print("\niteration from main:%d\n"%(i))
for dp in dropout:
for lr in learning_rate:
for rg in reg:
# weight_scale=weight_scale,reg=rg, dtype=np.float64,normalization="batchnorm",dropout = dp)
model = FullyConnectedNet([hidden_size,hidden_size],input_dim=1*5,
weight_scale=weight_scale,reg=rg, dtype=np.float64,normalization="batchnorm",dropout = dp)
solver = Solver(model,data,
print_every=10, num_epochs=num_epochs, batch_size=batch_size,
update_rule="rmsprop",verbose=True,
optim_config={
'learning_rate': lr,
}
)
solver.train()
result = solver.check_accuracy_for_saccarde( X, y2, num_samples=None, batch_size=546)
if result > acc:
acc = result
marker = "cs231n/data_record/3layer_itteration_%d_hidden_%d_lr_%f_end_rg_%f_end_dp_%f_acc_%f"%(i,hidden_size,lr,rg,dp,acc)
solver.checkpoint_name = marker
solver._save_checkpoint()
print("\niteration = %d\n"%(i))
print("\nlearning rate = %f\n"%(lr))
print("\nreg = %f\n"%(rg))
print("\n dropout = %f\n"%(dp))
print("\ntest accuracy = %f\n"%(acc))
else:
