app.py

import pandas as pd
from io import StringIO
import pandas as pd
import numpy as np
import xgboost as xgb
from math import sqrt
from sklearn.metrics import mean_squared_error
from sklearn.model_selection import train_test_split
import plotly.express as px
import logging

from datetime import datetime

import plotly.graph_objects as go
import numpy as np
import matplotlib.pyplot as plt
import plotly.graph_objs as go
from plotly.subplots import make_subplots
from matplotlib import pyplot
import whisper
from openai import AzureOpenAI
from openai import OpenAI
import json
import re
import gradio as gr

# Configure logging
logging.basicConfig(
    filename='demand_forecasting.log',  # You can adjust the log file name here
    filemode='a',
    format='[%(asctime)s] [%(levelname)s] [%(filename)s] [%(lineno)s:%(funcName)s()] %(message)s',
    datefmt='%Y-%b-%d %H:%M:%S'
)
LOGGER = logging.getLogger(__name__)

log_level_env = 'INFO'  # You can adjust the log level here
log_level_dict = {
    'DEBUG': logging.DEBUG,
    'INFO': logging.INFO,
    'WARNING': logging.WARNING,
    'ERROR': logging.ERROR,
    'CRITICAL': logging.CRITICAL
}
if log_level_env in log_level_dict:
    log_level = log_level_dict[log_level_env]
else:
    log_level = log_level_dict['INFO']
LOGGER.setLevel(log_level)

class DemandForecasting:
    def __init__(self):
        self.client = OpenAI()


    def get_column(self,train_csv_path: str):
        # Load the training data from the specified CSV file
        train_df = pd.read_csv(train_csv_path)

        column_names = train_df.columns.tolist()
        return column_names

    def load_data(self, train_csv_path: str) -> pd.DataFrame:
        """
        Load training data from a CSV file.

        Args:
            train_csv_path (str): Path to the training CSV file.

        Returns:
            pd.DataFrame: DataFrame containing the training data.
        """
        try:
            # Load the training data from the specified CSV file
            train_df = pd.read_csv(train_csv_path)


            # Return a tuple containing the training DataFrame
            return train_df

        except Exception as e:
            # Log an error message if an exception occurs during data loading
            LOGGER.error(f"Error loading data: {e}")

            # Return None
            return None


    def find_date_column(self, df_data: pd.DataFrame) -> str:
        """
        Find the column containing date-type values from the DataFrame.

        Args:
        - df_data (pd.DataFrame): Input DataFrame.

        Returns:
        - str: Name of the column containing date-type values.
        """
        for column in df_data.columns:
            # Check if the column can be converted to datetime
            try:
                pd.to_datetime(df_data[column])
                return column
            except ValueError:
                pass

        # Return None if no date column is found
        return None

    def preprocess_data(self, df_data: pd.DataFrame, list_columns: list, target_column: str) -> pd.DataFrame:
        """
        Transform date-related data in the DataFrame.

        Args:
        - df_data (pd.DataFrame): Input DataFrame.
        - list_columns (list): List of column names to retain.
        - target_column (str): Name of the target column.

        Returns:
        - pd.DataFrame: Transformed DataFrame.
        """
        # Make a copy of the input DataFrame to avoid modifying the original data
        df_data = df_data.copy()

        list_columns.append(target_column)

        # Drop columns not in list_columns
        columns_to_drop = [col for col in df_data.columns if col not in list_columns]
        df_data.drop(columns=columns_to_drop, inplace=True)

        # Find the date column
        date_column = self.find_date_column(df_data)

        if date_column is None:
            raise ValueError("No date column found in the provided list of columns.")
        else:
            print("date_column", date_column)
            # Parse date information only if a valid date column is found
            df_data[date_column] = pd.to_datetime(df_data[date_column])  # Convert 'date' column to datetime format
            df_data['day'] = df_data[date_column].dt.day  # Extract day of the month
            df_data['month'] = df_data[date_column].dt.month  # Extract month
            df_data['year'] = df_data[date_column].dt.year  # Extract year

            # Cyclical Encoding for Months
            df_data['month_sin'] = np.sin(2 * np.pi * df_data['month'] / 12)  # Cyclical sine encoding for month
            df_data['month_cos'] = np.cos(2 * np.pi * df_data['month'] / 12)  # Cyclical cosine encoding for month

            # Day of the Week
            df_data['day_of_week'] = df_data[date_column].dt.weekday  # Extract day of the week (0 = Monday, 6 = Sunday)

            # Week of the Year
            df_data['week_of_year'] = df_data[date_column].dt.isocalendar().week.astype(int)  # Extract week of the year as integer

            df_data.drop(columns=[date_column], axis=1, inplace=True)  # Drop the original date column

        return df_data

    def train_model(self, train: pd.DataFrame, target_column, list_columns) -> tuple:
        """
        Train an XGBoost model using the provided training data.

        Args:
        - train (pd.DataFrame): DataFrame containing training data.

        Returns:
        - tuple: A tuple containing the trained model, true validation labels, and predicted validation labels.
        """
        try:

            # Extract features and target variable
            X = train.drop(columns=[target_column])
            y = train[target_column]

            # Cannot use cross validation because it will use future data
            X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.2, random_state=333)

            # Convert data into DMatrix format for XGBoost
            dtrain = xgb.DMatrix(X_train, label=y_train)
            dval = xgb.DMatrix(X_val, label=y_val)

            # Parameters for XGBoost
            param = {
                'max_depth': 9,
                'eta': 0.3,
                'objective': 'reg:squarederror'
            }

            num_round = 60

            # Train the model
            model_xgb = xgb.train(param, dtrain, num_round)

            # Validate the model
            y_val_pred = model_xgb.predict(dval)     # Predict validation set labels

            # Calculate mean squared error
            mse = mean_squared_error(y_val, y_val_pred)

            # Print validation RMSE
            validation = f"Validation RMSE: {np.sqrt(mse)}"

            # Return trained model, true validation labels, and predicted validation labels
            return model_xgb, y_val, y_val_pred, validation

        except Exception as e:
            # Log an error message if an exception occurs during model training
            LOGGER.error(f"Error training model: {e}")

            # Return None for all outputs in case of an error
            return None, None, None

    def plot_line_graph(self, y_val, y_val_pred):

        # Take only the first 1000 data points
        num_data_points = 1000
        y_val = y_val[:num_data_points]
        y_val_pred = y_val_pred[:num_data_points]    

        # Create Plotly figure
        fig = make_subplots(rows=1, cols=1)

        # Add actual vs predicted traces to the figure (line plot)
        fig.add_trace(go.Scatter(x=np.arange(len(y_val)), y=y_val, mode='lines', name='Actual'), row=1, col=1)
        fig.add_trace(go.Scatter(x=np.arange(len(y_val)), y=y_val_pred, mode='lines', name='Predicted'), row=1, col=1)

        # Update layout
        fig.update_layout(title='Actual vs Predicted Over Time', xaxis_title='Time', yaxis_title='Value')

        # Show interactive plot
        fig.show()
        return fig

    def plot_scatter_plot(self, y_val, y_val_pred):

        # Take only the first 1000 data points
        num_data_points = 1000
        y_val = y_val[:num_data_points]
        y_val_pred = y_val_pred[:num_data_points]  

        # Create Plotly figure
        fig = make_subplots(rows=1, cols=1)

        # Add scatter plots for actual vs predicted (scatter plot)
        fig.add_trace(go.Scatter(x=np.arange(len(y_val)), y=y_val, mode='markers', name='Actual', marker=dict(color='blue', size=8)), row=1, col=1)
        fig.add_trace(go.Scatter(x=np.arange(len(y_val)), y=y_val_pred, mode='markers', name='Predicted', marker=dict(color='orange', size=8)), row=1, col=1)

        # Update layout
        fig.update_layout(title='Actual vs Predicted Over Time (Scatter Plot)', xaxis_title='Time', yaxis_title='Value')

        # Show interactive plot
        fig.show()

        return fig    


    def predict_sales_for_date(self, input_data, model: xgb.Booster) -> float:
        """
        Predict the sales for a specific date using the trained model.

        Args:
        - date_input (str): Date for which sales prediction is needed (in 'YYYY-MM-DD' format).
        - model (xgb.Booster): Trained XGBoost model.
        - features (pd.DataFrame): DataFrame containing features for the date.

        Returns:
        - float: Predicted sales value.
        """
        try:
            input_features = pd.DataFrame([input_data])

            # Regular expression pattern for date in the format 'dd-mm-yyyy'
            for key, value in input_data.items():
                if isinstance(value, str) and re.match(r'\d{2}-\d{2}-\d{4}', value):
                  date_column = key

            if date_column:
                # # Assuming date_input is a datetime object
                date_input = pd.to_datetime(input_features[date_column])

                # Extract day of the month
                input_features['day'] = date_input.dt.day

                # Extract month
                input_features['month'] = date_input.dt.month

                # Extract year
                input_features['year'] = date_input.dt.year

                # Cyclical sine encoding for month
                input_features['month_sin'] = np.sin(2 * np.pi * input_features['month'] / 12)

                # Cyclical cosine encoding for month
                input_features['month_cos'] = np.cos(2 * np.pi * input_features['month'] / 12)

