Spaces:
Sleeping
Sleeping
Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,201 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import numpy as np
|
3 |
+
import pandas as pd
|
4 |
+
import torch
|
5 |
+
import matplotlib.pyplot as plt
|
6 |
+
from mlp_utils import (
|
7 |
+
MLP, generate_dataset, split_data, train_model, plot_training_history,
|
8 |
+
visualize_weights, plot_weight_optimization, visualize_network,
|
9 |
+
plot_confusion_matrix, plot_classification_metrics, ACTIVATION_MAP
|
10 |
+
)
|
11 |
+
|
12 |
+
st.set_page_config(page_title="Interactive MLP Learning Platform", layout="wide")
|
13 |
+
|
14 |
+
st.title("Interactive MLP Learning Platform")
|
15 |
+
st.markdown("""
|
16 |
+
This application helps you learn about Multi-Layer Perceptrons (MLPs) through interactive experimentation.
|
17 |
+
You can generate synthetic data, design your own MLP architecture, and observe the training process.
|
18 |
+
""")
|
19 |
+
|
20 |
+
# Sidebar for dataset configuration
|
21 |
+
st.sidebar.header("Dataset Configuration")
|
22 |
+
n_samples = st.sidebar.slider("Number of Samples", 100, 1000, 500)
|
23 |
+
n_features = st.sidebar.slider("Number of Features", 2, 10, 4)
|
24 |
+
n_classes = st.sidebar.slider("Number of Classes", 2, 5, 3)
|
25 |
+
|
26 |
+
# Data split percentages
|
27 |
+
st.sidebar.subheader("Data Split (%)")
|
28 |
+
def_percent = 20
|
29 |
+
val_percent = st.sidebar.slider("Validation %", 0, 50, def_percent)
|
30 |
+
test_percent = st.sidebar.slider("Test %", 0, 50, def_percent)
|
31 |
+
train_percent = 100 - val_percent - test_percent
|
32 |
+
if train_percent < 1:
|
33 |
+
st.sidebar.error("Train % must be at least 1%.")
|
34 |
+
|
35 |
+
# Generate dataset
|
36 |
+
if st.sidebar.button("Generate Dataset"):
|
37 |
+
X, y = generate_dataset(n_samples, n_features, n_classes)
|
38 |
+
(X_train, y_train), (X_val, y_val), (X_test, y_test) = split_data(
|
39 |
+
X, y, val_percent/100, test_percent/100)
|
40 |
+
st.session_state['X_train'] = X_train
|
41 |
+
st.session_state['y_train'] = y_train
|
42 |
+
st.session_state['X_val'] = X_val
|
43 |
+
st.session_state['y_val'] = y_val
|
44 |
+
st.session_state['X_test'] = X_test
|
45 |
+
st.session_state['y_test'] = y_test
|
46 |
+
st.session_state['dataset_generated'] = True
|
47 |
+
st.session_state['network_confirmed'] = False
|
48 |
+
st.session_state['training_complete'] = False
|
49 |
+
st.session_state['testing_complete'] = False
|
50 |
+
|
51 |
+
# Main content area
|
52 |
+
if 'dataset_generated' in st.session_state:
|
53 |
+
st.header("Dataset Information")
|
54 |
+
st.write(f"Train: {len(st.session_state['X_train'])} samples | "
|
55 |
+
f"Validation: {len(st.session_state['X_val'])} samples | "
|
56 |
+
f"Test: {len(st.session_state['X_test'])} samples")
|
57 |
+
|
58 |
+
# Display dataset statistics
|
59 |
+
df = pd.DataFrame(st.session_state['X_train'], columns=[f'Feature {i+1}' for i in range(n_features)])
|
60 |
+
df['Class'] = st.session_state['y_train']
|
61 |
+
