app.py

import os
import re
import mne
import numpy as np
import pandas as pd
import tempfile
import zipfile
from pathlib import Path
import gradio as gr
import plotly.graph_objects as go

mne.set_log_level('WARNING')

# Global state (okay for local use)
raw_data = None
events_df_global = None
events_wide_global = None

# =============================================================================
# UTILITIES
# =============================================================================

def discover_stim_channels(raw):
    stim_candidates = []
    for ch_name, ch_type in zip(raw.ch_names, raw.get_channel_types()):
        if ch_type in ['stim', 'misc']:
            stim_candidates.append(ch_name)
        elif any(kw in ch_name.upper() for kw in ['STI', 'TRIG', 'TTL', 'STIM', 'TRIGGER']):
            stim_candidates.append(ch_name)
        elif re.match(r'^\d+[a-zA-Z]+$', ch_name):
            stim_candidates.append(ch_name)
    return sorted(set(stim_candidates))


def extract_events_from_channels(raw, stim_channels):
    events = []
    sfreq = raw.info['sfreq']
    for ch in stim_channels:
        if ch not in raw.ch_names:
            continue
        data = raw.get_data(picks=[ch])[0]
        digital = (data > 0.5).astype(int)
        diff = np.diff(digital, prepend=0)
        onset_samples = np.where(diff == 1)[0]
        onset_times = onset_samples / sfreq
        for t, s in zip(onset_times, onset_samples):
            events.append({'channel': ch, 'time_sec': t, 'sample': int(s)})
    if not events:
        return pd.DataFrame()
    df = pd.DataFrame(events)
    return df.sort_values('time_sec').reset_index(drop=True)


def events_df_to_wide_format(events_df):
    if events_df.empty:
        return pd.DataFrame()
    
    cleaned_events = []
    for ch in sorted(events_df['channel'].unique()):
        ch_events = events_df[events_df['channel'] == ch].copy()
        ch_events = ch_events.sort_values('time_sec').reset_index(drop=True)
        times = ch_events['time_sec'].values
        
        i = 0
        while i < len(times):
            burst_start = times[i]
            cleaned_events.append({'channel': ch, 'time_sec': burst_start})
            j = i
            while j + 1 < len(times) and (times[j + 1] - times[j]) <= 0.9:
                j += 1
            i = j + 1
    
    cleaned_df = pd.DataFrame(cleaned_events)
    
    wide_data = {}
    max_events = 0
    for ch in sorted(cleaned_df['channel'].unique()):
        times = cleaned_df[cleaned_df['channel'] == ch]['time_sec'].values
        wide_data[ch] = times
        max_events = max(max_events, len(times))
    
    for ch in wide_data:
        times = wide_data[ch]
        if len(times) < max_events:
            wide_data[ch] = np.pad(times, (0, max_events - len(times)), constant_values=np.nan)
    
    df_wide = pd.DataFrame(wide_data)
    df_wide.index = np.arange(1, len(df_wide) + 1)
    df_wide.index.name = 'Event_Index'
    return df_wide


def plot_events_timeline_plotly(events_df, raw_duration):
    if events_df.empty:
        return None
    fig = go.Figure()
    for i, ch in enumerate(sorted(events_df['channel'].unique())):
        ch_events = events_df[events_df['channel'] == ch]
        fig.add_trace(go.Scatter(
            x=ch_events['time_sec'],
            y=[i] * len(ch_events),
            mode='markers',
            name=f"{ch} (n={len(ch_events)})",
            marker=dict(size=8, opacity=0.7)
        ))
    fig.update_layout(
        xaxis_title='Time (seconds)',
        yaxis_title='Channel',
        xaxis=dict(range=[0, raw_duration]),
        height=500,
        showlegend=True,
        hovermode='closest'
    )
    return fig


def plot_raw_data_plotly(raw, stim_channels, duration=120.0):
    stim_picks = [raw.ch_names.index(ch) for ch in stim_channels if ch in raw.ch_names]
    all_picks = np.arange(len(raw.ch_names))
    non_stim_picks = np.setdiff1d(all_picks, stim_picks)
    
    if len(non_stim_picks) == 0:
        return None
        
    n_channels = min(45, len(non_stim_picks))
    picks = non_stim_picks[:n_channels]
    
