Jeremy Live commited on
Commit
82420e4
·
1 Parent(s): 1e24325
Files changed (1) hide show
  1. app.py +156 -151
app.py CHANGED
@@ -572,7 +572,6 @@ async def stream_agent_response(question: str, chat_history: List[List[str]]) ->
572
  response_text = str(response)
573
 
574
  logger.info(f"Extracted response text: {response_text[:200]}...")
575
-
576
  # Check if the response contains an SQL query and it truly looks like SQL
577
  sql_query = extract_sql_query(response_text)
578
  if sql_query and looks_like_sql(sql_query):
@@ -580,167 +579,173 @@ async def stream_agent_response(question: str, chat_history: List[List[str]]) ->
580
  # Execute the query and update the response
581
  db_connection, _ = setup_database_connection()
582
  if db_connection:
583
- query_result = execute_sql_query(sql_query, db_connection)
584
-
585
- # Add the query and its result to the response
586
- response_text += f"\n\n### 🔍 Resultado de la consulta:\n```sql\n{sql_query}\n```\n\n{query_result}"
587
-
588
- # Try to generate an interactive chart if the result is tabular
589
  try:
590
- if isinstance(query_result, str) and '|' in query_result and '---' in query_result:
591
- # Clean up the markdown table
592
- lines = [line.strip() for line in query_result.split('\n')
593
- if line.strip() and '---' not in line and '|' in line]
594
- if len(lines) > 1: # At least header + 1 data row
595
- # Get column names from the first line
596
- columns = [col.strip() for col in lines[0].split('|')[1:-1]]
597
- # Get data rows
598
- data = []
599
- for line in lines[1:]:
600
- values = [val.strip() for val in line.split('|')[1:-1]]
601
- if len(values) == len(columns):
602
- data.append(dict(zip(columns, values)))
603
-
604
- if data and len(columns) >= 2:
605
- # Determine chart type from user's question
606
- _, desired_type = detect_chart_preferences(question)
607
-
608
- # Choose x/y columns (assume first is category, second numeric)
609
- x_col = columns[0]
610
- y_col = columns[1]
611
 
612
- # Coerce numeric values for y
613
- for row in data:
614
- try:
615
- row[y_col] = float(re.sub(r"[^0-9.\-]", "", str(row[y_col])))
616
- except Exception:
617
- pass
618
-
619
- chart_fig = generate_chart(
620
- data=data,
621
- chart_type=desired_type,
622
- x=x_col,
623
- y=y_col,
624
- title=f"{y_col} por {x_col}"
625
- )
626
- if chart_fig is not None:
627
- logger.info(f"Chart generated from SQL table: type={desired_type}, x={x_col}, y={y_col}, rows={len(data)}")
628
- except Exception as e:
629
- logger.error(f"Error generating chart: {str(e)}", exc_info=True)
630
- # Don't fail the whole request if chart generation fails
631
- response_text += "\n\n⚠️ No se pudo generar la visualización de los datos."
632
- else:
633
- response_text += "\n\n⚠️ No se pudo conectar a la base de datos para ejecutar la consulta."
634
- elif sql_query and not looks_like_sql(sql_query):
635
- logger.info("Detected code block but it does not look like SQL; skipping execution.")
636
 
637
- # If we still have no chart but the user clearly wants one,
638
- # try a second pass to get ONLY a SQL query from the agent and execute it.
639
- if chart_fig is None:
640
- wants_chart, default_type = detect_chart_preferences(question)
641
- if wants_chart:
642
  try:
643
- logger.info("Second pass: asking agent for ONLY SQL query in fenced block.")
644
- sql_only_prompt = (
645
- "Devuelve SOLO la consulta SQL en un bloque ```sql``` para responder a: "
646
- f"{question}. No incluyas explicación ni texto adicional."
647
- )
648
- sql_only_resp = await agent.ainvoke({"input": sql_only_prompt})
649
- sql_only_text = str(sql_only_resp)
650
- sql_query2 = extract_sql_query(sql_only_text)
651
- if sql_query2 and looks_like_sql(sql_query2):
652
- logger.info(f"Second pass SQL detected: {sql_query2}")
653
- db_connection, _ = setup_database_connection()
654
- if db_connection:
655
- query_result = execute_sql_query(sql_query2, db_connection)
656
- # Try to parse table-like text into DataFrame if possible
657
- data = None
658
- if isinstance(query_result, str):
 
