import os import requests import joblib import time import matplotlib.dates as mdates from datetime import datetime, timedelta from zoneinfo import ZoneInfo from dateutil import parser as date_parser from typing import Union from geopy.geocoders import Nominatim from timezonefinder import TimezoneFinder from tabulate import tabulate import pandas as pd import matplotlib.pyplot as plt from langchain.schema import AIMessage try: from langchain_openai import ChatOpenAI except ImportError: from langchain.chat_models import ChatOpenAI #os.environ["OPENAI_API_KEY"] = "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" #os.environ["WEATHER_API_KEY"] = "b9b44f4c0e8949bb95a90524250204" os.environ["OPENAI_API_KEY"] = os.getenv("OPENAI_API_KEY") os.environ["WEATHER_API_KEY"] = os.getenv("WEATHER_API_KEY") try: llm_gpt4 = ChatOpenAI(model="gpt-4o-mini", temperature=0) except Exception: try: llm_gpt4 = ChatOpenAI(model_name="gpt-4o-mini", temperature=0) except Exception as e: print(f"LangChain initialization failed: {e}") llm_gpt4 = None def location_to_timezone(location: str) -> str: try: geo = Nominatim(user_agent="time_agent_demo") loc = geo.geocode(location) if not loc: return "Europe/London" tf = TimezoneFinder() return tf.timezone_at(lng=loc.longitude, lat=loc.latitude) or "Europe/London" except Exception: return "Europe/London" def get_time_tool2(query: str) -> tuple[datetime, int, str]: try: location_prompt = f""" You are a location extractor. Given a user's query about time or date, return the location mentioned in it. If not found, return "London". Query: "{query}" """ location_response = llm_gpt4.invoke(location_prompt) location = location_response.content.strip() if isinstance(location_response, AIMessage) else str(location_response).strip() #print(f"[DEBUG] Extracted Location: {location}") tz_str = location_to_timezone(location) #print(f"[DEBUG] Timezone: {tz_str}") now = datetime.now(ZoneInfo(tz_str)) #print(f"[DEBUG] Local Time at {location}: {now}") examples = [ # Pure hourly relative expression ("five hours later", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 5"), ("later", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 2"), ("soon", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("shortly", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("after a while", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), # Expressions throughout the day (no specific time) ("today", f"START_TIME: {now.replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("tomorrow", f"START_TIME: {(now + timedelta(days=1)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("yesterday", f"START_TIME: {(now - timedelta(days=1)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("the day before yesterday", f"START_TIME: {(now - timedelta(days=2)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("the day after tomorrow", f"START_TIME: {(now + timedelta(days=2)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("昨天", f"START_TIME: {(now - timedelta(days=1)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("今天", f"START_TIME: {now.replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("明天", f"START_TIME: {(now + timedelta(days=1)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("前天", f"START_TIME: {(now - timedelta(days=2)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), # Specific time point (single point) ("tomorrow at 3pm", f"START_TIME: {(now + timedelta(days=1)).replace(hour=15,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("tomorrow at 10am", f"START_TIME: {(now + timedelta(days=1)).replace(hour=10,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("yesterday at 5pm", f"START_TIME: {(now - timedelta(days=1)).replace(hour=17,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("today at 6pm", f"START_TIME: {now.replace(hour=18,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("the day after tomorrow at 10am", f"START_TIME: {(now + timedelta(days=2)).replace(hour=10,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("昨天下午五點", f"START_TIME: {(now - timedelta(days=1)).replace(hour=17,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("昨天早上八點", f"START_TIME: {(now - timedelta(days=1)).replace(hour=8,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("下週一下午三點", f"START_TIME: {(now + timedelta(days=(7 - now.weekday() + 0) % 7)).replace(hour=15,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("last Monday 9am", f"START_TIME: {(now - timedelta(days=(now.weekday() + 7))).replace(hour=9,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), # Day of the week (all day) ("next Monday", f"START_TIME: {(now + timedelta(days=(7 - now.weekday()))).