Spaces:
Running
on
T4
Running
on
T4
Nikita
commited on
Commit
·
7fb6e58
1
Parent(s):
bf87f19
second round of edits from günther and andreas
Browse files- app.py +128 -76
- data/.DS_Store +0 -0
- data/{air_passengers_forecast_256.pt → air_passengers_forecast_512.pt} +0 -0
- data/ett2.csv +0 -0
- data/ett2_forecast_512.pt +0 -0
- data/loop.csv +0 -0
- data/loop_forecast_512.pt +0 -0
- data/merged_ett2_loop.csv +0 -0
- data/merged_ett2_loop_forecast_256.pt +0 -0
app.py
CHANGED
@@ -6,7 +6,9 @@ from PIL import Image
|
|
6 |
import numpy as np
|
7 |
import gradio as gr
|
8 |
import os
|
9 |
-
from
|
|
|
|
|
10 |
|
11 |
# ----------------------------
|
12 |
# Helper functions (logic mostly unchanged)
|
@@ -15,21 +17,23 @@ from tirex import load_model, ForecastModel
|
|
15 |
torch.manual_seed(42)
|
16 |
|
17 |
def model_forecast(input_data, forecast_length=256, file_name=None):
|
18 |
-
if os.path.basename(file_name) == "
|
19 |
-
_forecast_tensor = torch.load("data/
|
|
|
|
|
|
|
20 |
return _forecast_tensor[:,:forecast_length,:]
|
21 |
elif os.path.basename(file_name) == "air_passangers.csv":
|
22 |
-
_forecast_tensor = torch.load("data/
|
23 |
return _forecast_tensor[:,:forecast_length,:]
|
24 |
else:
|
25 |
-
model: ForecastModel = load_model("NX-AI/TiRex",device='cuda')
|
26 |
-
forecast = model.forecast(context=input_data, prediction_length=forecast_length)
|
27 |
-
return forecast[0]
|
|
|
28 |
|
29 |
|
30 |
|
31 |
-
|
32 |
-
|
33 |
def plot_forecast_plotly(timeseries, quantile_predictions, timeseries_name):
|
34 |
"""
|
35 |
- timeseries: 1D list/array of historical values.
|
@@ -66,7 +70,7 @@ def plot_forecast_plotly(timeseries, quantile_predictions, timeseries_name):
|
|
66 |
y=lower_q,
|
67 |
mode="lines",
|
68 |
line=dict(color="rgba(0, 0, 0, 0)", width=0),
|
69 |
-
name=f"{timeseries_name} –
|
70 |
hovertemplate="Lower: %{y:.2f}<extra></extra>"
|
71 |
))
|
72 |
|
@@ -78,7 +82,7 @@ def plot_forecast_plotly(timeseries, quantile_predictions, timeseries_name):
|
|
78 |
line=dict(color="rgba(0, 0, 0, 0)", width=0),
|
79 |
fill="tonexty",
|
80 |
fillcolor="rgba(128, 128, 128, 0.3)",
|
81 |
-
name=f"{timeseries_name} –
|
82 |
hovertemplate="Upper: %{y:.2f}<extra></extra>"
|
83 |
))
|
84 |
|
@@ -139,24 +143,6 @@ def load_table(file_path):
|
|
139 |
raise ValueError("Unsupported format. Use CSV, XLS, XLSX, or PARQUET.")
|
140 |
|
141 |
|
142 |
-
# def extract_names_and_update(file, preset_filename):
|
143 |
-
# try:
|
144 |
-
# if file is not None:
|
145 |
-
# df = load_table(file.name)
|
146 |
-
# else:
|
147 |
-
# if not preset_filename:
|
148 |
-
# return gr.update(choices=[], value=[]), []
|
149 |
-
# df = load_table(preset_filename)
|
150 |
-
|
151 |
-
# if df.shape[1] > 0 and df.iloc[:, 0].dtype == object and not df.iloc[:, 0].str.isnumeric().all():
|
152 |
-
# names = df.iloc[:, 0].tolist()
|
153 |
-
# else:
|
154 |
-
# names = [f"Series {i}" for i in range(len(df))]
|
155 |
-
# return gr.update(choices=names, value=names), names
|
156 |
-
# except Exception:
|
157 |
-
# return gr.update(choices=[], value=[]), []
|
158 |
-
|
159 |
-
|
160 |
def extract_names_and_update(file, preset_filename):
|
161 |
try:
|
162 |
# Determine which file to use and get default forecast length
|
@@ -206,7 +192,7 @@ def get_default_forecast_length(file_path):
|
|
206 |
return 64
|
207 |
|
208 |
filename = os.path.basename(file_path)
|
209 |
-
if filename == "
|
210 |
return 256
|
211 |
elif filename == "air_passangers.csv":
|
212 |
return 48
|
@@ -216,17 +202,19 @@ def get_default_forecast_length(file_path):
|
|
216 |
|
217 |
def display_filtered_forecast(file, preset_filename, selected_names, forecast_length):
|
218 |
try:
|
219 |
-
# If no file
|
220 |
if file is None and (preset_filename is None or preset_filename == "-- No preset selected --"):
|
221 |
return None, "No file selected."
