MNIST MLP implemented in Rust
To use:
cargo build --release (will be significantly slower if you just cargo run)
./target/release/corroded_classifier
Training should look like:
❯ ./target/release/corroded_classifier
Starting training...
Epoch 1/5
[00:00:06] ======================================== 390/390 Epoch avg loss: 0.0549 | Train Accuracy: 96.26% | Val Accuracy: 95.86%
Epoch 2/5
[00:00:06] ======================================== 390/390 Epoch avg loss: 0.0291 | Train Accuracy: 97.11% | Val Accuracy: 96.44%
Epoch 3/5
[00:00:06] ======================================== 390/390 Epoch avg loss: 0.0221 | Train Accuracy: 97.90% | Val Accuracy: 96.92%
Epoch 4/5
[00:00:06] ======================================== 390/390 Epoch avg loss: 0.0183 | Train Accuracy: 97.99% | Val Accuracy: 96.89%
Epoch 5/5
[00:00:06] ======================================== 390/390 Epoch avg loss: 0.0149 | Train Accuracy: 98.44% | Val Accuracy: 97.24%
and log some ascii graphs at the end.
CIFAR10 can be used, but currently we are treating colour very naively (flattening) to use as MLP input, so performance is poor. Will explore implementing convlayers as a follow up.