app.py

import gradio as gr
import pandas as pd
import numpy as np
from sklearn.preprocessing import StandardScaler
import scipy
from scipy import signal
import pickle
from sklearn.svm import SVC
from sklearn.model_selection import train_test_split, GridSearchCV

# Global variable to store the uploaded data
global_data = None

def get_data_preview(file):
    global global_data
    global_data = pd.read_csv(file.name)
    global_data['label'] = np.nan  # Initialize a label column
    global_data['label'] = global_data['label'].astype(object)  # Ensure the label column can hold different types
    return global_data.head()

def label_data(ranges):
    global global_data
    print("Ranges received for labeling:", ranges)
    for i, (start, end, label) in enumerate(ranges.values):
        start = int(start)
        end = int(end)
        if start < 0 or start >= len(global_data):
            continue
        if end >= len(global_data):
            end = len(global_data) - 1
        global_data.loc[start:end, 'label'] = label
    return global_data.tail()

def preprocess_data():
    global global_data
    try:
        global_data.drop(columns=global_data.columns[0], axis=1, inplace=True)
        global_data.columns = ['raw_eeg', 'label']
        raw_data = global_data['raw_eeg']
        labels_old = global_data['label']

        sampling_rate = 512
        notch_freq = 50.0
        lowcut, highcut = 0.5, 30.0

        nyquist = (0.5 * sampling_rate)
        notch_freq_normalized = notch_freq / nyquist
        b_notch, a_notch = signal.iirnotch(notch_freq_normalized, Q=0.05, fs=sampling_rate)

        lowcut_normalized = lowcut / nyquist
        highcut_normalized = highcut / nyquist
        b_bandpass, a_bandpass = signal.butter(4, [lowcut_normalized, highcut_normalized], btype='band')

        features = []
        labels = []

        def calculate_psd_features(segment, sampling_rate):
            f, psd_values = scipy.signal.welch(segment, fs=sampling_rate, nperseg=len(segment))
            alpha_indices = np.where((f >= 8) & (f <= 13))
            beta_indices = np.where((f >= 14) & (f <= 30))
            theta_indices = np.where((f >= 4) & (f <= 7))
            delta_indices = np.where((f >= 0.5) & (f <= 3))
            energy_alpha = np.sum(psd_values[alpha_indices])
            energy_beta = np.sum(psd_values[beta_indices])
            energy_theta = np.sum(psd_values[theta_indices])
            energy_delta = np.sum(psd_values[delta_indices])
            alpha_beta_ratio = energy_alpha / energy_beta
            return {
                'E_alpha': energy_alpha,
                'E_beta': energy_beta,
                'E_theta': energy_theta,
                'E_delta': energy_delta,
                'alpha_beta_ratio': alpha_beta_ratio
            }

        def calculate_additional_features(segment, sampling_rate):
            f, psd = scipy.signal.welch(segment, fs=sampling_rate, nperseg=len(segment))
            peak_frequency = f[np.argmax(psd)]
            spectral_centroid = np.sum(f * psd) / np.sum(psd)
            log_f = np.log(f[1:])
            log_psd = np.log(psd[1:])
            spectral_slope = np.polyfit(log_f, log_psd, 1)[0]
            return {
                'peak_frequency': peak_frequency,
                'spectral_centroid': spectral_centroid,
                'spectral_slope': spectral_slope
            }

        for i in range(0, len(raw_data) - 512, 256):
            print(f"Processing segment {i} to {i + 512}")
            segment = raw_data.loc[i:i+512]
            segment = pd.to_numeric(segment, errors='coerce')
            segment = signal.filtfilt(b_notch, a_notch, segment)
            segment = signal.filtfilt(b_bandpass, a_bandpass, segment)
            segment_features = calculate_psd_features(segment, 512)
            additional_features = calculate_additional_features(segment, 512)
            segment_features = {**segment_features, **additional_features}
            features.append(segment_features)
            labels.append(labels_old[i])

        columns = ['E_alpha', 'E_beta', 'E_theta', 'E_delta', 'alpha_beta_ratio', 'peak_frequency', 'spectral_centroid', 'spectral_slope']
        df_features = pd.DataFrame(features, columns=columns)
        df_features['label'] = labels

        scaler = StandardScaler()
        X_scaled = scaler.fit_transform(df_features.drop('label', axis=1))
        df_scaled = pd.DataFrame(X_scaled, columns=columns)
        df_scaled['label'] = df_features['label']

        processed_data_filename = 'processed_data.csv'
        df_scaled.to_csv(processed_data_filename, index=False)

        scaler_filename = 'scaler.pkl'
        with open(scaler_filename, 'wb') as file:
            pickle.dump(scaler, file)

        return "Data preprocessing complete! Download the processed data and scaler below.", processed_data_filename, scaler_filename
    
    except Exception as e:
        return f"An error occurred during preprocessing: {e}", None, None

def train_model():
    global global_data
    try:
        preprocess_status, processed_data_filename, scaler_filename = preprocess_data()
        if processed_data_filename is None:
            return preprocess_status, None, None
        
        df_scaled = pd.read_csv(processed_data_filename)
        X = df_scaled.drop('label', axis=1)
        y = df_scaled['label']
        
        X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
        
        param_grid = {'C': [0.1, 1, 10, 100], 'gamma': ['scale', 'auto', 0.1, 0.01, 0.001, 0.0001], 'kernel': ['rbf']}
        svc = SVC(probability=True)
        grid_search = GridSearchCV(estimator=svc, param_grid=param_grid, cv=5, verbose=2, n_jobs=-1)
        grid_search.fit(X_train, y_train)
        
        model = grid_search.best_estimator_
        model_filename = 'model.pkl'

        with open(model_filename, 'wb') as file:
            pickle.dump(model, file)

        return "Training complete! Download the model and scaler below.", model_filename, scaler_filename
    
    except Exception as e:
        print(f"An error occurred during training: {e}")
        return f"An error occurred during training: {e}", None, None

with gr.Blocks() as demo:
    file_input = gr.File(label="Upload CSV File")
    data_preview = gr.Dataframe(label="Data Preview", interactive=False)
    ranges_input = gr.Dataframe(headers=["Start Index", "End Index", "Label"], label="Ranges for Labeling")
    labeled_data_preview = gr.Dataframe(label="Labeled Data Preview", interactive=False)

    training_status = gr.Textbox(label="Training Status")
    model_file = gr.File(label="Download Trained Model")
    scaler_file = gr.File(label="Download Scaler")

    file_input.upload(get_data_preview, inputs=file_input, outputs=data_preview)
    label_button = gr.Button("Label Data")
    label_button.click(label_data, inputs=[ranges_input], outputs=labeled_data_preview, queue=True)
    train_button = gr.Button("Train Model")
    train_button.click(train_model, outputs=[training_status, model_file, scaler_file])

demo.launch()