simplify torch imports of Tensor (#169)

* Tensor
* venv
* PT version
* setuptools
* torch-sparse
* 95

Author: Borda

Diff for: .actions/assistant.py

@@ -182,7 +182,7 @@ def _find_meta(folder: str) -> str:
 
     @staticmethod
     def _load_meta(folder: str, strict: bool = False) -> Optional[dict]:
-        """Loading meta data for a particular notebook with given folder path.
+        """Loading meta-data for a particular notebook with given folder path.
 
         Args:
             folder: path to the folder with python script, meta and artefacts
@@ -259,7 +259,7 @@ def _parse_requirements(folder: str) -> Tuple[str, str]:
             for k, v in meta.items()
             if k.startswith(AssistantCLI._META_PIP_KEY)
         }
-        pip_args = []
+        pip_args = ["--extra-index-url https://download.pytorch.org/whl/" + _RUNTIME_VERSIONS.get("DEVICE")]
         for pip_key in meta_pip_args:
             if not isinstance(meta_pip_args[pip_key], (list, tuple, set)):
                 meta_pip_args[pip_key] = [meta_pip_args[pip_key]]

Diff for: .azure/ipynb-publish.yml

@@ -38,6 +38,7 @@ jobs:
         whereis nvidia
         nvidia-smi
         python --version
+        pip list | grep torch
       displayName: 'Image info & NVIDIA'
 
     - script: |

Diff for: .azure/ipynb-tests.yml

@@ -9,7 +9,7 @@ jobs:
 
   - job: nbval
     # how long to run the job before automatically cancelling
-    timeoutInMinutes: 55
+    timeoutInMinutes: 95
     # how much time to give 'run always even if cancelled tasks' before stopping them
     cancelTimeoutInMinutes: 2
 
@@ -32,6 +32,7 @@ jobs:
         whereis nvidia
         nvidia-smi
         python --version
+        pip list | grep torch
       displayName: 'Image info & NVIDIA'
 
     - script: |

Diff for: Makefile

@@ -33,5 +33,7 @@ clean:
 	rm -f ./*-folders.txt
 	rm -f ./*/**/*.ipynb
 	rm -rf ./*/**/.ipynb_checkpoints
+	rm -rf ./*/**/venv
+	rm -rf ./*/**/logs
 	rm -rf ./*/**/lightning_logs
 	rm -f ./*/**/requirements.txt

Diff for: course_UvA-DL/01-introduction-to-pytorch/Introduction_to_PyTorch.py

@@ -33,6 +33,7 @@
 # %matplotlib inline
 from IPython.display import set_matplotlib_formats
 from matplotlib.colors import to_rgba
+from torch import Tensor
 from tqdm.notebook import tqdm  # Progress bar
 
 set_matplotlib_formats("svg", "pdf")
@@ -82,10 +83,10 @@
 #
 # Let's first start by looking at different ways of creating a tensor.
 # There are many possible options, the most simple one is to call
-# `torch.Tensor` passing the desired shape as input argument:
+# `Tensor` passing the desired shape as input argument:
 
 # %%
-x = torch.Tensor(2, 3, 4)
+x = Tensor(2, 3, 4)
 print(x)
 
 # %% [markdown]
@@ -101,7 +102,7 @@
 
 # %%
 # Create a tensor from a (nested) list
-x = torch.Tensor([[1, 2], [3, 4]])
+x = Tensor([[1, 2], [3, 4]])
 print(x)
 
 # %%
@@ -648,9 +649,9 @@ def __getitem__(self, idx):
 
 # %%
 def visualize_samples(data, label):
-    if isinstance(data, torch.Tensor):
+    if isinstance(data, Tensor):
         data = data.cpu().numpy()
-    if isinstance(label, torch.Tensor):
+    if isinstance(label, Tensor):
         label = label.cpu().numpy()
     data_0 = data[label == 0]
     data_1 = data[label == 1]
@@ -935,9 +936,9 @@ def eval_model(model, data_loader):
 # %%
 @torch.no_grad()  # Decorator, same effect as "with torch.no_grad(): ..." over the whole function.
 def visualize_classification(model, data, label):
-    if isinstance(data, torch.Tensor):
+    if isinstance(data, Tensor):
         data = data.cpu().numpy()
-    if isinstance(label, torch.Tensor):
+    if isinstance(label, Tensor):
         label = label.cpu().numpy()
     data_0 = data[label == 0]
     data_1 = data[label == 1]
@@ -952,8 +953,8 @@ def visualize_classification(model, data, label):
 
