| title | Single Layer Perceptron |
|---|---|
| emoji | 🧠 |
| colorFrom | blue |
| colorTo | purple |
| sdk | gradio |
| sdk_version | 5.9.1 |
| app_file | app.py |
| python_version | 3.12 |
| pinned | false |
| license | mit |
A single-layer neural network built without external ML libraries — only NumPy and OpenCV.
Use the Gradio app to capture webcam images for 4 classes, train the perceptron, classify live frames, and visualize learned weights.
Trained weights are converted into color or black-and-white images so you can visually inspect what the network learned.
For local use on Windows with keyboard controls, run main_windows.py:
| Key | Action |
|---|---|
q |
Gather training images (a/s/d/f for classes 0–3) |
z |
Train |
m |
Test / classify |
p |
Export weights |
esc |
Exit |
Requires the keyboard package and a local webcam.
- A neuron holds a number between 0 and 1 (its activation).
- Weights connect input neurons to output neurons; they can be positive or negative.
- Bias is a threshold for each output neuron.
- If weighted sum > bias, the neuron fires.
(1×n) · (n×k) = (1×k)
Where n is the number of input neurons and k is the number of output neurons.
ReLU: max(0, x)
Sigmoid: 1 / (1 + e^-x)