Merge pull request #2 from SaM-92/dev_b

Resolve Midnight Data Gaps & Integrate ElevenLabs Audio Generation

.gitignore

@@ -45,3 +45,4 @@ __pycache__/
 # for workflow
 
 docker_and_azure.txt
+eirgrid_api.ipynb

main.py

@@ -11,6 +11,7 @@
     ConversationHandler,
     CallbackContext,
 )
+from elevenlabs import generate
 from subs.energy_api import *
 from subs.openai_script import *
 from dotenv import load_dotenv
@@ -20,6 +21,8 @@
 load_dotenv()
 Telegram_energy_api = os.environ.get("Telegram_energy_api")
 CHANNEL_ID_FOR_FEEDBACK = os.environ.get("CHANNEL_ID_FOR_FEEDBACK")
+ELEVEN_API_KEY = os.environ.get("ELEVEN_API_KEY")
+
 # Set up logging
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
@@ -123,16 +126,28 @@ async def energy_api_func(update: Update, context: CallbackContext):
             summary_text, df_with_trend = find_optimized_relative_periods(df_)
             today_date = df_with_trend.index[0].strftime("%d/%m/%Y")
             eu_summary_text = optimize_categorize_periods(df_with_trend)
-            quantile_summary_text, _ = find_optimized_relative_periods(
+            quantile_summary_text, df_with_trend_ = find_optimized_relative_periods(
                 df_with_trend
             )  # Generate this based on your DataFrame
 
             prompt = create_combined_gpt_prompt(
                 today_date, eu_summary_text, quantile_summary_text
             )
+
+            # get generated prompt
             gpt_recom = opt_gpt_summarise(prompt)
+            # slice the energy saving actions part
+            energy_saving_actions = get_energy_actions(gpt_recom)
+            audio_msg = generate_voice(energy_saving_actions)
+            await context.bot.send_voice(
+                update.effective_chat.id,
+                audio_msg,
+                caption="Here's your energy-saving tips 🎙️",
+            )
             await update.message.reply_text(gpt_recom)
-            await send_co2_intensity_plot(update, context, df_with_trend)
+            if len(df_with_trend) > 1:
+                await send_co2_intensity_plot(update, context, df_with_trend)
+            del audio_msg
 
     else:
         await update.message.reply_text(
@@ -172,7 +187,7 @@ async def energy_status(update: Update, context: ContextTypes.DEFAULT_TYPE):
 
     # send the list of options and ask the user to select one
     await update.message.reply_text(
-        "I'm here to help you with energy insights. Which category would you like more information about?",
+        "I'm here to help you with energy insights. Which category would you like more information about?💡🌍🔍",
         reply_markup=reply_markup,
     )
     # Set the conversation state to SELECT_COLUMN

requirements.txt

@@ -5,4 +5,5 @@ matplotlib==3.7.0
 requests==2.31.0
 openai==1.11.1
 python-dotenv==1.0.1
-seaborn==0.13.2
+seaborn==0.13.2
+elevenlabs==0.2.27

subs/openai_script.py

@@ -5,10 +5,12 @@
 import numpy as np
 import pandas as pd
 from dotenv import load_dotenv
+from elevenlabs import generate
 
 # Load environment variables from .env file
 load_dotenv()
 OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY")
+ELEVEN_API_KEY = os.environ.get("ELEVEN_API_KEY")
 
 
 def optimize_categorize_periods(df):
@@ -55,52 +57,60 @@ def optimize_categorize_periods(df):
 
 
 def find_optimized_relative_periods(df):
-    # Normalize CO2 values to a 0-1 scale
-    df["normalized"] = (df["Value"] - df["Value"].min()) / (
-        df["Value"].max() - df["Value"].min()
-    )
-
-    # Define thresholds for relative categorization
-    low_threshold = df["normalized"].quantile(0.33)
-    high_threshold = df["normalized"].quantile(0.66)
-
-    # Categorize each timestamp
-    df["category"] = pd.cut(
-        df["normalized"],
-        bins=[-np.inf, low_threshold, high_threshold, np.inf],
-        labels=["Low", "Medium", "High"],
-    )
-
-    # Find consecutive periods with the same category
-    df["group"] = (df["category"] != df["category"].shift()).cumsum()
 
-    # Prepare summary text
-    # summary_text = "Considering absolute CO2 emission values, determined by data trends, distinct periods are identified as:\n\n"
-    summary_text = ""
-
-    # Initialize a dictionary to store concatenated periods for each category
-    period_summary = {"Low": [], "Medium": [], "High": []}
+    if len(df) > 1:
+        # Normalize CO2 values to a 0-1 scale
+        df["normalized"] = (df["Value"] - df["Value"].min()) / (
+            df["Value"].max() - df["Value"].min()
+        )
 
-    # Define emojis for each category
-    emoji_dict = {"Low": "🟢", "Medium": "🟡", "High": "🔴"}
+        # Define thresholds for relative categorization
+        low_threshold = df["normalized"].quantile(0.33)
+        high_threshold = df["normalized"].quantile(0.66)
 
