File size: 8,551 Bytes
d9c2086
4390020
d9c2086
 
 
4390020
 
d9c2086
4390020
d9c2086
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d724e84
d9c2086
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
# === Imports ===
import streamlit as st
import torch
import torch.nn as nn
import torch.optim as optim
import numpy as np
import matplotlib.pyplot as plt
from sklearn.datasets import make_moons, make_circles, make_classification
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import accuracy_score
import io

# === Streamlit Config ===
st.set_page_config(page_title="🧠 ANN Visualizer", layout="wide")
st.markdown("<style>.main { background: linear-gradient(to right, #ece9e6, #ffffff); }</style>", unsafe_allow_html=True)

# === Sidebar ===
st.sidebar.title("⚙️ Model Configuration")

dataset = st.sidebar.selectbox("Select Dataset", ["moons", "circles", "classification"])
n_samples = st.sidebar.slider("Number of Samples", 100, 5000, 500, step=100)
noise = st.sidebar.slider("Noise Level", 0.0, 1.0, 0.2, step=0.05)

st.sidebar.subheader("Network Architecture")
n_layers = st.sidebar.number_input("Number of Hidden Layers", 1, 10, 2)

layer_config = []
activation_options = ["ReLU", "Tanh", "Sigmoid"]
reg_options = ["None", "L1", "L2", "L1_L2"]

for i in range(n_layers):
    with st.sidebar.expander(f"Layer {i+1}"):
        units = st.number_input(f"Units", 1, 512, 8, key=f"units_{i}")
        activation = st.selectbox("Activation", activation_options, key=f"act_{i}")
        dropout = st.slider("Dropout", 0.0, 0.9, 0.0, step=0.05, key=f"drop_{i}")
        reg = st.selectbox("Regularization", reg_options, key=f"reg_{i}")
        reg_strength = st.number_input("Reg Strength", 0.0001, 1.0, 0.001, key=f"reg_strength_{i}") if reg != "None" else 0.0
        layer_config.append((units, activation, dropout, reg, reg_strength))

st.sidebar.subheader("Training Settings")
learning_rate = st.sidebar.number_input("Learning Rate", 0.0001, 1.0, 0.01)
epochs = st.sidebar.slider("Epochs", 100, 5000, 500)
early_stop = st.sidebar.checkbox("Early Stopping", True)
patience = st.sidebar.slider("Patience", 1, 20, 5) if early_stop else None
min_delta = st.sidebar.number_input("Min Improvement (Delta)", 0.0001, 0.1, 0.001) if early_stop else None

# === Dataset Loader ===
def load_data(name, n_samples, noise):
    if name == "moons":
        X, y = make_moons(n_samples=n_samples, noise=noise, random_state=42)
    elif name == "circles":
        X, y = make_circles(n_samples=n_samples, noise=noise, factor=0.5, random_state=42)
    else:
        X, y = make_classification(n_samples=n_samples, n_features=2, n_informative=2, n_redundant=0, n_clusters_per_class=1, random_state=42)
    return X, y

# === Model Definition ===
activation_map = {
    "ReLU": nn.ReLU,
    "Tanh": nn.Tanh,
    "Sigmoid": nn.Sigmoid
}

class CustomLayer(nn.Module):
    def __init__(self, in_dim, out_dim, activation, dropout, reg, reg_strength):
        super().__init__()
        self.linear = nn.Linear(in_dim, out_dim)
        self.activation = activation_map[activation]()
        self.dropout = nn.Dropout(dropout)
        self.reg = reg
        self.reg_strength = reg_strength

    def forward(self, x):
        x = self.linear(x)
        x = self.activation(x)
        x = self.dropout(x)
        return x

    def reg_loss(self):
        if self.reg == "None":
            return 0
        elif self.reg == "L1":
            return self.reg_strength * torch.sum(torch.abs(self.linear.weight))
        elif self.reg == "L2":
            return self.reg_strength * torch.sum(self.linear.weight.pow(2))
        elif self.reg == "L1_L2":
            return self.reg_strength * (torch.sum(torch.abs(self.linear.weight)) + torch.sum(self.linear.weight.pow(2)))

class ANN(nn.Module):
    def __init__(self, input_dim, config, output_dim=2):
        super().__init__()
        self.layers = nn.ModuleList()
        self.reg_layers = []
        prev_dim = input_dim
        for units, act, drop, reg, reg_strength in config:
            layer = CustomLayer(prev_dim, units, act, drop, reg, reg_strength)
            self.layers.append(layer)
            self.reg_layers.append(layer)
            prev_dim = units
        self.output = nn.Linear(prev_dim, output_dim)

    def forward(self, x):
        for layer in self.layers:
            x = layer(x)
        return self.output(x)

    def total_reg_loss(self):
        return sum(layer.reg_loss() for layer in self.reg_layers)

