alphas = []
mmcs = []
nzs = []
aucs = []
bestValidationAcc = 0.
for epoch in range(epochs):
# alpha = min(1, max(0, (epoch**0.1-1.5)/0.6))
# alpha = epoch/epochs
alpha = maxAlpha * epoch / epochs
# r2 = min(0.01, 0.01 - epoch * 0.06 / 5000)
# r2 = 1.
for i, batch in enumerate(dl_train):
loss, x, y, y_pred, z = step(model, optimizer, batch, alpha, r2)
accuracy, loss = eval_step(model, x_train, y_train, alpha, r2)
#if epoch % 10 == 0:
# print('train: epoch {}, accuracy {:.3f}, loss {:.3f}'.format(epoch, accuracy, loss))
testacc, testloss = eval_step(model, x_validate, y_validate, alpha, r2)
if testacc > bestValidationAcc:
# include both learnable param and registered buffer
# PATH = outputDir+'/csnn_2_csnn_layers_run{}_r2{:.1f}_maxAlpha{:.1f}_affine_false.pth'.format(run, r2, maxAlpha)
# torch.save({'net':model, 'alpha':alpha, 'r2':r2}, PATH)
bestValidationAcc = testacc
if epoch % 5 == 0:
#print('validation: epoch {}, fc1 shape {}, loss {:.3f}'.format(epoch, model.fc1.weight.data.shape[0], loss))
losses.append(loss)
losses_validate.append(testloss)
accuracies.append(accuracy)
accuracies_validate.append(testacc)
bestValidationAcc = 0.
np.set_printoptions(precision=4)
for epoch in range(epochs):
# alpha = maxAlpha * epoch/epochs
alpha = maxAlpha
for i, batch in enumerate(dl_train):
if learnable_r:
loss_ce, loss_penalty, loss_l2, x, y, y_pred, z = step(
model, optimizer, batch, alpha, r2, learnable_r=True)
else:
loss, x, y, y_pred, z = step(model, optimizer, batch, alpha,
r2)
if learnable_r:
accuracy, loss_ce, loss_penalty, loss_l2 = eval_step(
model, x_train, y_train, alpha, r2, learnable_r=True)
testacc, testloss_ce, testloss_penalty, testloss_l2 = eval_step(
model, x_validate, y_validate, alpha, r2, learnable_r=True)
else:
accuracy, loss = eval_step(model, x_train, y_train, alpha, r2)
testacc, testloss = eval_step(model, x_validate, y_validate, alpha,
r2)
if testacc > bestValidationAcc:
bestValidationAcc = testacc
if epoch % 5 == 0:
if learnable_r:
losses.append(loss_ce + loss_penalty)
losses_validate.append(testloss_ce + testloss_penalty)
else:
losses.append(loss)
nzs = []
aucs = []
bestValidationAcc = 0.
for epoch in range(epochs):
# alpha = min(1, max(0, (epoch**0.1-1.5)/0.6))
# alpha = epoch/epochs
alpha = maxAlpha * epoch / epochs
# alpha = 0.5
# r2 = min(0.01, 0.01 - epoch * 0.06 / 5000)
# r2 = 1.
for i, batch in enumerate(dl_train):
# loss_ce, loss_r, loss_w, x, y, y_pred, z = step(model, optimizer, batch, alpha, r2, learnable_r)
step(model, optimizer, batch, alpha, r2, learnable_r)
if learnable_r:
accuracy, loss_ce, loss_r, loss_w = eval_step(
model, x_train, y_train, alpha, r2, learnable_r)
testacc, testloss_ce, testloss_r, testloss_w = eval_step(
model, x_validate, y_validate, alpha, r2, learnable_r)
else:
accuracy, loss = eval_step(model, x_train, y_train, alpha, r2,
learnable_r)
testacc, testloss = eval_step(model, x_validate, y_validate,
alpha, r2, learnable_r)
#if epoch % 10 == 0:
# print('train: epoch {}, accuracy {:.3f}, loss {:.3f}'.format(epoch, accuracy, loss))
if testacc > bestValidationAcc:
# include both learnable param and registered buffer
# PATH = outputDir+'/csnn_2_csnn_layers_run{}_r2{:.1f}_maxAlpha{:.1f}_affine_false.pth'.format(run, r2, maxAlpha)
# torch.save({'net':model, 'alpha':alpha, 'r2':r2}, PATH)
bestValidationAcc = testacc