|
222 |
|
223 |
-
# Load
|
224 |
if file is not None:
|
225 |
df = load_table(file.name)
|
|
|
226 |
else:
|
227 |
df = load_table(preset_filename)
|
|
|
228 |
|
229 |
-
# Determine names
|
230 |
if (
|
231 |
df.shape[1] > 0
|
232 |
and df.iloc[:, 0].dtype == object
|
@@ -238,63 +226,128 @@ def display_filtered_forecast(file, preset_filename, selected_names, forecast_le
|
|
238 |
all_names = [f"Series {i}" for i in range(len(df))]
|
239 |
data_only = df.astype(float)
|
240 |
|
241 |
-
# Build
|
242 |
mask = [name in selected_names for name in all_names]
|
243 |
if not any(mask):
|
244 |
return None, "No timeseries chosen to plot."
|
245 |
|
246 |
-
|
247 |
-
filtered_data = data_only.iloc[mask, :].values
|
248 |
filtered_names = [all_names[i] for i, m in enumerate(mask) if m]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
249 |
|
250 |
-
|
251 |
-
file_path = file.name if file is not None else preset_filename
|
252 |
-
out = model_forecast(filtered_data, forecast_length=forecast_length, file_name=file_path)
|
253 |
-
inp = torch.tensor(filtered_data) # shape = (n_selected, seq_len)
|
254 |
-
|
255 |
-
# If only one series is selected, we can just call plot_forecast_plotly directly:
|
256 |
-
if inp.shape[0] == 1:
|
257 |
-
ts = inp[0].numpy().tolist()
|
258 |
-
qp = out[0].numpy()
|
259 |
-
fig = plot_forecast_plotly(ts, qp, filtered_names[0])
|
260 |
-
return fig, ""
|
261 |
-
|
262 |
-
# If multiple series are selected, build a master figure by concatenating traces
|
263 |
-
master_fig = go.Figure()
|
264 |
-
for idx in range(inp.shape[0]):
|
265 |
ts = inp[idx].numpy().tolist()
|
266 |
qp = out[idx].numpy()
|
267 |
series_name = filtered_names[idx]
|
268 |
|
269 |
-
#
|
270 |
-
|
271 |
-
|
272 |
-
|
273 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
274 |
|
275 |
-
#
|
276 |
-
|
277 |
template="plotly_dark",
|
|
|
278 |
title=dict(
|
279 |
text="Forecasts for Selected Timeseries",
|
280 |
x=0.5,
|
281 |
-
font=dict(size=
|
282 |
),
|
283 |
-
xaxis=dict(
|
284 |
-
rangeslider=dict(visible=True), # <-- put rangeslider here
|
285 |
-
fixedrange=False
|
286 |
-
),
|
287 |
-
xaxis_title="Time",
|
288 |
-
yaxis_title="Value",
|
289 |
hovermode="x unified",
|
290 |
-
|
291 |
-
|
292 |
-
autosize=True,
|
293 |
)
|
294 |
-
|
|
|
295 |
|
296 |
except Exception as e:
|
297 |
-
return None, f"Error: {e}
|
|
|
298 |
|
299 |
|
300 |
# ----------------------------
|
@@ -313,7 +366,7 @@ with gr.Blocks(fill_width=True,theme=gr.themes.Ocean()) as demo:
|
|
313 |
label="Upload CSV / XLSX / PARQUET",
|
314 |
file_types=[".csv", ".xls", ".xlsx", ".parquet"]
|
315 |
)
|
316 |
-
preset_choices = ["-- No preset selected --", "data/
|
317 |
|
318 |
preset_dropdown = gr.Dropdown(
|
319 |
label="Or choose a preset:",
|
@@ -347,7 +400,6 @@ with gr.Blocks(fill_width=True,theme=gr.themes.Ocean()) as demo:
|
|
347 |
gr.Image("static/nxai_logo.png", width=150, show_label=False, container=False)
|
348 |
gr.Image("static/tirex.jpeg", width=150, show_label=False, container=False)
|
349 |
|
350 |
-