 
 
 
 
 
 
659
  try:
660
- import pandas as pd
661
- df = pd.read_csv(io.StringIO(query_result), sep="|")
662
- data = df
663
  except Exception:
664
  pass
665
- # As a fallback, don't rely on text table; just skip charting here
666
- if data is not None and hasattr(data, "empty") and not data.empty:
667
- # Heuristics: choose first column as x and second as y if numeric
668
- x_col = data.columns[0]
669
- # pick first numeric column different to x
670
- y_col = None
671
- for col in data.columns[1:]:
672
- try:
673
- pd.to_numeric(data[col])
674
- y_col = col
675
- break
676
- except Exception:
677
- continue
678
- if y_col:
679
- desired_type = default_type
680
- chart_fig = generate_chart(
681
- data=data,
682
- chart_type=desired_type,
683
- x=x_col,
684
- y=y_col,
685
- title=f"{y_col} por {x_col}"
686
- )
687
- if chart_fig is not None:
688
- logger.info("Chart generated from second-pass SQL execution.")
689
- else:
690
- logger.info("No DB connection on second pass; skipping.")
691
  except Exception as e:
692
- logger.error(f"Second-pass SQL synthesis failed: {e}")
693
-
694
- # Fallback: if user asked for a chart and we didn't get SQL or chart yet,
695
- # parse the most recent assistant text for lines like "LABEL: NUMBER" (bulleted or plain).
696
- if chart_fig is None:
697
- wants_chart, desired_type = detect_chart_preferences(question)
698
- if wants_chart:
699
- # Find the most recent assistant message with usable numeric pairs
700
- candidate_text = ""
701
- if chat_history:
702
- for pair in reversed(chat_history):
703
- if len(pair) >= 2 and isinstance(pair[1], str) and pair[1].strip():
704
- candidate_text = pair[1]
705
- break
706
- # Also consider current response_text as a data source
707
- if not candidate_text and isinstance(response_text, str) and response_text.strip():
708
- candidate_text = response_text
709
- if candidate_text:
710
- raw_lines = candidate_text.split('\n')
711
- # Normalize lines: strip bullets and markdown symbols
712
- norm_lines = []
713
- for l in raw_lines:
714
- s = l.strip()
715
- if not s:
716
- continue
717
- s = s.lstrip("•*-\t ")
718
- # Remove surrounding markdown emphasis from labels later
719
- norm_lines.append(s)
720
- data = []
721
- for l in norm_lines:
722
- # Accept patterns like "**LABEL**: 123" or "LABEL: 1,234"
723
- m = re.match(r"^(.+?):\s*([0-9][0-9.,]*)$", l)
724
- if m:
725
- label = m.group(1).strip()
726
- # Strip common markdown emphasis
727
- label = re.sub(r"[*_`]+", "", label).strip()
 
 
 
 
 
 
 
 
 
 
728
  try:
729
- val = float(m.group(2).replace(',', ''))
 
 
730
  except Exception:
731
  continue
732
- data.append({"label": label, "value": val})
733
- logger.info(f"Fallback parse from text: extracted {len(data)} items for potential chart")
734
- if len(data) >= 2:
735
- chart_fig = generate_chart(
736
- data=data,
737
- chart_type=desired_type,
738
- x="label",
739
- y="value",
740
- title="Distribución"
741
- )
742
- if chart_fig is not None:
743
- logger.info(f"Chart generated from text fallback: type={desired_type}, items={len(data)}")
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
744
 
745
  # Update the assistant's message with the response
746
  assistant_message["content"] = response_text
 
572
  response_text = str(response)
573
 
574
  logger.info(f"Extracted response text: {response_text[:200]}...")
 
