'first working version of the demo'

app.py

@@ -3,17 +3,35 @@ import streamlit as st
 import mne
 import matplotlib.pyplot as plt
 import os
+import streamlit as st
+import random
+
 from misc import *
 
+import streamlit as st
+
+# Create two columns with st.columns (new way)
+col1, col2 = st.columns(2)
 
+# Create the upload button in the first column
 # Load the edf file
-edf_file = st.file_uploader("Upload an EEG edf file", type="edf")
+edf_file = col1.file_uploader("Upload an EEG edf file", type="edf")
+# Create the result placeholder button in the second column
+col2.button('Result:')
 
 
 if edf_file is not None:
-
+    
     # Read the file
     raw = read_file(edf_file)
 
     # Preprocess and plot the data
     preprocessing_and_plotting(raw)
+
+    # Build the model
+    clf = build_model(model_name='deep4net', n_classes=2, n_chans=21, input_window_samples=6000)
+
+    output = predict(raw,clf)
+
+    # # Print the output
+    set_button_state (output,col2)

misc.py

@@ -3,7 +3,99 @@ import mne
 import streamlit as st
 import matplotlib.pyplot as plt
 
+from braindecode import EEGClassifier
+from braindecode.models import Deep4Net,ShallowFBCSPNet,EEGNetv4, TCN
+from braindecode.training.losses import CroppedLoss
 
+import torch
+import numpy as np
+
+def set_button_state(output,col):
+        # Generate a random output value of 0 or 1
+    # output = 2023 #random.randint(0, 1)
+
+    # Store the output value in session state
+    st.session_state.output = output
+
+    # Define the button color and text based on the output value
+    if st.session_state.output == 0:
+        button_color = "green"
+        button_text = "Normal"
+    elif st.session_state.output == 1:
+        button_color = "red"
+        button_text = "Abnormal"
+    # elif st.session_state.output == 3:
+    #     button_color = "yellow"
+    #     button_text = "Waiting"
+    else:
+        button_color = "gray"
+        button_text = "Unknown"
+
+    # Create a custom HTML button with CSS styling
+    col.markdown(f"""
+    <style>
+    .custom-button {{
+        background-color: {button_color};
+        color: black;
+        padding: 10px 20px;
+        border: none;
+        border-radius: 5px;
+        cursor: pointer;
+    }}
+    </style>
+    <button class="custom-button">Output: {button_text}</button>
+    """, unsafe_allow_html=True)
+
+
+def predict(raw,clf):
+    x = np.expand_dims(raw.get_data()[:21, :6000], axis=0)
+    output = clf.predict(x)
+    return output
+
+
+def build_model(model_name, n_classes, n_chans, input_window_samples, drop_prob=0.5, lr=0.01):#, weight_decay, batch_size, n_epochs, wandb_run, checkpoint, optimizer__param_groups, window_train_set, window_val):
+    n_start_chans = 25
+    final_conv_length = 1
+    n_chan_factor = 2
+    stride_before_pool = True
+    # input_window_samples =6000
+    model = Deep4Net(
+                n_chans, n_classes,
+                n_filters_time=n_start_chans,
+                n_filters_spat=n_start_chans,
+                input_window_samples=input_window_samples,
+                n_filters_2=int(n_start_chans * n_chan_factor),
+                n_filters_3=int(n_start_chans * (n_chan_factor ** 2.0)),
+                n_filters_4=int(n_start_chans * (n_chan_factor ** 3.0)),
+                final_conv_length=final_conv_length,
+                stride_before_pool=stride_before_pool,
+                drop_prob=drop_prob)
+     
+    clf = EEGClassifier(
+            model,
+            cropped=True,
+            criterion=CroppedLoss,
+            # criterion=CroppedLoss_sd,
+            criterion__loss_function=torch.nn.functional.nll_loss,
+            optimizer=torch.optim.AdamW,
+            optimizer__lr=lr,
+            iterator_train__shuffle=False,
+            # iterator_train__sampler = ImbalancedDatasetSampler(window_train_set, labels=window_train_set.get_metadata().target),
+            # batch_size=batch_size,
+            callbacks=[
+                # EarlyStopping(patience=5),
+                # StochasticWeightAveraging(swa_utils, swa_start=1, verbose=1, swa_lr=lr),
+                # "accuracy", "balanced_accuracy","f1",("lr_scheduler", LRScheduler('CosineAnnealingLR', T_max=n_epochs - 1)),
+                # checkpoint,
+                ],  #"accuracy",
+            # device='cuda'
+            )
+    clf.initialize()
+    pt_path = './Deep4Net_trained_tuh_scaling_wN_WAug_DefArgs_index8_number2700_state_dict_100.pt'
+    clf.load_params(f_params=pt_path)
+    
+    return clf
+    
 
 def preprocessing_and_plotting(raw):
     # Select the first channel