MarcosRodrigo/tensorflow-test

1import numpy as np2import matplotlib.pyplot as plt3import tensorflow as tf4from tensorflow.keras.models import Sequential5from tensorflow.keras.layers import Dense6import streamlit as st7 8# Function to generate synthetic data9def generate_data(dataset_type, noise, n_samples=500):10    np.random.seed(0)11    if dataset_type == 'moons':12        from sklearn.datasets import make_moons13        X, y = make_moons(n_samples=n_samples, noise=noise)14    elif dataset_type == 'circles':15        from sklearn.datasets import make_circles16        X, y = make_circles(n_samples=n_samples, noise=noise, factor=0.5)17    elif dataset_type == 'linear':18        X = np.random.randn(n_samples, 2)19        y = (X[:, 0] > X[:, 1]).astype(int)20    else:21        X = np.random.randn(n_samples, 2)22        y = np.random.randint(0, 2, n_samples)23    return X, y24 25# Function to create model26def create_model(input_shape, hidden_layers, activation, learning_rate, regularization_rate):27    model = Sequential()28    model.add(Dense(hidden_layers[0], input_shape=input_shape, activation=activation,29                    kernel_regularizer=tf.keras.regularizers.l2(regularization_rate)))30    for units in hidden_layers[1:]:31        model.add(Dense(units, activation=activation,32                        kernel_regularizer=tf.keras.regularizers.l2(regularization_rate)))33    model.add(Dense(1, activation='sigmoid'))34    model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate),35                  loss='binary_crossentropy',36                  metrics=['accuracy'])37    return model38 39# Streamlit UI40st.title('Interactive Neural Network Visualization')41st.sidebar.header('Model Parameters')42 43# Dataset selection44dataset_type = st.sidebar.selectbox('Select dataset', ['moons', 'circles', 'linear'])45noise = st.sidebar.slider('Noise level', 0.0, 1.0, 0.2)46X, y = generate_data(dataset_type, noise)47split = st.sidebar.slider('Train/Test split ratio', 0.1, 0.9, 0.5)48split_idx = int(split * len(X))49X_train, X_test = X[:split_idx], X[split_idx:]50y_train, y_test = y[:split_idx], y[split_idx:]51 52# Model parameters53learning_rate = st.sidebar.slider('Learning rate', 0.001, 0.1, 0.01)54activation = st.sidebar.selectbox('Activation function', ['relu', 'tanh', 'sigmoid'])55regularization_rate = st.sidebar.slider('Regularization rate', 0.0, 0.1, 0.01)56hidden_layers = [st.sidebar.slider('Layer 1 units', 1, 10, 4),57                 st.sidebar.slider('Layer 2 units', 1, 10, 2)]58 59# Create and train model60model = create_model((2,), hidden_layers, activation, learning_rate, regularization_rate)61history = model.fit(X_train, y_train, epochs=100, verbose=0, validation_split=0.1)62 63# Evaluation64train_loss, train_acc = model.evaluate(X_train, y_train, verbose=0)65test_loss, test_acc = model.evaluate(X_test, y_test, verbose=0)66st.write(f'Training loss: {train_loss:.4f}, Training accuracy: {train_acc:.4f}')67st.write(f'Test loss: {test_loss:.4f}, Test accuracy: {test_acc:.4f}')68 69# Plot data and decision boundary70fig, ax = plt.subplots()71ax.scatter(X_test[:, 0], X_test[:, 1], c=y_test, cmap='viridis', marker='o', edgecolor='k', s=50)72xx, yy = np.meshgrid(np.linspace(X_test[:, 0].min(), X_test[:, 0].max(), 100),73                     np.linspace(X_test[:, 1].min(), X_test[:, 1].max(), 100))74Z = model.predict(np.c_[xx.ravel(), yy.ravel()])75Z = Z.reshape(xx.shape)76ax.contourf(xx, yy, Z, alpha=0.5, cmap='viridis')77ax.set_title('Data and Model Decision Boundary')78st.pyplot(fig)79