575
  # Check if the response contains an SQL query and it truly looks like SQL
576
  sql_query = extract_sql_query(response_text)
577
  if sql_query and looks_like_sql(sql_query):
 
579
  # Execute the query and update the response
580
  db_connection, _ = setup_database_connection()
581
  if db_connection:
 
 
 
 
 
 
582
  try:
583
+ query_result = execute_sql_query(sql_query, db_connection)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
584
 
585
+ # Add the query and its result to the response
586
+ response_text += f"\n\n### 🔍 Resultado de la consulta:\n```sql\n{sql_query}\n```\n\n{query_result}"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
587
 
588
+ # Try to generate an interactive chart if the result is tabular
589
+ if isinstance(query_result, str) and '|' in query_result and '---' in query_result:
 
 
 
590
  try:
591
+ # Clean up the markdown table
592
+ lines = [line.strip() for line in query_result.split('\n')
593
+ if line.strip() and '---' not in line and '|' in line]
594
+ if len(lines) > 1: # At least header + 1 data row
595
+ # Get column names from the first line
596
+ columns = [col.strip() for col in lines[0].split('|')[1:-1]]
597
+ # Get data rows
598
+ data = []
599
+ for line in lines[1:]:
600
+ values = [val.strip() for val in line.split('|')[1:-1]]
601
+ if len(values) == len(columns):
602
+ data.append(dict(zip(columns, values)))
603
+
604
+ if data and len(columns) >= 2:
605
+ # Determine chart type from user's question
606
+ _, desired_type = detect_chart_preferences(question)
607
+
608
+ # Choose x/y columns (assume first is category, second numeric)
609
+ x_col = columns[0]
610
+ y_col = columns[1]
611
+
612
+ # Coerce numeric values for y
613
+ for row in data:
614
  try:
615
+ row[y_col] = float(re.sub(r"[^0-9.\-]", "", str(row[y_col])))
 
 
616
  except Exception:
617
  pass
618
+
619
+ chart_fig = generate_chart(
620
+ data=data,
621
+ chart_type=desired_type,
622
+ x=x_col,
623
+ y=y_col,
624
+ title=f"{y_col} por {x_col}"
625
+ )
626
+ if chart_fig is not None:
627
+ logger.info(
628
+ f"Chart generated from SQL table: type={desired_type}, x={x_col}, y={y_col}, rows={len(data)}"
629
+ )
 
 
 
 
 
 
 
 
 
 
 
 
 
