Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -6,28 +6,27 @@ from transformers import AutoTokenizer, AutoModelForCausalLM
|
|
| 6 |
import torch
|
| 7 |
import os
|
| 8 |
|
| 9 |
-
# Load an open-source LLM model with no additional training
|
| 10 |
model_name = "tiiuae/falcon-7b-instruct"
|
| 11 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 12 |
model = AutoModelForCausalLM.from_pretrained(
|
| 13 |
model_name,
|
| 14 |
trust_remote_code=True,
|
| 15 |
torch_dtype=torch.float16,
|
| 16 |
-
device_map="auto"
|
| 17 |
)
|
| 18 |
|
| 19 |
def compute_band_power(psd, freqs, fmin, fmax):
|
| 20 |
-
"""Compute mean band power in the given frequency range."""
|
| 21 |
freq_mask = (freqs >= fmin) & (freqs <= fmax)
|
| 22 |
-
# Take the mean across channels and frequencies
|
| 23 |
band_psd = psd[:, freq_mask].mean()
|
| 24 |
return float(band_psd)
|
| 25 |
|
| 26 |
-
def load_eeg_data(file_path):
|
| 27 |
"""
|
| 28 |
-
Load EEG data from a file.
|
| 29 |
-
If FIF
|
| 30 |
-
If CSV
|
|
|
|
|
|
|
| 31 |
"""
|
| 32 |
_, file_ext = os.path.splitext(file_path)
|
| 33 |
file_ext = file_ext.lower()
|
|
@@ -35,24 +34,33 @@ def load_eeg_data(file_path):
|
|
| 35 |
if file_ext == '.fif':
|
| 36 |
raw = mne.io.read_raw_fif(file_path, preload=True)
|
| 37 |
elif file_ext == '.csv':
|
| 38 |
-
# Assume first column is 'time', and subsequent columns are channels
|
| 39 |
df = pd.read_csv(file_path)
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 56 |
ch_types = ['eeg'] * len(ch_names)
|
| 57 |
info = mne.create_info(ch_names=ch_names, sfreq=sfreq, ch_types=ch_types)
|
| 58 |
|
|
@@ -62,31 +70,24 @@ def load_eeg_data(file_path):
|
|
| 62 |
|
| 63 |
return raw
|
| 64 |
|
| 65 |
-
def process_eeg(file):
|
| 66 |
-
|
| 67 |
-
raw = load_eeg_data(file.name)
|
| 68 |
|
| 69 |
-
# Compute PSD (Power Spectral Density) between 1 and 40 Hz
|
| 70 |
psd, freqs = mne.time_frequency.psd_welch(raw, fmin=1, fmax=40)
|
| 71 |
-
|
| 72 |
-
# Compute simple band powers
|
| 73 |
alpha_power = compute_band_power(psd, freqs, 8, 12)
|
| 74 |
beta_power = compute_band_power(psd, freqs, 13, 30)
|
| 75 |
|
| 76 |
-
# Create a short summary of the extracted features
|
| 77 |
data_summary = (
|
| 78 |
f"Alpha power: {alpha_power:.3f}, Beta power: {beta_power:.3f}. "
|
| 79 |
f"The EEG shows stable alpha rhythms and slightly elevated beta activity."
|
| 80 |
)
|
| 81 |
|
| 82 |
-
# Prepare the prompt for the language model
|
| 83 |
prompt = f"""You are a neuroscientist analyzing EEG features.
|
| 84 |
Data Summary: {data_summary}
|
| 85 |
|
| 86 |
Provide a concise, user-friendly interpretation of these findings in simple terms.
|
| 87 |
"""
|
| 88 |
|
| 89 |
-
# Generate the summary using the LLM
|
| 90 |
inputs = tokenizer.encode(prompt, return_tensors="pt").to(model.device)
|
| 91 |
outputs = model.generate(
|
| 92 |
inputs, max_length=200, do_sample=True, top_k=50, top_p=0.95
|
|
@@ -97,12 +98,18 @@ Provide a concise, user-friendly interpretation of these findings in simple term
|
|
| 97 |
|
| 98 |
iface = gr.Interface(
|
| 99 |
fn=process_eeg,
|
| 100 |
-
inputs=
|
|
|
|
|
|
|
|
|
|
|
|
|
| 101 |
outputs="text",
|
| 102 |
-
title="NeuroNarrative-Lite: EEG Summary",
|
| 103 |
-
description=(
|
| 104 |
-
|
| 105 |
-
|
|
|
|
|
|
|
| 106 |
)
|
| 107 |
|
| 108 |
if __name__ == "__main__":
|
|
|
|
| 6 |
import torch
|
| 7 |
import os
|
| 8 |
|
|
|
|
| 9 |
model_name = "tiiuae/falcon-7b-instruct"
|
| 10 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 11 |
model = AutoModelForCausalLM.from_pretrained(
|
| 12 |
model_name,
|
| 13 |
trust_remote_code=True,
|
| 14 |
torch_dtype=torch.float16,
|
| 15 |
+
device_map="auto"
|
| 16 |
)
|
| 17 |
|
| 18 |
def compute_band_power(psd, freqs, fmin, fmax):
|
|
|
|
| 19 |
freq_mask = (freqs >= fmin) & (freqs <= fmax)
|
|
|
|
| 20 |
band_psd = psd[:, freq_mask].mean()
|
| 21 |
return float(band_psd)
|
| 22 |
|
| 23 |
+
def load_eeg_data(file_path, default_sfreq=256.0, time_col='time'):
|
| 24 |
"""
|
| 25 |
+
Load EEG data from a file with flexible CSV handling.
