micrograd: C++ vs Python

Decision boundary

class +1 class −1 prediction −1 · 0 · +1

Click the plot to add a point of the selected class; Shift-click adds the other class.

Not trained yet: press Train and the background will show what the network predicts.

Preset Points Noise Data seed

Click adds class +1 −1

CSV import / export

Paste CSV: one x,y,label per line, x and y in [−2, 2], label 0/1 or −1/1

Example: project/data/moons_sample.csv in the repo (60 points).

Hidden layers Activation Learning rate Epochs Weights seed

Neurons per hidden layer, comma separated (e.g. 8,8). The output layer (1 neuron) is added for you.

Per data point. Re-initialize picks new random starting weights and trains again; relu sometimes starts “dead” (a flat output) and never learns.

log scale

The loss curve appears after you press Train.

Runs the same training (same data, starting weights, epochs and learning rate) in both implementations, a few times each, one after the other.

No benchmark yet. Press Benchmark to time C++ and Python on the current data and settings.

Times are measured on the machine running the server, not on your device.