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("last Friday", f"START_TIME: {(now - timedelta(days=(now.weekday() - 4 + 7) % 7)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("next Friday", f"START_TIME: {(now + timedelta(days=(4 - now.weekday() + 7) % 7)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("this Monday", f"START_TIME: {(now - timedelta(days=now.weekday())).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("this Sunday", f"START_TIME: {(now + timedelta(days=(6 - now.weekday()))).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), # X days before/after (all day) ("5 days ago", f"START_TIME: {(now - timedelta(days=5)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("3 days ago", f"START_TIME: {(now - timedelta(days=3)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("in 5 days", f"START_TIME: {(now + timedelta(days=5)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), # Multi-day range ("past 3 days", f"START_TIME: {(now - timedelta(days=2)).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 72"), ("last 5 days", f"START_TIME: {(now - timedelta(days=4)).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 120"), ("next 7 days", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 168"), ("past 3 days", f"START_TIME: {(now - timedelta(days=3)).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 72"), ("last 5 days", f"START_TIME: {(now - timedelta(days=5)).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 120"), ("last 10 days", f"START_TIME: {(now - timedelta(days=10)).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 240"), # Weekly/Monthly Range ("this week", f"START_TIME: {(now - timedelta(days=now.weekday())).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 168"), ("last week", f"START_TIME: {(now - timedelta(days=now.weekday() + 7)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 168"), ("next week", f"START_TIME: {(now + timedelta(days=(7 - now.weekday()))).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 168"), ("last month", f"START_TIME: {(now - timedelta(days=30)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 720"), ("本週", f"START_TIME: {(now - timedelta(days=now.weekday())).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 168"), ("上週", f"START_TIME: {(now - timedelta(days=now.weekday() + 7)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 168"), # Past X hours ("過去 24 小時", f"START_TIME: {(now - timedelta(hours=24)).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 24"), ("past 48 hours", f"START_TIME: {(now - timedelta(hours=48)).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 48"), ("last 12 hours", f"START_TIME: {(now - timedelta(hours=12)).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 12"), # The next X hours ("in 10 hours", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 10"), ("in 2 hours", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 2"), ("in one hour", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("next 2 hours", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 2"), ("next 8 hours", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 8"), # A few minutes of expression (regarded as one hour) ("in 30 minutes", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("in a few minutes", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), # Special Segment ("later this evening", f"START_TIME: {now.replace(hour=20,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("this weekend", f"START_TIME: {(now + timedelta(days=(5 - now.weekday()) % 7)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 48"), ("next weekend", f"START_TIME: {(now + timedelta(days=((5 - now.weekday()) % 