
f"Select Metrics ({len(available_in_category)} available)",
|
500 |
options=available_in_category,
|
501 |
default=st.session_state.selected_metrics_category,
|
502 |
key="metrics_multiselect_category",
|
503 |
+
help="Choose metrics to visualize",
|
504 |
)
|
505 |
+
|
506 |
elif selection_mode == "Search/Filter":
|
507 |
search_term = st.sidebar.text_input(
|
508 |
"Search Metrics",
|
509 |
placeholder="Enter keywords to filter metrics...",
|
510 |
+
help="Search for metrics containing specific terms",
|
511 |
)
|
512 |
+
|
513 |
if search_term:
|
514 |
+
filtered_metrics = [
|
515 |
+
m for m in available_metrics if search_term.lower() in m.lower()
|
516 |
+
]
|
517 |
else:
|
518 |
filtered_metrics = available_metrics
|
519 |
+
|
520 |
st.sidebar.write(f"Found {len(filtered_metrics)} metrics")
|
521 |
+
|
522 |
# Add select all button for search results
|
523 |
col1, col2 = st.sidebar.columns(2)
|
524 |
with col1:
|
|
|
527 |
with col2:
|
528 |
if st.button("Clear All", key="clear_all_search"):
|
529 |
st.session_state.selected_metrics_search = []
|
530 |
+
|
531 |
# Use session state for persistence
|
532 |
if "selected_metrics_search" not in st.session_state:
|
533 |
+
st.session_state.selected_metrics_search = (
|
534 |
+
filtered_metrics[:5]
|
535 |
+
if len(filtered_metrics) > 5
|
536 |
+
else filtered_metrics[:3]
|
537 |
+
)
|
538 |
+
|
539 |
selected_metrics = st.sidebar.multiselect(
|
540 |
"Select Metrics",
|
541 |
options=filtered_metrics,
|
542 |
default=st.session_state.selected_metrics_search,
|
543 |
key="metrics_multiselect_search",
|
544 |
+
help="Choose metrics to visualize",
|
545 |
)
|
546 |
+
|
547 |
else: # Select All
|
548 |
# Add select all button for all metrics
|
549 |
col1, col2 = st.sidebar.columns(2)
|
|
|
553 |
with col2:
|
554 |
if st.button("Clear All", key="clear_all_all"):
|
555 |
st.session_state.selected_metrics_all = []
|
556 |
+
|
557 |
# Use session state for persistence
|
558 |
if "selected_metrics_all" not in st.session_state:
|
559 |
+
st.session_state.selected_metrics_all = available_metrics[
|
560 |
+
:10
|
561 |
+
] # Limit default to first 10 for performance
|
562 |
+
|
563 |
selected_metrics = st.sidebar.multiselect(
|
564 |
f"All Metrics ({len(available_metrics)} total)",
|
565 |
options=available_metrics,
|
566 |
default=st.session_state.selected_metrics_all,
|
567 |
key="metrics_multiselect_all",
|
568 |
+
help="All available metrics - be careful with performance for large selections",
|
569 |
)
|
570 |
+
|
571 |
# Show selection summary
|
572 |
if selected_metrics:
|
573 |
st.sidebar.success(f"Selected {len(selected_metrics)} metrics")
|
574 |
+
|
575 |
# Performance warning for large selections
|
576 |
if len(selected_metrics) > 20:
|
577 |
+
st.sidebar.warning(
|
578 |
+
f"β οΈ Large selection ({len(selected_metrics)} metrics) may impact performance"
|
579 |
+
)
|
580 |
elif len(selected_metrics) > 50:
|
581 |
+
st.sidebar.error(
|
582 |
+
f"π¨ Very large selection ({len(selected_metrics)} metrics) - consider reducing for better performance"
|
583 |
+
)
|
584 |
else:
|
585 |
st.sidebar.warning("No metrics selected")
|
586 |
+
|
587 |
# Metric info expander
|
588 |
with st.sidebar.expander("βΉοΈ Metric Information", expanded=False):
|
589 |
st.write(f"**Total Available Metrics:** {len(available_metrics)}")
|
590 |
st.write(f"**Categories Found:** {len(metric_categories)}")
|
591 |
+
|
592 |
if st.checkbox("Show all metric names", key="show_all_metrics"):
|
593 |
st.write("**All Available Metrics:**")
|
594 |
for i, metric in enumerate(available_metrics, 1):
|
595 |
st.write(f"{i}. `{metric}`")
|
596 |
+
|
597 |
# Main content tabs
|
598 |
+
tab1, tab2, tab3, tab4, tab5 = st.tabs(
|
599 |
+
[
|
600 |
+
"π Distributions",
|
601 |
+
"π Correlations",
|
602 |
+
"π Comparisons",
|
603 |
+
"π Conversation",
|
604 |
+
"π― Details",
|
605 |
+
]
|
606 |
+
)
|
607 |
+
|
608 |
with tab1:
|
609 |
st.header("Distribution Analysis")
|
610 |
+
|
611 |
if not selected_metrics:
|
612 |
st.warning("Please select at least one metric to visualize.")
