Interactive neural network visualization

This app shows a compact Multi-Layer Perceptron (MLP) trained on MNIST. Draw a digit and watch activations propagate in real time through all fully connected layers.

How it works:

• Draw: Click and drag in the 2D grid (top left) to sketch a digit
• Watch: See your sketch flow through the network layers in 3D
• Predict: Check the probability for each digit (0–9) in the chart (top right)

Network architecture (default export):

• Input layer: 28×28 pixel grid (your drawing)
• Dense layer 1: 784 → 64 neurons with ReLU
• Dense layer 2: 64 → 32 neurons with ReLU
• Output layer: 32 → 10 logits → softmax probabilities

3D controls:

• Rotate: Hold left mouse button and drag
• Pan: Hold right mouse button and drag
• Zoom: Use mouse wheel

Color coding:

• Nodes: Color shows activation strength (dark blues for low/negative values, strong coral for high positive activations)
• Connections: Warm colors for strong positive contributions, cool tones for negative, faint lines near zero.

Training your own model:

• Run python training/mlp_train.py to train the MLP (with Apple Metal acceleration if available).
• The script writes exports/mlp_weights.json, which the visualizer loads on startup.
• Change hidden neuron count, epochs, or export paths via the CLI options documented in training/mlp_train.py.

Real-time features:

• Layer activations: Spheres show per-neuron activations with color-coded strength.
• Strong connections: Each target neuron highlights its strongest incoming weights for readability.
• Live probabilities: The bar chart updates logits → softmax values in real time.

The network is kept compact for smooth real-time visualization. You can retrain with other layer sizes; keep the architecture lean so the 3D view stays responsive.