# Right column: interactive plot + instructions
|
351 |
with gr.Column(scale=5):
|
352 |
gr.Markdown("## Forecast Plot")
|
353 |
plot_output = gr.Plot()
|
@@ -418,10 +470,10 @@ with gr.Blocks(fill_width=True,theme=gr.themes.Ocean()) as demo:
|
|
418 |
|
419 |
# Plot button
|
420 |
plot_button.click(
|
421 |
-
|
422 |
-
|
423 |
-
|
424 |
-
|
425 |
demo.launch()
|
426 |
|
427 |
|
|
|
6 |
import numpy as np
|
7 |
import gradio as gr
|
8 |
import os
|
9 |
+
from plotly.subplots import make_subplots
|
10 |
+
|
11 |
+
# from tirex import load_model, ForecastModel
|
12 |
|
13 |
# ----------------------------
|
14 |
# Helper functions (logic mostly unchanged)
|
|
|
17 |
torch.manual_seed(42)
|
18 |
|
19 |
def model_forecast(input_data, forecast_length=256, file_name=None):
|
20 |
+
if os.path.basename(file_name) == "loop.csv":
|
21 |
+
_forecast_tensor = torch.load("data/loop_forecast_512.pt")
|
22 |
+
return _forecast_tensor[:,:forecast_length,:]
|
23 |
+
elif os.path.basename(file_name) == "ett2.csv":
|
24 |
+
_forecast_tensor = torch.load("data/ett2_forecast_512.pt")
|
25 |
return _forecast_tensor[:,:forecast_length,:]
|
26 |
elif os.path.basename(file_name) == "air_passangers.csv":
|
27 |
+
_forecast_tensor = torch.load("data/air_passengers_forecast_512.pt")
|
28 |
return _forecast_tensor[:,:forecast_length,:]
|
29 |
else:
|
30 |
+
# model: ForecastModel = load_model("NX-AI/TiRex",device='cuda')
|
31 |
+
# forecast = model.forecast(context=input_data, prediction_length=forecast_length)
|
32 |
+
# return forecast[0]
|
33 |
+
pass
|
34 |
|
35 |
|
36 |
|
|
|
|
|
37 |
def plot_forecast_plotly(timeseries, quantile_predictions, timeseries_name):
|
38 |
"""
|
39 |
- timeseries: 1D list/array of historical values.
|
|
|
70 |
y=lower_q,
|
71 |
mode="lines",
|
72 |
line=dict(color="rgba(0, 0, 0, 0)", width=0),
|
73 |
+
name=f"{timeseries_name} – 10% Quantile",
|
74 |
hovertemplate="Lower: %{y:.2f}<extra></extra>"
|
75 |
))
|
76 |
|
|
|
82 |
line=dict(color="rgba(0, 0, 0, 0)", width=0),
|
83 |
fill="tonexty",
|
84 |
fillcolor="rgba(128, 128, 128, 0.3)",
|
85 |
+
name=f"{timeseries_name} – 90% Quantile",
|
86 |
hovertemplate="Upper: %{y:.2f}<extra></extra>"
|
87 |
))
|
88 |
|
|
|
143 |
raise ValueError("Unsupported format. Use CSV, XLS, XLSX, or PARQUET.")
|
144 |
|
145 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
146 |
def extract_names_and_update(file, preset_filename):
|
147 |
try:
|
148 |
# Determine which file to use and get default forecast length
|
|
|
192 |
return 64
|
193 |
|
194 |
filename = os.path.basename(file_path)
|
195 |
+
if filename == "loop.csv" or filename == "ett2.csv":
|
196 |
return 256
|
197 |
elif filename == "air_passangers.csv":
|
198 |
return 48
|
|
|
202 |
|
203 |
def display_filtered_forecast(file, preset_filename, selected_names, forecast_length):
|
204 |
try:
|
205 |
+
# 1) If no file or preset selected, show an error
|
206 |
if file is None and (preset_filename is None or preset_filename == "-- No preset selected --"):
|
207 |
return None, "No file selected."