|
613 |
return
|
614 |
+
|
615 |
+
# Create buttons for each metric to prevent loading all at once
|
616 |
+
st.info(
|
617 |
+
f"π Select a metric to plot its distribution ({len(selected_metrics)} metrics available)"
|
618 |
+
)
|
619 |
+
|
620 |
+
# Organize buttons in columns for better layout
|
621 |
+
cols_per_row = 3
|
622 |
+
for i in range(0, len(selected_metrics), cols_per_row):
|
623 |
+
cols = st.columns(cols_per_row)
|
624 |
+
for j, metric in enumerate(selected_metrics[i : i + cols_per_row]):
|
625 |
+
with cols[j]:
|
626 |
+
friendly_name = get_human_friendly_metric_name(metric)
|
627 |
+
# Truncate button text if too long
|
628 |
+
button_text = (
|
629 |
+
friendly_name[:25] + "..."
|
630 |
+
if len(friendly_name) > 25
|
631 |
+
else friendly_name
|
632 |
+
)
|
633 |
+
|
634 |
+
if st.button(
|
635 |
+
f"π {button_text}",
|
636 |
+
key=f"plot_{metric}",
|
637 |
+
help=f"Plot distribution for {friendly_name}",
|
638 |
+
):
|
639 |
+
render_metric_distribution(
|
640 |
+
metric, filtered_df_exploded, selected_types
|
641 |
+
)
|
642 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
643 |
with tab2:
|
644 |
st.header("Correlation Analysis")
|
645 |
+
|
646 |
if len(selected_metrics) < 2:
|
647 |
st.warning("Please select at least 2 metrics for correlation analysis.")
|
648 |
else:
|
649 |
+
# Add button to trigger correlation analysis
|
650 |
+
st.info(
|
651 |
+
f"π Ready to analyze correlations between {len(selected_metrics)} metrics"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
652 |
)
|
653 |
+
|
654 |
+
col1, col2 = st.columns([1, 3])
|
655 |
+
with col1:
|
656 |
+
run_correlation = st.button(
|
657 |
+
"π Run Correlation Analysis",
|
658 |
+
help="Calculate and display correlation matrix and scatter plots",
|
659 |
+
)
|
660 |
+
with col2:
|
661 |
+
if len(selected_metrics) > 10:
|
662 |
+
st.warning(
|
663 |
+
f"β οΈ Large analysis ({len(selected_metrics)} metrics) - may take some time"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
664 |
)
|
665 |
+
|
666 |
+
if run_correlation:
|
667 |
+
with st.spinner("Calculating correlations..."):
|
668 |
+
# Prepare correlation data
|
669 |
+
corr_columns = [f"turn.turn_metrics.{m}" for m in selected_metrics]
|
670 |
+
corr_data = filtered_df_exploded[corr_columns + ["type"]].copy()
|
671 |
+
|
672 |
+
# Clean column names for display
|
673 |
+
corr_data.columns = [
|
674 |
+
(
|
675 |
+
get_human_friendly_metric_name(
|
676 |
+
col.replace("turn.turn_metrics.", "")
|
677 |
+
)
|
678 |
+
if col.startswith("turn.turn_metrics.")