7) + 7)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 48"), ("tonight", f"START_TIME: {now.replace(hour=20,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 4"), ("this morning", f"START_TIME: {now.replace(hour=6,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 6"), ("this afternoon", f"START_TIME: {now.replace(hour=12,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 6"), # Current Time ("now", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("right now", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), ("現在", f"START_TIME: {now.strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 1"), # Scope of Expression ("from today 15:00 to 20:00", f"START_TIME: {now.replace(hour=15,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 5"), ("從今天下午 3 點到晚上 8 點", f"START_TIME: {now.replace(hour=15,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 5"), ("from tomorrow 14:00 to tomorrow 18:00", f"START_TIME: {(now + timedelta(days=1)).replace(hour=14,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 4"), ("from 3pm today to 2am tomorrow", f"START_TIME: {now.replace(hour=15,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 11"), ("from 4pm to 8pm today", f"START_TIME: {now.replace(hour=16,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 4"), ("Provide a 6-day weather summary for Tokyo ending today.", f"START_TIME: {(now - timedelta(days=5)).replace(hour=0,minute=0).strftime('%Y-%m-%d %H:%M')}\nDURATION_HOURS: 144"), ] # few‐shot prompt examples_header = f"""Assume the current local time in {location} is exactly: **{now.strftime('%Y-%m-%d %H:%M')}** (timezone: {tz_str}) Use this exact time to reason all examples below. """ examples_str = "\n".join([f'User Query: "{q}"\n→ {out}' for q, out in examples]) time_query_prompt = f""" You are a timezone-aware time reasoner. Based on the user's query, calculate: 1. START_TIME (format: YYYY-MM-DD HH:MM) in the local timezone. 2. DURATION_HOURS (integer hours) for how many hours this query spans. Examples: {examples_str} Now process: User Query: "{query}" → """ time_response = llm_gpt4.invoke(time_query_prompt) resp_lines = time_response.content.strip().splitlines() # # Parse START_TIME and DURATION_HOURS from the return. start_time = now duration_hours = 1 for line in resp_lines: if line.startswith("START_TIME:"): t_str = line.split(":", 1)[1].strip() try: start_time = datetime.strptime(t_str, "%Y-%m-%d %H:%M") start_time = start_time.replace(tzinfo=ZoneInfo(tz_str)) except: pass elif line.startswith("DURATION_HOURS:"): try: duration_hours = int(line.split(":", 1)[1].strip()) except: pass return start_time, duration_hours, location except Exception as e: # If parsing fails, fallback to "current time + 1 hour". tz_str_fallback = "Europe/London" try: tz_str_fallback = location_to_timezone(location) except: pass now = datetime.now(ZoneInfo(tz_str_fallback)) return now, 1, location def render_chart(df: pd.DataFrame, location: str, title: str = "Weather Forecast") -> str: try: chart_path = "/tmp/weather_chart.png" if os.path.exists(chart_path): os.remove(chart_path) # Converting time fields to datetime objects if "time" in df.columns: df_plot = df.copy() df_plot["datetime"] = pd.to_datetime(df_plot["time"]) # Create Submap fig, axes = plt.subplots(2, 2, figsize=(15, 10)) fig.suptitle(f"{title}", fontsize=16) # Temperature Chart if "temp_c" in df_plot.columns: axes[0, 0].plot(df_plot["datetime"], df_plot["temp_c"], marker='o', color='red', linewidth=2, markersize=4) axes[0, 0].set_title("Temperature (°C)") axes[0, 0].set_ylabel("Temperature (°C)") axes[0, 0].grid(True, alpha=0.3) axes[0, 0].tick_params(axis='x', rotation=45) # Humidity Chart if "humidity" in df_plot.columns: axes[0, 1].plot(df_plot["datetime"], df_plot["humidity"], marker='s', color='blue', linewidth=2, markersize=4) axes[0, 1].set_title("Humidity (%)") axes[0, 1].set_ylabel("Humidity (%)") axes[0, 1].grid(True, alpha=0.3) axes[0, 1].tick_params(axis='x', rotation=45) # Wind Speed Chart if "wind_kph" in df_plot.columns: axes[1, 0].plot(df_plot["datetime"], df_plot["wind_kph"], marker='^', color='green', linewidth=2, markersize=4) axes[1, 0].set_title("Wind Speed (kph)") axes[1, 0].set_ylabel("Wind Speed (kph)") axes[1, 0].grid(True, alpha=0.3) axes[1, 0].tick_params(axis='x', rotation=45) # Rainfall Probability Chart if "chance_of_rain" in df_plot.columns: axes[1, 1].plot(df_plot["datetime"], df_plot["chance_of_rain"], marker='d', color='purple', linewidth=2, markersize=4) axes[1, 1].set_title("Chance of Rain (%)") axes[1, 1].set_ylabel("Chance of Rain (%)") axes[1, 1].grid(True, alpha=0.3) axes[1, 1].tick_params(axis='x', rotation=45) for ax in axes.flat: if len(df_plot) > 1: # Adjust the timeline format according to the data range time_span = (df_plot["datetime"].iloc[-1] - df_plot["datetime"].iloc[0]).total_seconds() / 3600 if time_span <= 24: # In 24 hours, show hours ax.xaxis.set_major_formatter(mdates.DateFormatter('%H:%M')) ax.xaxis.set_major_locator(mdates.HourLocator(interval=max(1, int(time_span/6)))) elif time_span <= 168: # Within one week, date and hour are displayed ax.xaxis.set_major_formatter(mdates.DateFormatter('%m/%d %H')) ax.xaxis.set_major_locator(mdates.HourLocator(interval=12)) else: # More than one week, only the date is displayed ax.xaxis.set_major_formatter(mdates.DateFormatter('%m/%d')) ax.xaxis.set_major_locator(mdates.DayLocator(interval=1)) timestamp = int(time.time()) plt.tight_layout() chart_path = "/tmp/weather_chart.png" plt.savefig(chart_path, dpi=300, bbox_inches='tight') print(f" Chart saved to: {chart_path}") plt.close() return chart_path else: # Returns an error if there is no time field return "Error: No time column found in DataFrame" except Exception as e: plt.close() # Ensure closure charts return f"Chart generation error: {str(e)}" def render_table(df: pd.DataFrame) -> str: try: if df.empty: return "No data available" # Create a more readable table format display_df = df.copy() # Rename fields to more friendly names column_mapping = { "time": "Time", "temp_c": "Temp (°C)", "feelslike_c": "Feels Like (°C)", "humidity": "Humidity (%)", "condition": "Condition", "chance_of_rain": "Rain (%)", "chance_of_snow": "Snow (%)", "wind_kph": "Wind (kph)", "uv": "UV Index", "cloud": "Cloud (%)", "vis_km": "Visibility (km)" } # Rename existing columns for old_name, new_name in column_mapping.items(): if old_name in display_df.columns: display_df = display_df.rename(columns={old_name: new_name}) # Retain 1 decimal place in the numeric field numeric_columns = display_df.select_dtypes(include=['float64', 'float32']).columns for col in numeric_columns: display_df[col] = display_df[col].round(1) # 使用 tabulate 創建完美對齊的表格 return tabulate( display_df, headers='keys', tablefmt='grid', showindex=False, numalign="center", stralign="center" ) except Exception as e: return f"Table generation error: {str(e)}" def render_text_summary(df: pd.DataFrame, location: str, time_type: str) -> str: try: lines = [f"Weather {time_type} summary for {location}:"] for _, row in df.iterrows(): if "time" in row and "temp_c" in row and "humidity" in row: time_str = row["time"] temp = row["temp_c"] humidity = row["humidity"] condition = row.get("condition", "N/A") lines.append(f"{time_str}: {temp}°C, {humidity}% humidity, {condition}") return "\n".join(lines) except Exception as e: return f"Text summary generation error: {str(e)}" def predict_weather_fallback(location: str, target_dt: datetime) -> dict: try: # Load previously trained model and metadata all_models = joblib.load("weather_multi_parameter_models.joblib") forecast_horizons = joblib.load("weather_forecast_horizons.joblib") # Get the actual weather of the location as input to the model weather_api_key = os.environ.get("WEATHER_API_KEY") current_url = f"http://api.weatherapi.com/v1/current.json?key={weather_api_key}&q={location}" current_data = requests.get(current_url).json() current = current_data["current"] # Prepare X_input: a DataFrame that is consistent with the "features" of the training X_input = pd.DataFrame([{ "temp": current["temp_c"], "humidity": current["humidity"], "pressure": current["pressure_mb"], "wind": current["wind_kph"] / 3.6, # km/h → m/s "hour": target_dt.hour, "day": target_dt.day, "month": target_dt.month, "day_of_week": target_dt.weekday(), "is_weekend": 1 if target_dt.weekday() >= 