st.subheader("Training Set Preview")
|
62 |
+
st.dataframe(df.head())
|
63 |
+
|
64 |
+
# MLP Configuration
|
65 |
+
st.header("MLP Configuration")
|
66 |
+
n_hidden_layers = st.slider("Number of Hidden Layers", 1, 5, 2)
|
67 |
+
hidden_sizes = []
|
68 |
+
activations = []
|
69 |
+
activation_options = list(ACTIVATION_MAP.keys())
|
70 |
+
for i in range(n_hidden_layers):
|
71 |
+
cols = st.columns([2, 2])
|
72 |
+
with cols[0]:
|
73 |
+
size = st.slider(f"Nodes in Hidden Layer {i+1}", 2, 20, 8, key=f"hsize_{i}")
|
74 |
+
with cols[1]:
|
75 |
+
act = st.selectbox(f"Activation for Layer {i+1}", activation_options[:-1], index=0, key=f"act_{i}")
|
76 |
+
hidden_sizes.append(size)
|
77 |
+
activations.append(act)
|
78 |
+
# Add activation for input to first hidden
|
79 |
+
activations = [activations[0]] + activations
|
80 |
+
|
81 |
+
# Confirm network button
|
82 |
+
if st.button("Confirm Network"):
|
83 |
+
st.session_state['hidden_sizes'] = hidden_sizes
|
84 |
+
st.session_state['activations'] = activations
|
85 |
+
st.session_state['network_confirmed'] = True
|
86 |
+
st.session_state['training_complete'] = False
|
87 |
+
st.session_state['testing_complete'] = False
|
88 |
+
|
89 |
+
# Show network configuration
|
90 |
+
if st.session_state.get('network_confirmed', False):
|
91 |
+
st.subheader("Network Architecture Visualization")
|
92 |
+
fig = visualize_network(n_features, hidden_sizes, n_classes)
|
93 |
+
st.pyplot(fig)
|
94 |
+
st.write(f"Input: {n_features} | Hidden: {hidden_sizes} | Output: {n_classes}")
|
95 |
+
st.write(f"Activations: {st.session_state['activations']}")
|
96 |
+
|
97 |
+
# Training parameters
|
98 |
+
st.subheader("Training Parameters")
|
99 |
+
epochs = st.slider("Number of Epochs", 10, 200, 50)
|
100 |
+
learning_rate = st.slider("Learning Rate", 0.001, 0.1, 0.01, 0.001)
|
101 |
+
batch_size = st.slider("Batch Size", 8, 128, 32)
|
102 |
+
|
103 |
+
# Train button
|
104 |
+
if st.button("Train MLP"):
|
105 |
+
model = MLP(n_features, hidden_sizes, n_classes, st.session_state['activations'])
|
106 |
+
train_losses, train_accuracies, val_losses, val_accuracies, weights_history = train_model(
|
107 |
+
model,
|
108 |
+
st.session_state['X_train'],
|
109 |
+
st.session_state['y_train'],
|
110 |
+
st.session_state['X_val'],
|
111 |
+
st.session_state['y_val'],
|
112 |
+
epochs,
|
113 |
+
learning_rate,
|
114 |
+
batch_size,
|
115 |
+
track_weights=True
|
116 |
+
)
|
117 |
+
st.session_state['model'] = model
|
118 |
+
st.session_state['train_losses'] = train_losses
|
119 |
+
st.session_state['train_accuracies'] = train_accuracies
|
120 |
+
st.session_state['val_losses'] = val_losses
|
121 |
+
st.session_state['val_accuracies'] = val_accuracies
|
122 |
+
st.session_state['weights_history'] = weights_history
|
123 |
+
st.session_state['training_complete'] = True
|
124 |
+
st.session_state['testing_complete'] = False
|
125 |
+
|
126 |
+
# Show training results
|
127 |
+
if st.session_state.get('training_complete', False):
|
128 |
+