|
679 |
+
else col
|
680 |
+
)
|
681 |
+
for col in corr_data.columns
|
682 |
+
]
|
683 |
+
|
684 |
+
# Calculate correlation matrix
|
685 |
+
corr_matrix = corr_data.select_dtypes(include=[np.number]).corr()
|
686 |
+
|
687 |
+
# Create correlation heatmap
|
688 |
+
fig = px.imshow(
|
689 |
+
corr_matrix,
|
690 |
+
text_auto=True,
|
691 |
+
aspect="auto",
|
692 |
+
title="Correlation Matrix",
|
693 |
+
color_continuous_scale="RdBu_r",
|
694 |
+
zmin=-1,
|
695 |
+
zmax=1,
|
696 |
+
)
|
697 |
+
|
698 |
+
fig.update_layout(height=600)
|
699 |
st.plotly_chart(fig, use_container_width=True)
|
700 |
+
|
701 |
+
# Scatter plots for strong correlations
|
702 |
+
st.subheader("Strong Correlations")
|
703 |
+
|
704 |
+
# Find strong correlations (>0.7 or <-0.7)
|
705 |
+
strong_corrs = []
|
706 |
+
for i in range(len(corr_matrix.columns)):
|
707 |
+
for j in range(i + 1, len(corr_matrix.columns)):
|
708 |
+
corr_val = corr_matrix.iloc[i, j]
|
709 |
+
if abs(corr_val) > 0.7:
|
710 |
+
strong_corrs.append(
|
711 |
+
(
|
712 |
+
corr_matrix.columns[i],
|
713 |
+
corr_matrix.columns[j],
|
714 |
+
corr_val,
|
715 |
+
)
|
716 |
+
)
|
717 |
+
|
718 |
+
if strong_corrs:
|
719 |
+
for metric1, metric2, corr_val in strong_corrs[
|
720 |
+
:3
|
721 |
+
]: # Show top 3
|
722 |
+
fig = px.scatter(
|
723 |
+
corr_data,
|
724 |
+
x=metric1,
|
725 |
+
y=metric2,
|
726 |
+
color="type",
|
727 |
+
title=f"{metric1} vs {metric2} (r={corr_val:.3f})",
|
728 |
+
color_discrete_map=(
|
729 |
+
PLOT_PALETTE if len(selected_types) <= 3 else None
|
730 |
+
),
|
731 |
+
opacity=0.6,
|
732 |
+
)
|
733 |
+
st.plotly_chart(fig, use_container_width=True)
|
734 |
+
else:
|
735 |
+
st.info(
|
736 |
+
"No strong correlations (|r| > 0.7) found between selected metrics."
|
737 |
+
)
|
738 |
+
|
739 |
with tab3:
|
740 |
st.header("Type Comparisons")
|
741 |
+
|
742 |
if not selected_metrics:
|
743 |
st.warning("Please select at least one metric to compare.")
|
744 |
else:
|
745 |
# Box plots for each metric
|
746 |
for metric in selected_metrics:
|
747 |
full_metric_name = f"turn.turn_metrics.{metric}"
|
748 |
+
|
749 |
if full_metric_name not in filtered_df_exploded.columns:
|
750 |
continue
|
751 |
+
|
752 |
st.subheader(f"π¦ {get_human_friendly_metric_name(metric)} by Type")
|
753 |
+
|
754 |
# Create box plot
|
755 |
fig = px.box(
|
756 |
filtered_df_exploded.dropna(subset=[full_metric_name]),
|
757 |
+
x="type",
|
758 |
y=full_metric_name,
|
759 |
title=f"Distribution of {get_human_friendly_metric_name(metric)} by Type",
|
760 |
+
color="type",
|
761 |
+
color_discrete_map=(
|
762 |
+
PLOT_PALETTE if len(selected_types) <= 3 else None
|
763 |
+
),
|
764 |
)
|
765 |
+
|
766 |
fig.update_layout(
|
767 |
xaxis_title="Dataset Type",
|
768 |
yaxis_title=get_human_friendly_metric_name(metric),
|
769 |
+
height=400,
|
770 |
)
|
771 |
+
|
772 |
st.plotly_chart(fig, use_container_width=True)
|
773 |
+
|
774 |
with tab4:
|
775 |
st.header("Individual Conversation Analysis")
|
776 |
+
|
777 |
# Conversation selector
|
778 |
st.subheader("π Select Conversation")
|
779 |
+
|
780 |
# Get unique conversations with some metadata
|
781 |
conversation_info = []
|
782 |
for idx, row in filtered_df.iterrows():
|
783 |
+
conv_type = row["type"]
|
784 |
# Get basic info about the conversation
|
785 |
+
conv_turns = len(row.get("conversation", []))
|
786 |
+
conversation_info.append(
|
787 |
+
{
|
788 |
+
"index": idx,
|
789 |
+
"type": conv_type,
|
790 |
+
"turns": conv_turns,
|
791 |
+
"display": f"Conversation {idx} ({conv_type}) - {conv_turns} turns",
|
792 |
+
}
|
793 |
+
)
|
794 |
+
|
795 |
# Sort by type and number of turns for better organization
|
796 |
+
conversation_info = sorted(
|
797 |
+
conversation_info, key=lambda x: (x["type"], -x["turns"])
|
798 |
+
)
|
799 |
+
|
800 |
# Conversation selection
|
801 |
col1, col2 = st.columns([3, 1])
|
802 |
+
|
803 |
with col1:
|
804 |
selected_conv_display = st.selectbox(
|
805 |
"Choose a conversation to analyze",
|
806 |
+
options=[conv["display"] for conv in conversation_info],
|
807 |
+
help="Select a conversation to view detailed metrics and content",
|
808 |
)
|
809 |
+
|
810 |
with col2:
|
811 |
if st.button("π² Random", help="Select a random conversation"):
|
812 |
import random
|
813 |
+
|
814 |
+
selected_conv_display = random.choice(
|
815 |
+
[conv["display"] for conv in conversation_info]
|
816 |
+
)
|
817 |
st.rerun()
|
818 |
+
|
819 |
# Get the selected conversation data
|
820 |
+
selected_conv_info = next(
|
821 |
+
conv
|
822 |
+
for conv in conversation_info
|
823 |
+
if conv["display"] == selected_conv_display
|
824 |
+
)
|
825 |
+
selected_idx = selected_conv_info["index"]
|
826 |
selected_conversation = filtered_df.iloc[selected_idx]
|
827 |
+
|
828 |
# Display conversation metadata
|
829 |
st.subheader("π Conversation Overview")
|
830 |
+
|
831 |
# First row - basic info
|
832 |
col1, col2, col3, col4 = st.columns(4)
|
833 |
with col1:
|
834 |
+
st.metric("Type", selected_conversation["type"])
|
835 |
with col2:
|
836 |
st.metric("Index", selected_idx)
|
837 |
with col3:
|
838 |
+
st.metric("Total Turns", len(selected_conversation.get("conversation", [])))
|
839 |
with col4:
|
840 |
# Count user vs assistant turns
|
841 |
+
roles = [
|
842 |
+
turn.get("role", "unknown")
|
843 |
+
for turn in selected_conversation.get("conversation", [])
|
844 |
+
]
|
845 |
+
user_turns = roles.count("user")
|
846 |
+
assistant_turns = roles.count("assistant")
|
847 |
st.metric("User/Assistant", f"{user_turns}/{assistant_turns}")
|
848 |
+
|
849 |
# Second row - additional metadata
|
850 |
col1, col2, col3 = st.columns(3)
|
851 |
with col1:
|
852 |
+
provenance = selected_conversation.get("provenance_dataset", "Unknown")
|
853 |
st.metric("Dataset Source", provenance)
|
854 |
with col2:
|
855 |
+
language = selected_conversation.get("language", "Unknown")
|
856 |
+
st.metric("Language", language.upper() if language else "Unknown")
|
857 |
with col3:
|
858 |
+
timestamp = selected_conversation.get("timestamp", None)
|
859 |
if timestamp:
|
860 |
# Handle different timestamp formats
|
861 |
if isinstance(timestamp, str):
|
|
|
864 |
st.metric("Timestamp", str(timestamp))
|
865 |
else:
|
866 |
st.metric("Timestamp", "Not Available")
|
867 |
+
|
868 |
# Add toxicity summary
|
869 |
+
conversation_turns_temp = selected_conversation.get("conversation", [])
|
870 |
+
if hasattr(conversation_turns_temp, "tolist"):
|
871 |
conversation_turns_temp = conversation_turns_temp.tolist()
|
872 |
elif conversation_turns_temp is None:
|
873 |
conversation_turns_temp = []
|
874 |
+
|
875 |
if len(conversation_turns_temp) > 0:
|
876 |
# Calculate overall toxicity statistics
|
877 |
all_toxicities = []
|
878 |
for turn in conversation_turns_temp:
|
879 |
+
toxicities = turn.get("toxicities", {})
|
880 |
+
if toxicities and "toxicity" in toxicities:
|
881 |
+
all_toxicities.append(toxicities["toxicity"])
|
882 |
+
|
883 |
if all_toxicities:
|
884 |
avg_toxicity = sum(all_toxicities) / len(all_toxicities)
|
885 |
max_toxicity = max(all_toxicities)
|
886 |
+
|
887 |
st.markdown("**π Toxicity Summary:**")
|
888 |
col1, col2, col3 = st.columns(3)
|
889 |
with col1:
|
890 |
# Color code average toxicity
|
891 |
if avg_toxicity > 0.5:
|
892 |
+
st.metric(
|
893 |
+
"Average Toxicity",
|
894 |
+
f"{avg_toxicity:.4f}",
|
895 |
+
delta="HIGH",
|
896 |
+
delta_color="inverse",
|
897 |
+
)
|
898 |
elif avg_toxicity > 0.1:
|
899 |
+
st.metric(
|
900 |
+
"Average Toxicity",
|
901 |
+
f"{avg_toxicity:.4f}",
|
902 |
+
delta="MED",
|
903 |
+
delta_color="off",
|
904 |
+
)
|
905 |
else:
|
906 |
+
st.metric(
|
907 |
+
"Average Toxicity",
|
908 |
+
f"{avg_toxicity:.4f}",
|
909 |
+
delta="LOW",
|
910 |
+
delta_color="normal",
|
911 |
+
)
|
912 |
+
|
913 |
with col2:
|
914 |
# Color code max toxicity
|
915 |
if max_toxicity > 0.5:
|
916 |
+
st.metric(
|
917 |
+
"Max Toxicity",
|
918 |
+
f"{max_toxicity:.4f}",
|
919 |
+
delta="HIGH",
|
920 |
+
delta_color="inverse",
|
921 |
+
)
|
922 |
elif max_toxicity > 0.1:
|
923 |
+
st.metric(
|
924 |
+
"Max Toxicity",
|
925 |
+
f"{max_toxicity:.4f}",
|
926 |
+
delta="MED",
|
927 |
+
delta_color="off",
|
928 |
+
)
|
929 |
else:
|
930 |
+
st.metric(
|
931 |
+
"Max Toxicity",
|
932 |
+
f"{max_toxicity:.4f}",
|
933 |
+
delta="LOW",
|
934 |
+
delta_color="normal",
|
935 |
+
)
|
936 |
+
|
937 |
with col3:
|
938 |
high_tox_turns = sum(1 for t in all_toxicities if t > 0.5)
|
939 |
st.metric("High Toxicity Turns", high_tox_turns)
|
940 |
+
|
941 |
# Get conversation turns with metrics
|
942 |
+
conv_turns_data = filtered_df_exploded[
|
943 |
+
filtered_df_exploded.index.isin(
|
944 |
+
filtered_df_exploded[
|
945 |
+
filtered_df_exploded.index
|
946 |
+
// len(filtered_df_exploded)
|
947 |
+
* len(filtered_df)
|
948 |
+
+ filtered_df_exploded.index % len(filtered_df)
|
949 |
+
== selected_idx
|
950 |
+
].index
|
951 |
+
)
|
952 |
+
].copy()
|
953 |
+
|
954 |
# Alternative approach: filter by matching all conversation data
|
955 |
# This is more reliable but less efficient
|
956 |
conv_turns_data = []
|
957 |
start_idx = None
|
958 |
for idx, row in filtered_df_exploded.iterrows():
|
959 |
# Check if this row belongs to our selected conversation
|
960 |
+
if (
|
961 |
+
row["type"] == selected_conversation["type"]
|
962 |
+
and hasattr(row, "conversation")
|
963 |
+
and row.get("conversation") is not None
|
964 |
+
):
|
965 |
# This is a simplified approach - in reality you'd need better conversation matching
|
966 |
pass
|
967 |
+
|
968 |
# Simpler approach: get all turns from the conversation directly
|
969 |
+
conversation_turns = selected_conversation.get("conversation", [])
|
970 |
+
|
971 |
# Ensure conversation_turns is a list and handle different data types
|
972 |
+
if hasattr(conversation_turns, "tolist"):
|
973 |
conversation_turns = conversation_turns.tolist()
|
974 |
elif conversation_turns is None:
|
975 |
conversation_turns = []
|
976 |
+
|
977 |
if len(conversation_turns) > 0:
|
978 |
# Display conversation content with metrics
|
979 |
st.subheader("π¬ Conversation with Metrics")
|
980 |
+
|
981 |
# Get actual turn-level data for this conversation
|
982 |
turn_metric_columns = [f"turn.turn_metrics.{m}" for m in selected_metrics]
|
983 |
+
available_columns = [
|
984 |
+
col
|
985 |
+
for col in turn_metric_columns
|
986 |
+
if col in filtered_df_exploded.columns
|
987 |
+
]
|
988 |
+
|
989 |
# Get sample metrics for this conversation type (since exact matching is complex)
|
990 |
sample_metrics = None
|
991 |
if available_columns:
|
992 |
+
type_turns = filtered_df_exploded[
|
993 |
+
filtered_df_exploded["type"] == selected_conversation["type"]
|
994 |
+
]
|
995 |
sample_size = min(len(conversation_turns), len(type_turns))
|
996 |
if sample_size > 0:
|
997 |
sample_metrics = type_turns.head(sample_size)
|
998 |
+
|
999 |
# Display each turn with its metrics
|
1000 |
for i, turn in enumerate(conversation_turns):
|
1001 |
+
role = turn.get("role", "unknown")
|
1002 |
+
content = turn.get("content", "No content")
|
1003 |
+
|
1004 |
# Display turn content with role styling
|
1005 |
+
if role == "user":
|
1006 |
st.markdown(f"**π€ User (Turn {i+1}):**")
|
1007 |
st.info(content)
|
1008 |
+
elif role == "assistant":
|
1009 |
st.markdown(f"**π€ Assistant (Turn {i+1}):**")
|
1010 |
st.success(content)
|
1011 |
else:
|
1012 |
st.markdown(f"**β {role.title()} (Turn {i+1}):**")
|
1013 |
st.warning(content)
|
1014 |
+
|
1015 |
# Display metrics for this turn
|
1016 |
if sample_metrics is not None and i < len(sample_metrics):
|
1017 |
turn_row = sample_metrics.iloc[i]
|
1018 |
+
|
1019 |
# Create metrics display
|
1020 |
metrics_for_turn = {}
|
1021 |
for col in available_columns:
|
1022 |
+
metric_name = col.replace("turn.turn_metrics.", "")
|
1023 |
friendly_name = get_human_friendly_metric_name(metric_name)
|
1024 |
+
value = turn_row.get(col, "N/A")
|
1025 |
if pd.notna(value) and isinstance(value, (int, float)):
|
1026 |
metrics_for_turn[friendly_name] = round(value, 3)
|
1027 |
else:
|
1028 |
+
metrics_for_turn[friendly_name] = "N/A"
|
1029 |
+
|
1030 |
# Add toxicity metrics if available
|
1031 |
+
toxicities = turn.get("toxicities", {})
|
1032 |
if toxicities:
|
1033 |
st.markdown("**π Toxicity Scores:**")
|
1034 |
tox_cols = st.columns(4)
|
1035 |
tox_metrics = [
|
1036 |
+
("toxicity", "Overall Toxicity"),
|
1037 |
+
("severe_toxicity", "Severe Toxicity"),
|
1038 |
+
("identity_attack", "Identity Attack"),
|
1039 |
+
("insult", "Insult"),
|
1040 |
+
("obscene", "Obscene"),
|
1041 |
+
("sexual_explicit", "Sexual Explicit"),
|
1042 |
+
("threat", "Threat"),
|
1043 |
]
|
1044 |
+
|
1045 |
for idx, (tox_key, tox_name) in enumerate(tox_metrics):
|
1046 |
if tox_key in toxicities:
|
1047 |
col_idx = idx % 4
|
|
|
1050 |
if isinstance(tox_value, (int, float)):
|
1051 |
# Color code based on toxicity level
|
1052 |
if tox_value > 0.5:
|
1053 |
+
st.metric(
|
1054 |
+
tox_name,
|
1055 |
+
f"{tox_value:.4f}",
|
1056 |
+
delta="HIGH",
|
1057 |
+
delta_color="inverse",
|
1058 |
+
)
|
1059 |
elif tox_value > 0.1:
|
1060 |
+
st.metric(
|
1061 |
+
tox_name,
|
1062 |
+
f"{tox_value:.4f}",
|
1063 |
+
delta="MED",
|
1064 |
+
delta_color="off",
|
1065 |
+
)
|
1066 |
else:
|
1067 |
+
st.metric(
|
1068 |
+
tox_name,
|
1069 |
+
f"{tox_value:.4f}",
|
1070 |
+
delta="LOW",
|
1071 |
+
delta_color="normal",
|
1072 |
+
)
|
1073 |
else:
|
1074 |
st.metric(tox_name, str(tox_value))
|
1075 |
+
|
1076 |
# Display complexity metrics
|
1077 |
if metrics_for_turn:
|
1078 |
st.markdown("**π Complexity Metrics:**")
|
|
|
1080 |
num_cols = min(4, len(metrics_for_turn))
|
1081 |
if num_cols > 0:
|
1082 |
cols = st.columns(num_cols)
|
1083 |
+
for idx, (metric_name, value) in enumerate(
|
1084 |
+
metrics_for_turn.items()
|
1085 |
+
):
|
1086 |
col_idx = idx % num_cols
|
1087 |
with cols[col_idx]:
|
1088 |
+
if (
|
1089 |
+
isinstance(value, (int, float))
|
1090 |
+
and value != "N/A"
|
1091 |
+
):
|
1092 |
st.metric(metric_name, value)
|
1093 |
else:
|
1094 |
st.metric(metric_name, str(value))
|
1095 |
else:
|
1096 |
# Show toxicity even when no complexity metrics available
|
1097 |
+
toxicities = turn.get("toxicities", {})
