ium_z487177/MLflow/full.py

import subprocess
import zipfile
import os
import pandas as pd
import re
from sklearn.model_selection import train_test_split
import torch
import torch.nn as nn
import torch.optim as optim
import mlflow

def download_kaggle_dataset(dataset_id, destination_folder):
    try:
        result = subprocess.run(["kaggle", "datasets", "download", "-d", dataset_id, "-p", destination_folder], check=True, capture_output=True, text=True)
        zip_filename = re.search(r"(\S+\.zip)", result.stdout).group(1)
        print(f"Dataset {dataset_id} successfully downloaded.")
        return os.path.join(destination_folder, zip_filename)
    except subprocess.CalledProcessError as e:
        print(f"Error downloading dataset {dataset_id}: {e}")
        return None

def unzip_file(zip_filepath, destination_folder):
    try:
        with zipfile.ZipFile(zip_filepath, 'r') as zip_ref:
            zip_ref.extractall(destination_folder)
            print(f"Files extracted to {destination_folder}.")
    except Exception as e:
        print(f"Error unzipping file {zip_filepath}: {e}")

def combine_csv_files(train_file, test_file, output_file):
    try:
        train_df = pd.read_csv(train_file)
        test_df = pd.read_csv(test_file)
        combined_df = pd.concat([train_df, test_df], ignore_index=True)
        combined_df.to_csv(output_file, index=False)
        print(f"Combined CSV files saved to {output_file}.")
    except Exception as e:
        print(f"Error combining CSV files: {e}")

def split_data(data, train_ratio, dev_ratio, random_seed=42):
    train_data, temp_data = train_test_split(data, train_size=train_ratio, random_state=random_seed)
    dev_data, test_data = train_test_split(temp_data, train_size=dev_ratio / (1 - train_ratio), random_state=random_seed)
    return train_data, dev_data, test_data

class SimpleNN(nn.Module):
    def __init__(self, input_size, hidden_size, output_size):
        super(SimpleNN, self).__init__()
        self.fc1 = nn.Linear(input_size, hidden_size)
        self.fc2 = nn.Linear(hidden_size, output_size)
        self.softmax = nn.Softmax(dim=1)

    def forward(self, x):
        x = torch.relu(self.fc1(x))
        x = self.fc2(x)
        return self.softmax(x)

def main():
    with mlflow.start_run():
        dataset_id = "iabhishekofficial/mobile-price-classification"
        destination_folder = "/app/data"
        zip_filepath = download_kaggle_dataset(dataset_id, destination_folder)

        if zip_filepath is not None:
            unzip_file(zip_filepath, destination_folder)
            train_file = os.path.join(destination_folder, "train.csv")
            test_file = os.path.join(destination_folder, "test.csv")
            output_file = os.path.join(destination_folder, "combined.csv")
            combine_csv_files(train_file, test_file, output_file)

            data = pd.read_csv(output_file)
            train_data, dev_data, test_data = split_data(data, train_ratio=0.6, dev_ratio=0.2)

            output_dir = "/app/output"
            os.makedirs(output_dir, exist_ok=True)
            train_data.to_csv(os.path.join(output_dir, 'Train1.csv'), index=False)
            dev_data.to_csv(os.path.join(output_dir, 'Dev1.csv'), index=False)
            test_data.to_csv(os.path.join(output_dir, 'Test1.csv'), index=False)

            print(f"Liczba wierszy w pliku Train1.csv: {len(train_data)}")
            print(f"Liczba wierszy w pliku Dev1.csv: {len(dev_data)}")
            print(f"Liczba wierszy w pliku Test1.csv: {len(test_data)}")

        train_file_path = os.path.join(output_dir, 'Train1.csv')
        train_data = pd.read_csv(train_file_path)
        train_data = train_data.dropna(subset=['price_range'])

        valid_values = {0.0, 1.0, 2.0, 3.0}
        assert set(train_data['price_range'].unique()) <= valid_values, "Unexpected values in price_range"

        input_size = len(train_data.columns) - 2  
        hidden_size = 50
        output_size = len(valid_values)

        # Logowanie parametrów
        mlflow.log_param("input_size", input_size)
        mlflow.log_param("hidden_size", hidden_size)
        mlflow.log_param("output_size", output_size)

        model = SimpleNN(input_size, hidden_size, output_size)

        criterion = nn.CrossEntropyLoss()
        optimizer = optim.Adam(model.parameters(), lr=0.001)

        epochs = 15
        for epoch in range(epochs):
            inputs = torch.tensor(train_data.drop(['price_range', 'id'], axis=1).values, dtype=torch.float32)
            labels = torch.tensor(train_data['price_range'].values, dtype=torch.long)

            optimizer.zero_grad()
            outputs = model(inputs)
            loss = criterion(outputs, labels)
            loss.backward()
            optimizer.step()

            # Logowanie metryk
            mlflow.log_metric("loss", loss.item(), step=epoch)

            print(f"Epoch {epoch+1}/{epochs}, Loss: {loss.item()}")

        save_path = "model.pth"
        torch.save(model.state_dict(), save_path)

        model.load_state_dict(torch.load("model.pth"))
        model.eval() 

        test_file_path = os.path.join(output_dir, 'Test1.csv')
        test_data = pd.read_csv(test_file_path)

        inputs = torch.tensor(test_data.drop(['price_range', 'id'], axis=1).values, dtype=torch.float32)
        with torch.no_grad():
            predictions = model(inputs)
        predicted_classes = torch.argmax(predictions, dim=1)

        predicted_classes_df = pd.DataFrame(predicted_classes.numpy(), columns=['Predicted_Price_Range'])
        predicted_classes_df['Actual_Price_Range'] = test_data['price_range'].values

        output_path = 'predictions.csv'
        predicted_classes_df.to_csv(output_path, index=False)

if __name__ == "__main__":
    main()