Upgrade Python and Pytorch versions for some examples

examples/v1beta1/trial-images/darts-cnn-cifar10/Dockerfile.cpu

@@ -1,5 +1,4 @@
-# TODO (tenzen-y): Upgrade Python version and Pytorch version
-FROM python:3.7-slim
+FROM python:3.9-slim
 
 ENV TARGET_DIR /opt/darts-cnn-cifar10
 

examples/v1beta1/trial-images/darts-cnn-cifar10/architect.py

@@ -47,13 +47,13 @@ def virtual_step(self, train_x, train_y, xi, w_optim):
         gradients = torch.autograd.grad(loss, self.model.getWeights())
 
         # Do virtual step (Update gradient)
-        # Below opeartions do not need gradient tracking
+        # Below operations do not need gradient tracking
         with torch.no_grad():
             # dict key is not the value, but the pointer. So original network weight have to
             # be iterated also.
             for w, vw, g in zip(self.model.getWeights(), self.v_model.getWeights(), gradients):
                 m = w_optim.state[w].get("momentum_buffer", 0.) * self.w_momentum
-                vw.copy_(w - xi * (m + g + self.w_weight_decay * w))
+                vw.copy_(w - torch.FloatTensor(xi) * (m + g + self.w_weight_decay * w))
 
             # Sync alphas
             for a, va in zip(self.model.getAlphas(), self.v_model.getAlphas()):
@@ -85,7 +85,7 @@ def unrolled_backward(self, train_x, train_y, valid_x, valid_y, xi, w_optim):
         # Update final gradient = dalpha - xi * hessian
         with torch.no_grad():
             for alpha, da, h in zip(self.model.getAlphas(), dalpha, hessian):
-                alpha.grad = da - xi * h
+                alpha.grad = da - torch.FloatTensor(xi) * h
 
     def compute_hessian(self, dws, train_x, train_y):
         """

examples/v1beta1/trial-images/darts-cnn-cifar10/requirements.txt

@@ -1,3 +1,3 @@
-torch==1.0.0
-torchvision==0.2.1
-Pillow==6.2.2
+torch==1.11.0
+torchvision==0.12.0
+Pillow>=9.1.1

examples/v1beta1/trial-images/darts-cnn-cifar10/run_trial.py

@@ -146,15 +146,15 @@ def main():
     best_top1 = 0.
 
     for epoch in range(num_epochs):
-        lr_scheduler.step()
-        lr = lr_scheduler.get_lr()[0]
+        lr = lr_scheduler.get_last_lr()
 
         model.print_alphas()
 
         # Training
         print(">>> Training")
         train(train_loader, valid_loader, model, architect, w_optim, alpha_optim,
               lr, epoch, num_epochs, device, w_grad_clip, print_step)
+        lr_scheduler.step()
 
         # Validation
         print("\n>>> Validation")

examples/v1beta1/trial-images/darts-cnn-cifar10/utils.py

@@ -38,7 +38,7 @@ def update(self, val, n=1):
 
 
 def accuracy(output, target, topk=(1,)):
-    """ Computes the precision@k  for the specified values of k """
+    """ Computes the precision@k for the specified values of k """
     maxk = max(topk)
     batch_size = target.size(0)
 
@@ -53,7 +53,7 @@ def accuracy(output, target, topk=(1,)):
 
     res = []
     for k in topk:
-        correct_k = correct[:k].view(-1).float().sum(0)
+        correct_k = correct[:k].contiguous().view(-1).float().sum(0)
         res.append(correct_k.mul_(1.0 / batch_size))
 
     return res

examples/v1beta1/trial-images/pytorch-mnist/Dockerfile

@@ -1,5 +1,4 @@
-# TODO (tenzen-y): Upgrade Python version and Pytorch version
-FROM python:3.7-slim
+FROM python:3.9-slim
 
 ADD examples/v1beta1/trial-images/pytorch-mnist /opt/pytorch-mnist
 WORKDIR /opt/pytorch-mnist

examples/v1beta1/trial-images/pytorch-mnist/requirements.txt

@@ -1,4 +1,4 @@
 cloudml-hypertune==0.1.0.dev6
-torch==1.0.0
-torchvision==0.2.1
-Pillow==6.2.2
+torch==1.11.0
+torchvision==0.12.0
+Pillow>=9.1.1

hack/verify-yamllint.sh

@@ -25,4 +25,4 @@ if [ -z "$(command -v yamllint)" ]; then
 fi
 
 echo 'Running yamllint'
-yamllint -d "{extends: default, rules: {line-length: disable}}" examples/* manifests/*
+yamllint -d "{extends: default, rules: {line-length: disable}}" examples/* manifests/*