|
1098 |
if toxicities:
|
1099 |
st.markdown("**π Toxicity Scores:**")
|
1100 |
tox_cols = st.columns(4)
|
1101 |
tox_metrics = [
|
1102 |
+
("toxicity", "Overall Toxicity"),
|
1103 |
+
("severe_toxicity", "Severe Toxicity"),
|
1104 |
+
("identity_attack", "Identity Attack"),
|
1105 |
+
("insult", "Insult"),
|
1106 |
+
("obscene", "Obscene"),
|
1107 |
+
("sexual_explicit", "Sexual Explicit"),
|
1108 |
+
("threat", "Threat"),
|
1109 |
]
|
1110 |
+
|
1111 |
for idx, (tox_key, tox_name) in enumerate(tox_metrics):
|
1112 |
if tox_key in toxicities:
|
1113 |
col_idx = idx % 4
|
|
|
1116 |
if isinstance(tox_value, (int, float)):
|
1117 |
# Color code based on toxicity level
|
1118 |
if tox_value > 0.5:
|
1119 |
+
st.metric(
|
1120 |
+
tox_name,
|
1121 |
+
f"{tox_value:.4f}",
|
1122 |
+
delta="HIGH",
|
1123 |
+
delta_color="inverse",
|
1124 |
+
)
|
1125 |
elif tox_value > 0.1:
|
1126 |
+
st.metric(
|
1127 |
+
tox_name,
|
1128 |
+
f"{tox_value:.4f}",
|
1129 |
+
delta="MED",
|
1130 |
+
delta_color="off",
|
1131 |
+
)
|
1132 |
else:
|
1133 |
+
st.metric(
|
1134 |
+
tox_name,
|
1135 |
+
f"{tox_value:.4f}",
|
1136 |
+
delta="LOW",
|
1137 |
+
delta_color="normal",
|
1138 |
+
)
|
1139 |
else:
|
1140 |
st.metric(tox_name, str(tox_value))
|
1141 |
+
|
1142 |
# Show basic turn statistics when no complexity metrics available
|
1143 |
st.markdown("**π Basic Statistics:**")
|
1144 |
col1, col2, col3 = st.columns(3)
|
|
|
1148 |
st.metric("Words", len(content.split()))
|
1149 |
with col3:
|
1150 |
st.metric("Role", role.title())
|
1151 |
+
|
1152 |
# Add separator between turns
|
1153 |
st.divider()
|
1154 |
+
|
1155 |
# Plot metrics over turns with real data if available
|
1156 |
if available_columns and sample_metrics is not None:
|
1157 |
st.subheader("π Metrics Over Turns")
|
1158 |
+
|
1159 |
fig = go.Figure()
|
1160 |
+
|
1161 |
# Add traces for each selected metric (real data)
|
1162 |
for col in available_columns[:5]: # Limit to first 5 for readability
|
1163 |
+
metric_name = col.replace("turn.turn_metrics.", "")
|
1164 |
friendly_name = get_human_friendly_metric_name(metric_name)
|
1165 |
+
|
1166 |
# Get values for this metric
|
1167 |
y_values = []
|
1168 |
for _, turn_row in sample_metrics.iterrows():
|
|
|
1171 |
y_values.append(value)
|
1172 |
else:
|
1173 |
y_values.append(None)
|
1174 |
+
|
1175 |
if any(v is not None for v in y_values):
|
1176 |
+
fig.add_trace(
|
1177 |
+
go.Scatter(
|
1178 |
+
x=list(range(1, len(y_values) + 1)),
|
1179 |
+
y=y_values,
|
1180 |
+
mode="lines+markers",
|
1181 |
+
name=friendly_name,
|
1182 |
+
line=dict(width=2),
|
1183 |
+
marker=dict(size=8),
|
1184 |
+
connectgaps=False,
|
1185 |
+
)
|
1186 |
+
)
|
1187 |
+
|
1188 |
if fig.data: # Only show if we have data
|
1189 |
fig.update_layout(
|
1190 |
title="Complexity Metrics Across Conversation Turns",
|
1191 |
xaxis_title="Turn Number",
|
1192 |
yaxis_title="Metric Value",
|
1193 |
height=400,
|
1194 |
+
hovermode="x unified",
|
1195 |
)
|
1196 |
+
|
1197 |
st.plotly_chart(fig, use_container_width=True)
|
1198 |
else:
|
1199 |
+
st.info(
|
1200 |
+
"No numeric metric data available to plot for this conversation type."
|
1201 |
+
)
|
1202 |
+
|
1203 |
elif selected_metrics:
|
1204 |
+
st.info(
|
1205 |
+
"Select metrics that are available in the dataset to see turn-level analysis."
|
1206 |
+
)
|
1207 |
else:
|
1208 |
st.warning("Select some metrics to see detailed turn-level analysis.")
|
1209 |
+
|
1210 |
else:
|
1211 |
st.warning("No conversation data available for the selected conversation.")