5 else 0 }]) # fix the offset-naive vs offset-aware bug # Use location to get the timezone string, then take the now_aware of the current "offset-aware" tz_str_model = location_to_timezone(location) now_aware = datetime.now(ZoneInfo(tz_str_model)) hours_ahead = int((target_dt - now_aware).total_seconds() / 3600) # Find the nearest trained model to hours_ahead horizon closest_horizon = min(forecast_horizons, key=lambda h: abs(h - hours_ahead)) # The corresponding sub-models are used to make their own predictions predictions = {} if closest_horizon in all_models["temperature"]: predictions["temp_c"] = float(all_models["temperature"][closest_horizon].predict(X_input)[0]) else: # If the model doesn't have the horizon you want, fallback to 0 or some other default value predictions["temp_c"] = current["temp_c"] if closest_horizon in all_models["humidity"]: predictions["humidity"] = float(all_models["humidity"][closest_horizon].predict(X_input)[0]) else: predictions["humidity"] = current["humidity"] if closest_horizon in all_models["pressure"]: predictions["pressure"] = float(all_models["pressure"][closest_horizon].predict(X_input)[0]) else: predictions["pressure"] = current["pressure_mb"] if closest_horizon in all_models["wind"]: pred_wind_ms = float(all_models["wind"][closest_horizon].predict(X_input)[0]) predictions["wind_kph"] = pred_wind_ms * 3.6 # m/s → km/h else: predictions["wind_kph"] = current["wind_kph"] # Use predicted temperature and humidity to extrapolate other parameters (rain or shine, cloud cover etc.) temp = predictions["temp_c"] humidity = predictions["humidity"] if humidity > 80: condition = "Cloudy" if temp < 25 else "Partly cloudy" chance_of_rain = 60 if humidity > 90 else 40 elif temp > 30: condition = "Sunny" chance_of_rain = 5 else: condition = "Clear" if humidity < 60 else "Partly cloudy" chance_of_rain = 20 # Assembles the complete dictionary to be returned to the Agent return { "temp_c": round(predictions["temp_c"], 1), "feelslike_c": round(predictions["temp_c"] - 2, 1), "humidity": round(predictions["humidity"]), "pressure": round(predictions["pressure"]), "wind_kph": round(predictions["wind_kph"], 1), "condition": condition, "chance_of_rain": chance_of_rain, "uv": 5, # fix "cloud": round(humidity * 0.8), "vis_km": 10 if humidity < 80 else 5 } except Exception as e: print(f"ML prediction error: {e}") # If the model or other link throws an exception, it returns a simple default prediction return { "temp_c": 25, "feelslike_c": 23, "humidity": 60, "pressure": 1013, "wind_kph": 15, "condition": "Partly cloudy", "chance_of_rain": 20, "uv": 5, "cloud": 40, "vis_km": 10 } def weather_agent_tool(query: str) -> str: try: weather_api_key = os.environ.get("WEATHER_API_KEY") if not weather_api_key: return "Weather API key not found. Please set WEATHER_API_KEY env variable." # Use get_time_tool2 to get (start_dt, duration_hours, location) time_result = get_time_tool2(query) if not isinstance(time_result, tuple) or len(time_result) != 3: return "Error in retrieving time information." start_dt, duration_hours, location = time_result tz_str = location_to_timezone(location) start_dt = start_dt.replace(tzinfo=ZoneInfo(tz_str)) now = datetime.now(ZoneInfo(tz_str)) end_dt = start_dt + timedelta(hours=duration_hours) if start_dt < now - timedelta(days=7): return "Only supports up to 7 days of historical data." if end_dt > now + timedelta(days=13): return "Only supports up to 13 days of future forecast." weather_data = [] current_time = start_dt while current_time <= end_dt: time_diff_hours = (current_time - now).total_seconds() / 3600 if time_diff_hours > 72: # Over the next 3 days, using ML modelling try: model_result = predict_weather_fallback(location, current_time) weather_point = { "time": current_time.strftime('%Y-%m-%d %H:%M'), "condition": model_result["condition"], "temp_c": model_result["temp_c"], "feelslike_c": model_result["feelslike_c"], "humidity": model_result["humidity"], "chance_of_rain": model_result.get("chance_of_rain", 0), "chance_of_snow": model_result.get("chance_of_snow", 0), "wind_kph": model_result.get("wind_kph", 0), "uv": model_result.get("uv", 0), "cloud": model_result.get("cloud", 