st.header("Training Results")
|
129 |
+
fig = plot_training_history(
|
130 |
+
st.session_state['train_losses'],
|
131 |
+
st.session_state['train_accuracies'],
|
132 |
+
st.session_state['val_losses'],
|
133 |
+
st.session_state['val_accuracies']
|
134 |
+
)
|
135 |
+
st.pyplot(fig)
|
136 |
+
|
137 |
+
st.subheader("Weight Visualization (All Layers)")
|
138 |
+
weight_fig = visualize_weights(st.session_state['model'])
|
139 |
+
st.pyplot(weight_fig)
|
140 |
+
|
141 |
+
st.subheader("Weight Optimization (First Layer)")
|
142 |
+
opt_fig = plot_weight_optimization(st.session_state['weights_history'])
|
143 |
+
st.pyplot(opt_fig)
|
144 |
+
|
145 |
+
col1, col2, col3, col4 = st.columns(4)
|
146 |
+
with col1:
|
147 |
+
st.metric("Final Training Loss", f"{st.session_state['train_losses'][-1]:.4f}")
|
148 |
+
with col2:
|
149 |
+
st.metric("Final Training Accuracy", f"{st.session_state['train_accuracies'][-1]:.2%}")
|
150 |
+
with col3:
|
151 |
+
st.metric("Final Validation Loss", f"{st.session_state['val_losses'][-1]:.4f}")
|
152 |
+
with col4:
|
153 |
+
st.metric("Final Validation Accuracy", f"{st.session_state['val_accuracies'][-1]:.2%}")
|
154 |
+
|
155 |
+
# Test button
|
156 |
+
if st.button("Test on Unseen Data"):
|
157 |
+
model = st.session_state['model']
|
158 |
+
X_test = st.session_state['X_test']
|
159 |
+
y_test = st.session_state['y_test']
|
160 |
+
model.eval()
|
161 |
+
with torch.no_grad():
|
162 |
+
X_tensor = torch.FloatTensor(X_test)
|
163 |
+
outputs = model(X_tensor)
|
164 |
+
_, predicted = torch.max(outputs.data, 1)
|
165 |
+
test_accuracy = (predicted.numpy() == y_test).mean()
|
166 |
+
|
167 |
+
st.session_state['test_accuracy'] = test_accuracy
|
168 |
+
st.session_state['test_predictions'] = predicted.numpy()
|
169 |
+
st.session_state['testing_complete'] = True
|
170 |
+
|
171 |
+
if st.session_state.get('testing_complete', False):
|
172 |
+
st.header("Test Results")
|
173 |
+
st.success(f"Test Accuracy: {st.session_state['test_accuracy']:.2%}")
|
174 |
+
|
175 |
+
# Confusion Matrix
|
176 |
+
st.subheader("Confusion Matrix")
|
177 |
+
cm_fig = plot_confusion_matrix(
|
178 |
+
st.session_state['y_test'],
|
179 |
+
st.session_state['test_predictions'],
|
180 |
+
n_classes
|
181 |
+
)
|
182 |
+
st.pyplot(cm_fig)
|
183 |
+
|
184 |
+
# Classification Metrics
|
185 |
+
st.subheader("Classification Metrics")
|
186 |
+
metrics_df = plot_classification_metrics(
|
187 |
+
st.session_state['y_test'],
|
188 |
+
st.session_state['test_predictions'],
|
189 |
+
n_classes
|
190 |
+
)
|
191 |
+
st.dataframe(metrics_df)
|
192 |
+
|
193 |
+
# Additional Test Metrics
|
194 |
+
st.subheader("Additional Test Metrics")
|
195 |
+
col1, col2 = st.columns(2)
|
196 |
+
with col1:
|
197 |
+
st.metric("Test Accuracy", f"{st.session_state['test_accuracy']:.2%}")
|
198 |
+
with col2:
|
199 |
+
st.metric("Test Error Rate", f"{1 - st.session_state['test_accuracy']:.2%}")
|
200 |
+
else:
|
201 |
+
st.info("Please generate a dataset using the sidebar controls to begin.")
|