-    # Group by category and group to concatenate periods
-    for (category, group), data in df.groupby(["category", "group"]):
-        start_time = data.index.min().strftime("%H:%M")
-        end_time = data.index.max().strftime("%H:%M")
-        # For periods that start and end at the same time, just show one time
-        period_str = (
-            f"{start_time} to {end_time}" if start_time != end_time else start_time
+        # Categorize each timestamp
+        df["category"] = pd.cut(
+            df["normalized"],
+            bins=[-np.inf, low_threshold, high_threshold, np.inf],
+            labels=["Low", "Medium", "High"],
         )
-        period_summary[category].append(period_str)
 
-    # Format the summary text for each category
-    for category in ["Low", "Medium", "High"]:
-        if period_summary[category]:
-            periods = ", ".join(period_summary[category])
-            summary_text += f"- {emoji_dict[category]} {category} Emission: {periods}\n"
-        else:
-            summary_text += f"- {emoji_dict[category]} {category} Emission: No specific periods identified.\n"
+        # Find consecutive periods with the same category
+        df["group"] = (df["category"] != df["category"].shift()).cumsum()
+
+        # Prepare summary text
+        # summary_text = "Considering absolute CO2 emission values, determined by data trends, distinct periods are identified as:\n\n"
+        summary_text = ""
+
+        # Initialize a dictionary to store concatenated periods for each category
+        period_summary = {"Low": [], "Medium": [], "High": []}
+
+        # Define emojis for each category
+        emoji_dict = {"Low": "🟢", "Medium": "🟡", "High": "🔴"}
+
+        # Group by category and group to concatenate periods
+        for (category, group), data in df.groupby(["category", "group"]):
+            start_time = data.index.min().strftime("%H:%M")
+            end_time = data.index.max().strftime("%H:%M")
+            # For periods that start and end at the same time, just show one time
+            period_str = (
+                f"{start_time} to {end_time}" if start_time != end_time else start_time
+            )
+            period_summary[category].append(period_str)
+
+        # Format the summary text for each category
+        for category in ["Low", "Medium", "High"]:
+            if period_summary[category]:
+                periods = ", ".join(period_summary[category])
+                summary_text += (
+                    f"- {emoji_dict[category]} {category} Emission: {periods}\n"
+                )
+            else:
+                summary_text += f"- {emoji_dict[category]} {category} Emission: No specific periods identified.\n"
+    else:
+        summary_text = (
+            "Sorry, we do not have enough data to process data trend analysis."
+        )
 
     return summary_text, df
 
@@ -120,9 +130,9 @@ def create_combined_gpt_prompt(date, eu_summary_text, quantile_summary_text):
         "- 🔍 Data Trend Schedule: ONLY report it\n"
         f"- 💡 Energy-Saving Actions: Give an example of energy-saving actions for each category of CO2 emission trend, "
         f"considering the current season ({date}). Examples should cover:\n"
-        "   - Low Emission Periods: [Your Example Here]\n"
-        "   - Medium Emission Periods: [Your Example Here]\n"
-        "   - High Emission Periods: [Your Example Here]\n"
+        "   -🟢 Low Emission Periods: [Your Example Here]\n"
+        "   -🟡 Medium Emission Periods: [Your Example Here]\n"
+        "   -🔴High Emission Periods: [Your Example Here]\n"
     )
 
     prompt_text = (
@@ -131,7 +141,7 @@ def create_combined_gpt_prompt(date, eu_summary_text, quantile_summary_text):
         "advice, utilizing specific data trends.\n\n"
         f"{prompt_data}"
         "💡 In periods of low emissions, feel free to use energy-intensive appliances without much concern for reduction.\n\n"
-        f"👉 Please use the following format for your response: \n\n {structure_example}\n"
+        f"👉 Please use the following format for your response and avoid using * in your response: \n\n {structure_example}\n"
     )
 
     return prompt_text
@@ -166,3 +176,35 @@ def opt_gpt_summarise(prompt):
         return generated_text
     except Exception as e:
         return str(e)
+
+
+def get_energy_actions(text):
+    start_keyword = "- 💡 Energy-Saving Actions:"
+    end_keywords = [
+        "📋",
+        "- 🇪🇺",
+        "- 🔍",
+        "- 💡",
+    ]  # Add possible start of next sections if format varies
+    end_keyword = next(
+        (kw for kw in end_keywords if kw in text[text.find(start_keyword) :]),
+        None,
+    )
+
+    # Find start and end positions
+    start_pos = text.find(start_keyword)
+    end_pos = text.find(end_keyword, start_pos + 1) if end_keyword else len(text)
+
+    # Extract the section
+    energy_saving_actions = text[start_pos:end_pos].strip()
+    return energy_saving_actions
+
+
+def generate_voice(text):
+    return generate(
+        text=text,
+        voice="Callum",
+        model="eleven_multilingual_v1",
+        output_format="mp3_44100_128",
+        api_key=ELEVEN_API_KEY,
+    )