 
630
  except Exception as e:
631
+ logger.error(f"Error generating chart: {str(e)}", exc_info=True)
632
+ # Don't fail the whole request if chart generation fails
633
+ response_text += "\n\n⚠️ No se pudo generar la visualización de los datos."
634
+ except Exception as e:
635
+ logger.error(f"Error handling SQL result: {e}", exc_info=True)
636
+ response_text += "\n\n⚠️ Ocurrió un error al procesar la consulta."
637
+ else:
638
+ response_text += "\n\n⚠️ No se pudo conectar a la base de datos para ejecutar la consulta."
639
+ elif sql_query and not looks_like_sql(sql_query):
640
+ logger.info("Detected code block but it does not look like SQL; skipping execution.")
641
+
642
+ # If we still have no chart but the user clearly wants one,
643
+ # try a second pass to get ONLY a SQL query from the agent and execute it.
644
+ if chart_fig is None:
645
+ wants_chart, default_type = detect_chart_preferences(question)
646
+ if wants_chart:
647
+ try:
648
+ logger.info("Second pass: asking agent for ONLY SQL query in fenced block.")
649
+ sql_only_prompt = (
650
+ "Devuelve SOLO la consulta SQL en un bloque ```sql``` para responder a: "
651
+ f"{question}. No incluyas explicación ni texto adicional."
652
+ )
653
+ sql_only_resp = await agent.ainvoke({"input": sql_only_prompt})
654
+ sql_only_text = str(sql_only_resp)
655
+ sql_query2 = extract_sql_query(sql_only_text)
656
+ if sql_query2 and looks_like_sql(sql_query2):
657
+ logger.info(f"Second pass SQL detected: {sql_query2}")
658
+ db_connection, _ = setup_database_connection()
659
+ if db_connection:
660
+ query_result = execute_sql_query(sql_query2, db_connection)
661
+ # Try to parse table-like text into DataFrame if possible
662
+ data = None
663
+ if isinstance(query_result, str):
664
+ try:
665
+ import pandas as pd
666
+ df = pd.read_csv(io.StringIO(query_result), sep="|")
667
+ data = df
668
+ except Exception:
669
+ pass
670
+ # As a fallback, don't rely on text table; just skip charting here
671
+ if data is not None and hasattr(data, "empty") and not data.empty:
672
+ # Heuristics: choose first column as x and second as y if numeric
673
+ x_col = data.columns[0]
674
+ # pick first numeric column different to x
675
+ y_col = None
676
+ for col in data.columns[1:]:
677
  try:
678
+ pd.to_numeric(data[col])
679
+ y_col = col
680
+ break
681
  except Exception:
682
  continue
683
+ if y_col:
684
+ desired_type = default_type
685
+ chart_fig = generate_chart(
686
+ data=data,
687
+ chart_type=desired_type,
688
+ x=x_col,
689
+ y=y_col,
690
+ title=f"{y_col} por {x_col}"
691
+ )
692
+ if chart_fig is not None:
693
+ logger.info("Chart generated from second-pass SQL execution.")
694
+ else:
695
+ logger.info("No DB connection on second pass; skipping.")
696
+ except Exception as e:
697
+ logger.error(f"Second-pass SQL synthesis failed: {e}")
698
+
699
+ # Fallback: if user asked for a chart and we didn't get SQL or chart yet,
700
+ # parse the most recent assistant text for lines like "LABEL: NUMBER" (bulleted or plain).
701
+ if chart_fig is None:
702
+ wants_chart, desired_type = detect_chart_preferences(question)
703
+ if wants_chart:
704
+ # Find the most recent assistant message with usable numeric pairs
705
+ candidate_text = ""
706
+ if chat_history:
707
+ for pair in reversed(chat_history):
708
+ if len(pair) >= 2 and isinstance(pair[1], str) and pair[1].strip():
709
+ candidate_text = pair[1]
710
+ break
711
+ # Also consider current response_text as a data source
712
+ if not candidate_text and isinstance(response_text, str) and response_text.strip():
713
+ candidate_text = response_text
714
+ if candidate_text:
715
+ raw_lines = candidate_text.split('\n')
716
+ # Normalize lines: strip bullets and markdown symbols
717
+ norm_lines = []
718
+ for l in raw_lines:
719
+ s = l.strip()
720
+ if not s:
721
+ continue
722
+ s = s.lstrip("•*-\t ")
723
+ # Remove surrounding markdown emphasis from labels later
724
+ norm_lines.append(s)
725
+ data = []
726
+ for l in norm_lines:
727
+ # Accept patterns like "**LABEL**: 123" or "LABEL: 1,234"
728
+ m = re.match(r"^(.+?):\s*([0-9][0-9.,]*)$", l)
729
+ if m:
730
+ label = m.group(1).strip()
731
+ # Strip common markdown emphasis
732
+ label = re.sub(r"[*_`]+", "", label).strip()
733
+ try:
734
+ val = float(m.group(2).replace(',', ''))
735
+ except Exception:
736
+ continue
737
+ data.append({"label": label, "value": val})
738
+ logger.info(f"Fallback parse from text: extracted {len(data)} items for potential chart")
739
+ if len(data) >= 2:
740
+ chart_fig = generate_chart(
741
+ data=data,
742
+ chart_type=desired_type,
743
+ x="label",
744
+ y="value",
745
+ title="Distribución"
746
+ )
747
+ if chart_fig is not None:
748
+ logger.info(f"Chart generated from text fallback: type={desired_type}, items={len(data)}")
749
 
750
  # Update the assistant's message with the response
751
  assistant_message["content"] = response_text