|
208 |
|
209 |
+
# 2) Load DataFrame and remember which filename to pass to model_forecast
|
210 |
if file is not None:
|
211 |
df = load_table(file.name)
|
212 |
+
file_name = file.name
|
213 |
else:
|
214 |
df = load_table(preset_filename)
|
215 |
+
file_name = preset_filename
|
216 |
|
217 |
+
# 3) Determine whether first column is names or numeric
|
218 |
if (
|
219 |
df.shape[1] > 0
|
220 |
and df.iloc[:, 0].dtype == object
|
|
|
226 |
all_names = [f"Series {i}" for i in range(len(df))]
|
227 |
data_only = df.astype(float)
|
228 |
|
229 |
+
# 4) Build mask from selected_names
|
230 |
mask = [name in selected_names for name in all_names]
|
231 |
if not any(mask):
|
232 |
return None, "No timeseries chosen to plot."
|
233 |
|
234 |
+
filtered_data = data_only.iloc[mask, :].values # shape = (n_selected, seq_len)
|
|
|
235 |
filtered_names = [all_names[i] for i, m in enumerate(mask) if m]
|
236 |
+
n_selected = filtered_data.shape[0]
|
237 |
+
|
238 |
+
# 5) First call model_forecast on all series, then select only the masked rows
|
239 |
+
full_data = data_only.values # shape = (n_all, seq_len)
|
240 |
+
full_out = model_forecast(full_data, forecast_length=forecast_length, file_name=file_name)
|
241 |
+
|
242 |
+
# Now pick only the rows we actually filtered
|
243 |
+
out = full_out[mask, :, :] # shape = (n_selected, pred_len, n_q)
|
244 |
+
inp = torch.tensor(filtered_data)
|
245 |
+
|
246 |
+
# 6) Create one subplot per selected series, with vertical spacing
|
247 |
+
fig = make_subplots(
|
248 |
+
rows=n_selected,
|
249 |
+
cols=1,
|
250 |
+
shared_xaxes=False,
|
251 |
+
vertical_spacing=0.3, # more space between subplots
|
252 |
+
subplot_titles=filtered_names
|
253 |
+
)
|
254 |
|
255 |
+
for idx in range(n_selected):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
256 |
ts = inp[idx].numpy().tolist()
|
257 |
qp = out[idx].numpy()
|
258 |
series_name = filtered_names[idx]
|
259 |
|
260 |
+
# a) plot historical data (blue line)
|
261 |
+
x_hist = list(range(len(ts)))
|
262 |
+
fig.add_trace(
|
263 |
+
go.Scatter(
|
264 |
+
x=x_hist,
|
265 |
+
y=ts,
|
266 |
+
mode="lines",
|
267 |
+
name=f"{series_name} – Given Data",
|
268 |
+
line=dict(color="blue", width=2),
|
269 |
+
showlegend=False
|
270 |
+
),
|
271 |
+
row=idx + 1, col=1
|
272 |
+
)
|
273 |
+
|
274 |
+
# b) compute forecast indices
|
275 |
+
pred_len = qp.shape[0]
|
276 |
+
x_pred = list(range(len(ts) - 1, len(ts) - 1 + pred_len))
|
277 |
+
|
278 |
+
lower_q = qp[:, 0]
|
279 |
+
upper_q = qp[:, -1]
|
280 |
+
n_q = qp.shape[1]
|
281 |
+
median_idx = n_q // 2
|
282 |
+
median_q = qp[:, median_idx]
|
283 |
+
|
284 |
+
# c) lower‐bound (invisible)
|
285 |
+
fig.add_trace(
|
286 |
+
go.Scatter(
|
287 |
+
x=x_pred,
|
288 |
+
y=lower_q,
|
289 |
+
mode="lines",
|
290 |
+
line=dict(color="rgba(0,0,0,0)", width=0),
|
291 |
+
name=f"{series_name} – 10% Quantile",
|
292 |
+
hovertemplate="10% Quantile: %{y:.2f}<extra></extra>",
|
293 |
+
showlegend=False
|
294 |
+
),
|
295 |
+
row=idx + 1, col=1
|
296 |
+
)
|
297 |
+
|
298 |
+
# d) upper‐bound (shaded area)
|
299 |
+
fig.add_trace(
|
300 |
+
go.Scatter(
|
301 |
+
x=x_pred,
|
302 |
+
y=upper_q,
|
303 |