0), "vis_km": model_result.get("vis_km", 0) } except Exception as e: # ML model failing fallback weather_point = { "time": current_time.strftime('%Y-%m-%d %H:%M'), "condition": "Partly cloudy", "temp_c": 20, "feelslike_c": 18, "humidity": 60, "chance_of_rain": 20, "chance_of_snow": 0, "wind_kph": 15, "uv": 5, "cloud": 40, "vis_km": 10 } else: # Use WeatherAPI within API support. try: if time_diff_hours < -168: current_time += timedelta(hours=1) continue elif time_diff_hours < -24: # Over 1 day ago with historical APIs url = f"http://api.weatherapi.com/v1/history.json?key={weather_api_key}&q={location}&dt={current_time.strftime('%Y-%m-%d')}" else: # Use the Predictive API for everything within 1 day (including now) url = f"http://api.weatherapi.com/v1/forecast.json?key={weather_api_key}&q={location}&days=3&aqi=no&alerts=no" data = requests.get(url).json() # print(f"[DEBUG] Processing time: {current_time}, time difference: {time_diff_hours:.1f}hour") # Collect all available hourly data forecast_hours = [] if "forecast" in data: for day in data["forecast"]["forecastday"]: for hour in day["hour"]: forecast_hours.append(hour) # Find the closest hour min_diff = float("inf") closest_hour = None for hour_data in forecast_hours: hour_dt = date_parser.parse(hour_data["time"]).replace(tzinfo=ZoneInfo(tz_str)) diff = abs((hour_dt - current_time).total_seconds()) if diff < min_diff: min_diff = diff closest_hour = hour_data if closest_hour: #print(f"[DEBUG] best match: Objectives{current_time.strftime('%H:%M')}, Selected{date_parser.parse(closest_hour['time']).strftime('%H:%M')}, Temperature{closest_hour['temp_c']}°C") weather_point = { "time": current_time.strftime('%Y-%m-%d %H:%M'), "condition": closest_hour["condition"]["text"], "temp_c": closest_hour["temp_c"], "feelslike_c": closest_hour["feelslike_c"], "humidity": closest_hour["humidity"], "chance_of_rain": closest_hour.get("chance_of_rain", 0), "chance_of_snow": closest_hour.get("chance_of_snow", 0), "wind_kph": closest_hour.get("wind_kph", 0), "uv": closest_hour.get("uv", 0), "cloud": closest_hour.get("cloud", 0), "vis_km": closest_hour.get("vis_km", 0) } else: # API fallback when no data is available weather_point = { "time": current_time.strftime('%Y-%m-%d %H:%M'), "condition": "No data", "temp_c": 20, "feelslike_c": 18, "humidity": 60, "chance_of_rain": 0, "chance_of_snow": 0, "wind_kph": 0, "uv": 0, "cloud": 0, "vis_km": 0 } except Exception as e: # fallback when the API fails weather_point = { "time": current_time.strftime('%Y-%m-%d %H:%M'), "condition": "API Error", "temp_c": 20, "feelslike_c": 18, "humidity": 60, "chance_of_rain": 0, "chance_of_snow": 0, "wind_kph": 0, "uv": 0, "cloud": 0, "vis_km": 0 } weather_data.append(weather_point) # Hourly Sampling current_time += timedelta(hours=1) # Formatted as a DataFrame df = pd.DataFrame(weather_data) # Step 5：Variables for Summary Prompt if duration_hours == 1: # Single point enquiry time_description = f"at a specific time: {start_dt.strftime('%Y-%m-%d %H:%M')} in {location}" if len(weather_data) > 0: wd = weather_data[0] weather_data_text = f"""Location: {location} Time: {start_dt.strftime('%Y-%m-%d')} at {start_dt.strftime('%H:%M')} Condition: {wd['condition']} Temperature: {wd['temp_c']}°C (Feels like {wd['feelslike_c']}°C) Humidity: {wd['humidity']}% Chance of rain: {wd['chance_of_rain']}% Chance of snow: {wd['chance_of_snow']}% Wind speed: {wd['wind_kph']} kph UV index: {wd['uv']} Cloud cover: {wd['cloud']}% Visibility: {wd['vis_km']} km""" else: weather_data_text = "No weather data available." else: # Range Enquiry - Using Tabular Format time_description = f"from {start_dt.strftime('%Y-%m-%d %H:%M')} to {end_dt.strftime('%Y-%m-%d %H:%M')} in {location}" weather_data_text = f"Location: {location}\n\nWeather Data Table:\n{df.to_string(index=False)}" summary_prompt = f""" You are a helpful weather reasoning assistant with intelligent output selection. The user wants to know about the weather conditions {time_description}. Use the data below to answer their question. This may refer to the past, present, or future — do not assume it is the current weather. Based on the following weather data and the user's question, think step-by-step to extract the most relevant information, and give a natural, friendly, and cautious answer in British English. Avoid being overly confident — never say "Yes, it will..." or "Definitely." Instead, use expressions like: - "It is very likely that..." - "There is a high chance of..." - "Based on the available data, it seems that..." - "There may be..." Also, after answering the question, include a short weather summary and a useful suggestion. **Do not use markdown formatting such as `*`, `**`, or list symbols.