| { | |
| "model_type": "MLP", | |
| "widths": [ | |
| 3072, | |
| 128, | |
| 10 | |
| ], | |
| "hidden_activation": "relu", | |
| "output_activation": "identity (logits; softmax gives class probabilities)", | |
| "weights": "weights.npz: W0, b0, W1, b1, ...; logits = h W + b with W of shape (n_in, n_out)", | |
| "preprocessing": "flatten the 32x32x3 uint8 image to 3072 values (row-major, RGB last), divide by 255, subtract preprocessing.npz['mean'] and divide by preprocessing.npz['std']", | |
| "classes": [ | |
| "airplane", | |
| "automobile", | |
| "bird", | |
| "cat", | |
| "deer", | |
| "dog", | |
| "frog", | |
| "horse", | |
| "ship", | |
| "truck" | |
| ], | |
| "training": { | |
| "data": "uoft-cs/cifar10 train part (50,000 images), true labels", | |
| "optimizer": "SGD", | |
| "loss": "cross-entropy", | |
| "seed": 1076, | |
| "lr": 0.01, | |
| "batch_size": 64, | |
| "grad_accumulation_steps": 1, | |
| "epochs": 100 | |
| }, | |
| "results": { | |
| "test_acc": 0.4833, | |
| "test_loss": 2.6108671875, | |
| "train_acc": 0.90612, | |
| "train_loss": 0.3009627978515625 | |
| } | |
| } |