    # 🔧 SAFETY FIX: avoid IndexError if raw.times is empty
    recording_duration = raw.times[-1] if len(raw.times) > 0 else 0.0
    end_time = min(duration, recording_duration)
    end_sample = int(end_time * raw.info['sfreq'])
    
    data, times = raw[picks, :end_sample]
    
    fig = go.Figure()
    for i, ch_idx in enumerate(picks):
        ch_data = data[i]
        ch_data = ch_data - np.mean(ch_data)
        if np.std(ch_data) > 1e-15:
            ch_data = ch_data / np.std(ch_data)
        ch_data = ch_data + i * 3
        fig.add_trace(go.Scatter(
            x=times,
            y=ch_data,
            mode='lines',
            name=raw.ch_names[ch_idx],
            line=dict(width=0.8),
            showlegend=False
        ))
    
    fig.update_layout(
        xaxis_title='Time (s)',
        yaxis_title='Channels (normalized + offset)',
        height=min(600, 100 + 120 * n_channels),
        hovermode='x unified'
    )
    return fig


def validate_and_clip_crop_regions(crop_regions, max_time):
    validated = []
    warnings = []
    for start, end in crop_regions:
        if start > max_time:
            warnings.append(f"⚠️ Skipping region ({start:.2f}, {end:.2f}) — starts after recording ends.")
            continue
        clipped_end = min(end, max_time)
        if clipped_end <= start:
            warnings.append(f"⚠️ Skipping region ({start:.2f}, {end:.2f}) — no valid duration.")
            continue
        if end > max_time:
            warnings.append(f"⚠️ Region ({start:.2f}, {end:.2f}) clipped to ({start:.2f}, {clipped_end:.2f})")
        validated.append((start, clipped_end))
    return validated, warnings


# =============================================================================
# GRADIO FUNCTIONS
# =============================================================================

def inspect_eeg(file_obj):
    global raw_data, events_df_global, events_wide_global
    
    if file_obj is None:
        return None, None, "Please upload an EEG file", None
    
    temp_extract_dir = None
    try:
        file_path = file_obj.name
        stem = Path(file_path).stem
        suffixes = Path(file_path).suffixes
        
        load_path = file_path
        
        if len(suffixes) >= 2 and suffixes[-1] == '.zip':
            temp_extract_dir = tempfile.mkdtemp()
            with zipfile.ZipFile(file_path, 'r') as zip_ref:
                zip_ref.extractall(temp_extract_dir)
            
            mff_dir = None
            for item in Path(temp_extract_dir).iterdir():
                if item.is_dir() and item.suffix == '.mff':
                    mff_dir = item
                    break
            
            if mff_dir is None:
                return None, None, "ZIP must contain a .mff folder (e.g., 'data.mff/')", None
            load_path = str(mff_dir)
            raw = mne.io.read_raw_egi(load_path, preload=True, verbose=False)
            
        elif Path(file_path).suffix.lower() == '.fif':
            raw = mne.io.read_raw_fif(file_path, preload=True)
            
        else:
            return None, None, "Please upload a .fif file or a .zip containing a .mff folder", None
        
        raw_data = raw
        
        stim_channels = discover_stim_channels(raw)
        events_df = extract_events_from_channels(raw, stim_channels)
        events_df_global = events_df

        events_wide = events_df_to_wide_format(events_df) if not events_df.empty else pd.DataFrame()
        if not events_wide.empty:
            events_wide = events_wide.reset_index()
        events_wide_global = events_wide
        
        n_stim = len(stim_channels)
        n_data = len(raw.ch_names) - n_stim
        summary_text = f"✅ Loaded: {stem}\n"
        summary_text += f"Duration: {raw.times[-1]:.2f}s | SFreq: {raw.info['sfreq']:.1f} Hz\n"
        summary_text += f"Channels: {n_data} data + {n_stim} stim = {len(raw.ch_names)} total\n\n"
        summary_text += f"🔍 Stimulus channels found: {stim_channels}\n\n"

        raw_plot = plot_raw_data_plotly(raw, stim_channels, duration=120.0)
        timeline_plot = plot_events_timeline_plotly(events_df, raw.times[-1])
        
        return timeline_plot, summary_text, events_wide, raw_plot
        
    except Exception as e:
        error_msg = f"Error loading file: {str(e)}"
        return None, None, error_msg, None
    finally:
        pass


def crop_eeg(crop_regions_str):
    global raw_data, events_wide_global
    
    if raw_data is None:
        return [], "Please inspect an EEG file first"
    
