PJ1 of CV, a three-layer linear neural network for image classification
python 3.12.3
numpy 1.26.4
matplotlib 3.8.4
https://github.com/zalandoresearch/fashion-mnist
Open the file "train_test.py", set the patameter job to "train", and then run
Open the file "train_test.py", set the patameter job to "test", and then run
Net: the network, including lays and activations[ReLU,Sigmoid,Tanh]
lr, lr_scheduler: Set the learning rate and the scheduler of updating[steplr,mulitisteplr,explr]
L2_norm_rate(decay):In the SGD system optimizer without introducing momentum, weight decay is completely equivalent to L2 regularization.
batch_size:Usually 2^k
epochs:Set the maximum epochs, and if there are multiple epochs with no significant decrease, the training will be automatically stopped