|
1212 |
+
|
1213 |
with tab5:
|
1214 |
st.header("Detailed View")
|
1215 |
+
|
1216 |
+
# Add button to trigger detailed analysis
|
1217 |
+
st.info("π― Generate detailed dataset analysis and visualizations")
|
1218 |
+
|
1219 |
+
col1, col2 = st.columns([1, 3])
|
|
|
|
|
|
|
1220 |
with col1:
|
1221 |
+
show_details = st.button(
|
1222 |
+
"π Show Detailed Analysis",
|
1223 |
+
help="Generate comprehensive dataset overview and metric analysis",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1224 |
)
|
1225 |
+
with col2:
|
1226 |
+
if len(selected_metrics) > 20:
|
1227 |
+
st.warning("β οΈ Large metric selection - analysis may take some time")
|
1228 |
+
|
1229 |
+
if show_details:
|
1230 |
+
with st.spinner("Generating detailed analysis..."):
|
1231 |
+
# Data overview
|
1232 |
+
st.subheader("π Dataset Overview")
|
1233 |
+
|
1234 |
+
st.info(f"**Current Dataset:** `{selected_dataset}`")
|
1235 |
+
|
1236 |
+
col1, col2, col3 = st.columns(3)
|
1237 |
+
|
1238 |
+
with col1:
|
1239 |
+
st.metric("Total Conversations", len(filtered_df))
|
1240 |
+
|
1241 |
+
with col2:
|
1242 |
+
st.metric("Total Turns", len(filtered_df_exploded))
|
1243 |
+
|
1244 |
+
with col3:
|
1245 |
+
st.metric("Available Metrics", len(available_metrics))
|
1246 |
+
|
1247 |
+
# Type distribution
|
1248 |
+
st.subheader("π Type Distribution")
|
1249 |
+
type_counts = filtered_df["type"].value_counts()
|
1250 |
+
|
1251 |
+
fig = px.pie(
|
1252 |
+
values=type_counts.values,
|
1253 |
+
names=type_counts.index,
|
1254 |
+
title="Distribution of Conversation Types",
|
1255 |
+
color_discrete_map=PLOT_PALETTE if len(type_counts) <= 3 else None,
|
1256 |
+
)
|
1257 |
+
|
1258 |
+
st.plotly_chart(fig, use_container_width=True)
|
1259 |
+
|
1260 |
+
# Sample data
|
1261 |
+
st.subheader("π Sample Data")
|
1262 |
+
|
1263 |
+
if st.checkbox("Show raw data sample"):
|
1264 |
+
sample_cols = ["type"] + [
|
1265 |
+
f"turn.turn_metrics.{m}"
|
1266 |
+
for m in selected_metrics
|
1267 |
+
if f"turn.turn_metrics.{m}" in filtered_df_exploded.columns
|
1268 |
+
]
|
1269 |
+
sample_data = filtered_df_exploded[sample_cols].head(100)
|
1270 |
+
st.dataframe(sample_data)
|
1271 |
+
|
1272 |
+
# Metric availability
|
1273 |
+
st.subheader("π Metric Availability")
|
1274 |
+
|
1275 |
+
metric_completeness = {}
|
1276 |
+
for metric in selected_metrics:
|
1277 |
+
full_metric_name = f"turn.turn_metrics.{metric}"
|
1278 |
+
if full_metric_name in filtered_df_exploded.columns:
|
1279 |
+
completeness = (
|
1280 |
+
1
|
1281 |
+
- filtered_df_exploded[full_metric_name].isna().sum()
|
1282 |
+
/ len(filtered_df_exploded)
|
1283 |
+
) * 100
|
1284 |
+
metric_completeness[get_human_friendly_metric_name(metric)] = (
|
1285 |
+
completeness
|
1286 |
+
)
|
1287 |
+
|
1288 |
+
if metric_completeness:
|
1289 |
+
completeness_df = pd.DataFrame(
|
1290 |
+
list(metric_completeness.items()),
|
1291 |
+
columns=["Metric", "Completeness (%)"],
|
1292 |
+
)
|
1293 |
+
fig = px.bar(
|
1294 |
+
completeness_df,
|
1295 |
+
x="Metric",
|
1296 |
+
y="Completeness (%)",
|
1297 |
+
title="Data Completeness by Metric",
|
1298 |
+
color="Completeness (%)",
|
1299 |
+
color_continuous_scale="Viridis",
|
1300 |
+
)
|
1301 |
+
fig.update_layout(xaxis_tickangle=-45, height=400)
|
1302 |
+
st.plotly_chart(fig, use_container_width=True)
|
1303 |
+
|
1304 |
|
1305 |
if __name__ == "__main__":
|
1306 |
main()
|