** --- Weather Data --- {weather_data_text} --- User Question --- {query} --- Final Answer --- First, provide your weather analysis and recommendations. Then, intelligently decide if the user would benefit from visual aids: **Add "chart: true" if:** - The query involves trends, changes over time, or comparisons - Multiple time periods are mentioned (e.g., "next 3 days", "this week") - The user asks about patterns, variations, or forecasts - Weather data spans several hours/days - Questions like "how will it change", "show me", "what's the trend" **Add "chart: true" if:** - Weather data spans MORE than 1 hour (time series visualization helpful) - ANY time range query (next 2 hours, today, tomorrow, this week, etc.) - Multiple time points are involved (even implicit ranges) - Trend analysis would be useful for the user - DEFAULT: If duration > 1 hour → ALWAYS add chart: true **Add "table: true" if:** - User wants comprehensive details, precise values, or reference data - Multiple weather parameters need exact numbers - Detailed breakdown is specifically requested **Single point queries (1 hour or specific moment):** - "Will it rain at 3pm tomorrow?" → neither (just text answer) - "Temperature right now in London" → neither (single value) **Time range queries (>1 hour):** - "Weather today" → chart: true (shows daily trend) - "Next 2 hours" → chart: true (shows progression) - "This weekend" → chart: true (trend visualization) - "Tomorrow" → chart: true (daily pattern) **Remember: Humans prefer visual information. When in doubt about time range, lean towards providing charts.** Think about what would be most helpful for the user, even if they didn't explicitly ask. """ response = llm_gpt4.invoke(summary_prompt) response_text = response.content.strip() if isinstance(response, AIMessage) else str(response) # Capture output mode (chart / table) def extract_output_mode(text: str) -> list[str]: modes = [] lower_text = text.lower() if "chart: true" in lower_text: modes.append("chart") if "table: true" in lower_text: modes.append("table") return modes output_modes = extract_output_mode(response_text) # Remove the chart/table prompt string and keep only the narrative. clean_response = response_text.split("chart:")[0].split("table:")[0].strip() final_response = clean_response # Inserting a Chart or Table (as indicated by LLM) if "chart" in output_modes: try: chart_path = render_chart(df, location, f"Weather Data for {location}") final_response += f"\n\nHere is your chart:\n{chart_path}" except Exception as e: final_response += f"\n\nChart generation failed: {e}" if "table" in output_modes: try: table_text = render_table(df) final_response += f"\n\nHere is your table:\n{table_text}" except Exception as e: final_response += f"\n\nTable generation failed: {e}" return final_response except Exception as e: return f"Weather Agent Error: {e}" if __name__ == "__main__": test_queries = [ "Will it rain in Tokyo tomorrow at 3pm?", "Show me the weather for the next 2 hours in London.", "What's the temperature in New York from 3pm today to 2am tomorrow?", "Generate a 3-day weather chart for Paris starting next Wednesday.", "Give me a table of average humidity for the next 5 days in Sydney.", "What was the average temperature in Kaohsiung 5 days ago?", "Provide a 6-day weather summary for Tokyo ending today.", "Will it rain in Taipei from 4pm to 8pm today?", "What is the average temperature in London for the next 8 days?", "How hot was it in New York yesterday at 5pm?" ] for q in test_queries: print("─" * 60) print("Query:", q) print("Response:") print(weather_agent_tool(q)) print("\n")