+
mode="lines",
|
304 |
+
line=dict(color="rgba(0,0,0,0)", width=0),
|
305 |
+
fill="tonexty",
|
306 |
+
fillcolor="rgba(128,128,128,0.3)",
|
307 |
+
name=f"{series_name} – 90% Quantile",
|
308 |
+
hovertemplate="90% Quantile: %{y:.2f}<extra></extra>",
|
309 |
+
showlegend=False
|
310 |
+
),
|
311 |
+
row=idx + 1, col=1
|
312 |
+
)
|
313 |
+
|
314 |
+
# e) median forecast (orange line)
|
315 |
+
fig.add_trace(
|
316 |
+
go.Scatter(
|
317 |
+
x=x_pred,
|
318 |
+
y=median_q,
|
319 |
+
mode="lines",
|
320 |
+
name=f"{series_name} – Median Forecast",
|
321 |
+
line=dict(color="orange", width=2),
|
322 |
+
hovertemplate="Median: %{y:.2f}<extra></extra>",
|
323 |
+
showlegend=False
|
324 |
+
),
|
325 |
+
row=idx + 1, col=1
|
326 |
+
)
|
327 |
+
|
328 |
+
# f) label axes for each subplot
|
329 |
+
fig.update_xaxes(title_text="Time", row=idx + 1, col=1)
|
330 |
+
fig.update_yaxes(title_text="Value", row=idx + 1, col=1)
|
331 |
|
332 |
+
# 7) Global layout tweaks
|
333 |
+
fig.update_layout(
|
334 |
template="plotly_dark",
|
335 |
+
height=300 * n_selected, # 300px per subplot
|
336 |
title=dict(
|
337 |
text="Forecasts for Selected Timeseries",
|
338 |
x=0.5,
|
339 |
+
font=dict(size=20, family="Arial", color="white")
|
340 |
),
|
|
|
|
|
|
|
|
|
|
|
|
|
341 |
hovermode="x unified",
|
342 |
+
margin=dict(t=120, b=40, l=60, r=40),
|
343 |
+
showlegend=False
|
|
|
344 |
)
|
345 |
+
|
346 |
+
return fig, ""
|
347 |
|
348 |
except Exception as e:
|
349 |
+
return None, f"Error: {str(e)}"
|
350 |
+
|
351 |
|
352 |
|
353 |
# ----------------------------
|
|
|
366 |
label="Upload CSV / XLSX / PARQUET",
|
367 |
file_types=[".csv", ".xls", ".xlsx", ".parquet"]
|
368 |
)
|
369 |
+
preset_choices = ["-- No preset selected --", "data/loop.csv", "data/air_passangers.csv", 'data/ett2.csv']
|
370 |
|
371 |
preset_dropdown = gr.Dropdown(
|
372 |
label="Or choose a preset:",
|
|
|
400 |
gr.Image("static/nxai_logo.png", width=150, show_label=False, container=False)
|
401 |
gr.Image("static/tirex.jpeg", width=150, show_label=False, container=False)
|
402 |
|
|
|
403 |
with gr.Column(scale=5):
|
404 |
gr.Markdown("## Forecast Plot")
|
405 |
plot_output = gr.Plot()
|
|
|
470 |
|
471 |
# Plot button
|
472 |
plot_button.click(
|
473 |
+
fn=display_filtered_forecast,
|
474 |
+
inputs=[file_input, preset_dropdown, filter_checkbox, forecast_length_slider],
|
475 |
+
outputs=[plot_output, errbox]
|
476 |
+
)
|
477 |
demo.launch()
|
478 |
|
479 |
|
data/.DS_Store
CHANGED
Binary files a/data/.DS_Store and b/data/.DS_Store differ
|
|
data/{air_passengers_forecast_256.pt → air_passengers_forecast_512.pt}
RENAMED
Binary files a/data/air_passengers_forecast_256.pt and b/data/air_passengers_forecast_512.pt differ
|
|
data/ett2.csv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/ett2_forecast_512.pt
ADDED
Binary file (38.1 kB). View file
|
|
data/loop.csv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
data/loop_forecast_512.pt
ADDED
Binary file (38.1 kB). View file
|
|
data/merged_ett2_loop.csv
DELETED
The diff for this file is too large to render.
See raw diff
|
|
data/merged_ett2_loop_forecast_256.pt
DELETED
Binary file (38.2 kB)
|
|