    try:
        crop_regions = []
        for region in crop_regions_str.split(';'):
            region = region.strip()
            if not region:
                continue
            parts = [float(x.strip()) for x in region.split(',')]
            if len(parts) == 2:
                crop_regions.append((parts[0], parts[1]))
        
        if not crop_regions:
            return [], "No valid crop regions specified"
        
        max_time = raw_data.times[-1]
        crop_regions, warnings = validate_and_clip_crop_regions(crop_regions, max_time)
        
        if not crop_regions:
            return [], "No valid crop regions after validation"
        
        if len(crop_regions) == 1:
            start, end = crop_regions[0]
            raw_crop = raw_data.copy().crop(tmin=start, tmax=end)
            status = f"✂️ Applied single crop: {start:.2f}s → {end:.2f}s\n"
        else:
            segments = [raw_data.copy().crop(tmin=s, tmax=e) for s, e in crop_regions]
            raw_crop = mne.concatenate_raws(segments)
            status = f"✂️ Concatenated {len(crop_regions)} segments\n"
        
        status += f"Final cropped duration: {raw_crop.times[-1]:.2f}s\n"
        if warnings:
            status += "\nWarnings:\n" + "\n".join(warnings)
        
        temp_dir = tempfile.mkdtemp()
        output_files = []
        
        eeg_output = os.path.join(temp_dir, "cropped_eeg.fif")
        raw_crop.save(eeg_output, overwrite=True)
        output_files.append(eeg_output)
        
        if events_wide_global is not None and not events_wide_global.empty:
            csv_output = os.path.join(temp_dir, "events_wide.csv")
            events_wide_global.to_csv(csv_output)
            output_files.append(csv_output)
        
        return output_files, status
        
    except Exception as e:
        return [], f"Error during cropping: {str(e)}"


# =============================================================================
# GRADIO APP
# =============================================================================

with gr.Blocks(theme=gr.themes.Base(), title="EEG Stimulus Discovery & Interactive Cropping") as demo:
    gr.Markdown("# EEG Stimulus Discovery & Interactive Cropping")
    gr.Markdown("""
    Upload EEG data to discover stimulus channels, visualize events, and crop clean segments.
    - **`.fif`**: Upload directly  
    - **`.mff`**: ZIP the entire `.mff` folder first, then upload the `.zip` file
    """)
    
    with gr.Tabs():
        with gr.Tab("🔍 Inspect"):
            gr.Markdown("### Step 1: Load and Inspect EEG Data")
            with gr.Row():
                with gr.Column(scale=1):
                    eeg_input = gr.File(
                        label="Upload .fif directly, or .zip containing a .mff folder",
                        file_types=[".fif", ".zip"]
                    )
                    inspect_btn = gr.Button("🔎 Inspect EEG", variant="primary", size="lg")
                with gr.Column(scale=2):
                    summary = gr.Textbox(label="Summary", lines=5, interactive=False)
            
            gr.Markdown("### Stimulus Timeline")
            timeline_plot = gr.Plot(label="Stimulus Events")
            
            gr.Markdown("### Event Times")
            events_csv = gr.Dataframe(label="CSV", interactive=False)

            gr.Markdown("### Raw Data Preview (first 120s)")
            raw_plot = gr.Plot(label="Raw Data")
            
            inspect_btn.click(
                fn=inspect_eeg,
                inputs=eeg_input,
                outputs=[timeline_plot, summary, events_csv, raw_plot]
            )
        
        with gr.Tab("✂️ Crop & Download"):
            gr.Markdown("### Step 2: Define Crop Regions and Download")
            gr.Markdown("""
            **Instructions:**
            - Use the event timeline and raw preview to choose clean segments
            - Format: `start1,end1; start2,end2` (e.g., `25.0,29.0; 30.0,34.0`)
            - Times in seconds
            """)
            
            crop_regions = gr.Textbox(
                value="25.0, 29.0",
                label="Crop Regions (seconds)",
                lines=2
            )
            
            crop_btn = gr.Button("✂️ Crop & Generate Downloads", variant="primary", size="lg")
            crop_status = gr.Textbox(label="Crop Status", lines=5, interactive=False)
            output_files = gr.Files(label="📥 Download Cropped EEG + Events CSV")
            
            crop_btn.click(
                fn=crop_eeg,
                inputs=crop_regions,
                outputs=[output_files, crop_status]
            )

if __name__ == "__main__":
    demo.launch()