# === Plot Helpers ===
def plot_boundary(model, X, y, title):
    x_min, x_max = X[:,0].min()-0.5, X[:,0].max()+0.5
    y_min, y_max = X[:,1].min()-0.5, X[:,1].max()+0.5
    xx, yy = np.meshgrid(np.linspace(x_min, x_max, 400), np.linspace(y_min, y_max, 400))
    grid = np.c_[xx.ravel(), yy.ravel()]
    model.eval()
    with torch.no_grad():
        probs = torch.softmax(model(torch.tensor(grid, dtype=torch.float32)), dim=1)[:,1]
    Z = probs.reshape(xx.shape).cpu().numpy()

    fig, ax = plt.subplots()
    ax.contourf(xx, yy, Z, levels=50, cmap="Spectral", alpha=0.8)
    ax.scatter(X[:,0], X[:,1], c=y, edgecolors='k', cmap="Spectral", s=20)
    ax.set_title(title)
    ax.set_xticks([])
    ax.set_yticks([])
    return fig

# === Training Function ===
def train(model, X_train, y_train, X_test, y_test, epochs, lr, early_stop, patience, min_delta):
    optimizer = optim.Adam(model.parameters(), lr=lr)
    criterion = nn.CrossEntropyLoss()
    train_losses, test_losses = [], []
    best_loss = np.inf
    stop_counter = 0

    for epoch in range(epochs):
        model.train()
        optimizer.zero_grad()
        y_pred = model(X_train)
        loss = criterion(y_pred, y_train) + model.total_reg_loss()
        loss.backward()
        optimizer.step()
        train_losses.append(loss.item())

        model.eval()
        with torch.no_grad():
            test_pred = model(X_test)
            test_loss = criterion(test_pred, y_test) + model.total_reg_loss()
            test_losses.append(test_loss.item())

        # Early Stopping
        if early_stop:
            if test_losses[-1] < best_loss - min_delta:
                best_loss = test_losses[-1]
                stop_counter = 0
            else:
                stop_counter += 1
                if stop_counter >= patience:
                    st.warning(f"⏹️ Early Stopping at Epoch {epoch+1}")
                    break

        yield epoch+1, train_losses, test_losses

# === Main Page ===
st.title("🧠 Interactive ANN Visualizer")

if st.button("🚀 Start Training"):
    X, y = load_data(dataset, n_samples, noise)
    X = StandardScaler().fit_transform(X)
    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)

    X_train_tensor = torch.tensor(X_train, dtype=torch.float32)
    y_train_tensor = torch.tensor(y_train, dtype=torch.long)
    X_test_tensor = torch.tensor(X_test, dtype=torch.float32)
    y_test_tensor = torch.tensor(y_test, dtype=torch.long)

    model = ANN(X.shape[1], layer_config)

    tabs = st.tabs(["🌐 Decision Boundary", "📈 Training Curve", "📊 Results"])

    with tabs[0]:
        st.subheader("Initial Random Surface")
        fig_init = plot_boundary(model, X, y, "Initial Decision Surface")
        st.pyplot(fig_init)

    train_losses, test_losses = [], []
    progress = st.progress(0)

    for epoch, train_losses, test_losses in train(model, X_train_tensor, y_train_tensor, X_test_tensor, y_test_tensor, epochs, learning_rate, early_stop, patience, min_delta):
        if epoch % (epochs//5) == 0:
            with tabs[0]:
                st.subheader(f"Epoch {epoch} Decision Surface")
                fig = plot_boundary(model, X, y, f"Decision Surface at Epoch {epoch}")
                st.pyplot(fig)
        progress.progress(epoch/epochs)

    with tabs[1]:
        st.subheader("Loss Curves")
        fig_loss, ax = plt.subplots()
        ax.plot(train_losses, label="Train Loss")
        ax.plot(test_losses, label="Test Loss")
        ax.legend()
        st.pyplot(fig_loss)

    with tabs[2]:
        st.subheader("Final Metrics")
        model.eval()
        train_preds = model(X_train_tensor).argmax(dim=1)
        test_preds = model(X_test_tensor).argmax(dim=1)

        st.metric("Train Accuracy", f"{accuracy_score(y_train, train_preds):.2%}")
        st.metric("Test Accuracy", f"{accuracy_score(y_test, test_preds):.2%}")

        fig_final = plot_boundary(model, X, y, "Final Decision Surface")
        st.pyplot(fig_final)

        buf = io.BytesIO()
        fig_final.savefig(buf, format="png")
        st.download_button("⬇️ Download Final Boundary", buf.getvalue(), "final_boundary.png", "image/png")