                # Extract day of the week (0 = Monday, 6 = Sunday)
                input_features['day_of_week'] = date_input.dt.weekday

                # Extract week of the year as integer
                input_features['week_of_year'] = date_input.dt.isocalendar().week


            input_features.drop(columns=[date_column], inplace=True)

            # Convert input features to DMatrix format
            dinput = xgb.DMatrix(input_features)

            # Make predictions using the trained model
            predicted_sales = model.predict(dinput)[0]

            # Print the predicted sales value
            predicted_result = f"""Date: {input_data[str(date_column)]} Predicted Value: {predicted_sales}"""
            # Return the predicted sales value
            return predicted_result

        except Exception as e:
            # Log an error message if an exception occurs during sales prediction
            LOGGER.error(f"Error predicting sales: {e}")

            # Return None in case of an error
            return None

    def audio_to_text(self, audio_path):
        audio_file= open(audio_path, "rb")
        transcription = self.client.audio.transcriptions.create(
        model="whisper-1", 
        file=audio_file,
        language="en")
        print(transcription.text)
        return transcription.text


    def parse_text(self, text, column_list):

        # Define the prompt or input for the model
        conversation =[{"role": "system", "content": ""},
        {"role": "user", "content":f""" Extract the values for this given column list:{column_list}, from the given text. all values should be integer data type. if date in given text, the date format should be in dd-mm-YYYY.
        text```{text}```
        the text may contains other name key and values, use consine similarity to map with column list.
        the column names should be keys.
        return result should be in JSON format:
        """
        }] 

        # Generate a response from the GPT-3 model
        chat_completion = self.client.chat.completions.create(
            model = "gpt-3.5-turbo",
            messages = conversation,
            max_tokens=500,
            temperature=0,
            n=1,
            stop=None,
            response_format={ "type": "json_object" },
        )

        # Extract the generated text from the API response
        generated_text = chat_completion.choices[0].message.content
        print(generated_text)
        # # Assuming jsonString is your JSON string
        try:
            json_data = json.loads(generated_text)
        except Exception as e:
            return e
        # print("parse_text",json_data)
        return json_data

    def main(self, train_csv_path: str, audio_path, target_column, column_list) -> None:
        """
        Main function to execute the demand forecasting pipeline.

        Args:
        - train_csv_path (str): Path to the training CSV file.
        - date (str): Date for which sales prediction is needed (in 'YYYY-MM-DD' format).
        """
        try:


            # Split the string by comma and convert it into a list
            column_list = column_list.split(",")


            text = self.audio_to_text(audio_path)

            input_data = self.parse_text(text, column_list)

            #load data
            train_data = self.load_data(train_csv_path)

            #preprocess the train data
            train_df = self.preprocess_data(train_data, column_list, target_column)

            # Train model and get validation predictions
            trained_model, y_val, y_val_pred, validation = self.train_model(train_df, target_column, column_list)

            # Plot interactive evaluation for training
            line_graph = self.plot_line_graph(y_val, y_val_pred) 

            scatter_plot = self.plot_scatter_plot(y_val, y_val_pred) 

            # Predict sales for the specified date using the trained model
            predicted_value = self.predict_sales_for_date(input_data, trained_model)

            return validation, line_graph, scatter_plot, predicted_value

        except Exception as e:
            # Log an error message if an exception occurs in the main function
            LOGGER.error(f"Error in main function: {e}")

    def gradio_interface(self):
        with gr.Blocks(css="style.css", theme="freddyaboulton/test-blue") as demo:

            gr.HTML("""<center><h1 style="color:#fff">Demand Forecasting</h1></center>""")

            with gr.Row():
                with gr.Column(scale=0.50):
                    train_csv = gr.File(elem_classes="uploadbutton")
                with gr.Column(scale=0.50):
                    column_list = gr.Textbox(label="Column List")
    
            with gr.Row():                
                with gr.Column(scale=0.50):
                    audio_path = gr.Audio(sources=["microphone"], type="filepath")
            with gr.Row():
                with gr.Column(scale=0.50):
                    selected_column = gr.Textbox(label="Select column")
                with gr.Column(scale=0.50):
                    target_column = gr.Textbox(label="target column")    


            with gr.Row():
                validation = gr.Textbox(label="Validation")
            with gr.Row():
                predicted_result = gr.Textbox(label="Predicted Result")
            with gr.Row():                
                line_plot = gr.Plot()
            with gr.Row():                
                scatter_plot = gr.Plot()
                
            train_csv.upload(self.get_column, train_csv, column_list)
            audio_path.stop_recording(self.main, [train_csv, audio_path, target_column, selected_column], [validation, line_plot, scatter_plot, predicted_result])

        demo.launch(debug=True)

if __name__ == "__main__":
    
    demand = DemandForecasting()
    demand.gradio_interface()