|
| 26 |
+
- If FIF: Use read_raw_fif.
|
| 27 |
+
- If CSV:
|
| 28 |
+
* If `time_col` is present, use it as time.
|
| 29 |
+
* Otherwise, assume a default sfreq and treat all columns as channels.
|
| 30 |
"""
|
| 31 |
_, file_ext = os.path.splitext(file_path)
|
| 32 |
file_ext = file_ext.lower()
|
|
|
|
| 34 |
if file_ext == '.fif':
|
| 35 |
raw = mne.io.read_raw_fif(file_path, preload=True)
|
| 36 |
elif file_ext == '.csv':
|
|
|
|
| 37 |
df = pd.read_csv(file_path)
|
| 38 |
+
|
| 39 |
+
# Remove non-numeric columns except time_col
|
| 40 |
+
for col in df.columns:
|
| 41 |
+
if col != time_col:
|
| 42 |
+
# Drop non-numeric columns if any
|
| 43 |
+
if not pd.api.types.is_numeric_dtype(df[col]):
|
| 44 |
+
df = df.drop(columns=[col])
|
| 45 |
+
|
| 46 |
+
if time_col in df.columns:
|
| 47 |
+
# Use the provided time column
|
| 48 |
+
time = df[time_col].values
|
| 49 |
+
data_df = df.drop(columns=[time_col])
|
| 50 |
+
|
| 51 |
+
if len(time) < 2:
|
| 52 |
+
raise ValueError("Not enough time points to estimate sampling frequency.")
|
| 53 |
+
sfreq = 1.0 / np.mean(np.diff(time))
|
| 54 |
+
else:
|
| 55 |
+
# No explicit time column, assume uniform sampling at default_sfreq
|
| 56 |
+
sfreq = default_sfreq
|
| 57 |
+
data_df = df
|
| 58 |
+
|
| 59 |
+
# Channels are all remaining columns
|
| 60 |
+
ch_names = list(data_df.columns)
|
| 61 |
+
data = data_df.values.T # shape: (n_channels, n_samples)
|
| 62 |
+
|
| 63 |
+
# Create MNE info
|
| 64 |
ch_types = ['eeg'] * len(ch_names)
|
| 65 |
info = mne.create_info(ch_names=ch_names, sfreq=sfreq, ch_types=ch_types)
|
| 66 |
|
|
|
|
| 70 |
|
| 71 |
return raw
|
| 72 |
|
| 73 |
+
def process_eeg(file, default_sfreq, time_col):
|
| 74 |
+
raw = load_eeg_data(file.name, default_sfreq=float(default_sfreq), time_col=time_col)
|
|
|
|
| 75 |
|
|
|
|
| 76 |
psd, freqs = mne.time_frequency.psd_welch(raw, fmin=1, fmax=40)
|
|
|
|
|
|
|
| 77 |
alpha_power = compute_band_power(psd, freqs, 8, 12)
|
| 78 |
beta_power = compute_band_power(psd, freqs, 13, 30)
|
| 79 |
|
|
|
|
| 80 |
data_summary = (
|
| 81 |
f"Alpha power: {alpha_power:.3f}, Beta power: {beta_power:.3f}. "
|
| 82 |
f"The EEG shows stable alpha rhythms and slightly elevated beta activity."
|
| 83 |
)
|
| 84 |
|
|
|
|
| 85 |
prompt = f"""You are a neuroscientist analyzing EEG features.
|
| 86 |
Data Summary: {data_summary}
|
| 87 |
|
| 88 |
Provide a concise, user-friendly interpretation of these findings in simple terms.
|
| 89 |
"""
|
| 90 |
|
|
|
|
| 91 |
inputs = tokenizer.encode(prompt, return_tensors="pt").to(model.device)
|
| 92 |
outputs = model.generate(
|
| 93 |
inputs, max_length=200, do_sample=True, top_k=50, top_p=0.95
|
|
|
|
| 98 |
|
| 99 |
iface = gr.Interface(
|
| 100 |
fn=process_eeg,
|
| 101 |
+
inputs=[
|
| 102 |
+
gr.File(label="Upload your EEG data (FIF or CSV)"),
|
| 103 |
+
gr.Textbox(label="Default Sampling Frequency if no time column (Hz)", value="256"),
|
| 104 |
+
gr.Textbox(label="Time column name (if exists)", value="time")
|
| 105 |
+
],
|
| 106 |
outputs="text",
|
| 107 |
+
title="NeuroNarrative-Lite: EEG Summary (Flexible CSV Handling)",
|
| 108 |
+
description=(
|
| 109 |
+
"Upload EEG data in FIF or CSV format. "
|
| 110 |
+
"If CSV, either include a 'time' column or specify a default sampling frequency. "
|
| 111 |
+
"Non-numeric columns will be removed (except the chosen time column)."
|
| 112 |
+
)
|
| 113 |
)
|
| 114 |
|
| 115 |
if __name__ == "__main__":
|