     # Let's make use of a lot of operations we have learned above
     model.to(device)
-    c0 = torch.Tensor(to_rgba("C0")).to(device)
-    c1 = torch.Tensor(to_rgba("C1")).to(device)
+    c0 = Tensor(to_rgba("C0")).to(device)
+    c1 = Tensor(to_rgba("C1")).to(device)
     x1 = torch.arange(-0.5, 1.5, step=0.01, device=device)
     x2 = torch.arange(-0.5, 1.5, step=0.01, device=device)
     xx1, xx2 = torch.meshgrid(x1, x2)  # Meshgrid function as in numpy

Diff for: course_UvA-DL/04-inception-resnet-densenet/Inception_ResNet_DenseNet.py

@@ -197,7 +197,7 @@
 # %% [markdown]
 # Thus, in the future, we don't have to define our own `set_seed` function anymore.
 #
-# In PyTorch Lightning, we define `pl.LightningModule`'s (inheriting from `torch.nn.Module`) that organize our code into 5 main sections:
+# In PyTorch Lightning, we define `pl.LightningModule`'s (inheriting from `Module`) that organize our code into 5 main sections:
 #
 # 1. Initialization (`__init__`), where we create all necessary parameters/models
 # 2. Optimizers (`configure_optimizers`) where we create the optimizers, learning rate scheduler, etc.

Diff for: course_UvA-DL/06-graph-neural-networks/.meta.yml

@@ -27,5 +27,4 @@ pip__find-link:
   # - https://pytorch-geometric.com/whl/torch-1.8.0+cu101.html
   - https://pytorch-geometric.com/whl/torch-%(TORCH_MAJOR_DOT_MINOR)s.0+%(DEVICE)s.html
 accelerator:
-  - CPU
   - GPU

Diff for: course_UvA-DL/06-graph-neural-networks/GNN_overview.py

@@ -26,6 +26,7 @@
 
 # PL callbacks
 from pytorch_lightning.callbacks import ModelCheckpoint
+from torch import Tensor
 
 AVAIL_GPUS = min(1, torch.cuda.device_count())
 BATCH_SIZE = 256 if AVAIL_GPUS else 64
@@ -183,7 +184,7 @@ def forward(self, node_feats, adj_matrix):
 
 # %%
 node_feats = torch.arange(8, dtype=torch.float32).view(1, 4, 2)
-adj_matrix = torch.Tensor([[[1, 1, 0, 0], [1, 1, 1, 1], [0, 1, 1, 1], [0, 1, 1, 1]]])
+adj_matrix = Tensor([[[1, 1, 0, 0], [1, 1, 1, 1], [0, 1, 1, 1], [0, 1, 1, 1]]])
 
 print("Node features:\n", node_feats)
 print("\nAdjacency matrix:\n", adj_matrix)
@@ -195,8 +196,8 @@ def forward(self, node_feats, adj_matrix):
 
 # %%
 layer = GCNLayer(c_in=2, c_out=2)
-layer.projection.weight.data = torch.Tensor([[1.0, 0.0], [0.0, 1.0]])
-layer.projection.bias.data = torch.Tensor([0.0, 0.0])
+layer.projection.weight.data = Tensor([[1.0, 0.0], [0.0, 1.0]])
+layer.projection.bias.data = Tensor([0.0, 0.0])
 
 with torch.no_grad():
     out_feats = layer(node_feats, adj_matrix)
@@ -308,7 +309,7 @@ def __init__(self, c_in, c_out, num_heads=1, concat_heads=True, alpha=0.2):
 
         # Sub-modules and parameters needed in the layer
         self.projection = nn.Linear(c_in, c_out * num_heads)
-        self.a = nn.Parameter(torch.Tensor(num_heads, 2 * c_out))  # One per head
+        self.a = nn.Parameter(Tensor(num_heads, 2 * c_out))  # One per head
         self.leakyrelu = nn.LeakyReLU(alpha)
 
         # Initialization from the original implementation
@@ -376,9 +377,9 @@ def forward(self, node_feats, adj_matrix, print_attn_probs=False):
 
 # %%
 layer = GATLayer(2, 2, num_heads=2)
-layer.projection.weight.data = torch.Tensor([[1.0, 0.0], [0.0, 1.0]])
-layer.projection.bias.data = torch.Tensor([0.0, 0.0])
-layer.a.data = torch.Tensor([[-0.2, 0.3], [0.1, -0.1]])
+layer.projection.weight.data = Tensor([[1.0, 0.0], [0.0, 1.0]])
+layer.projection.bias.data = Tensor([0.0, 0.0])
+layer.a.data = Tensor([[-0.2, 0.3], [0.1, -0.1]])
 
 with torch.no_grad():
     out_feats = layer(node_feats, adj_matrix, print_attn_probs=True)

Diff for: course_UvA-DL/09-normalizing-flows/NF_image_modeling.py

@@ -25,6 +25,7 @@
 from IPython.display import HTML, display, set_matplotlib_formats
 from matplotlib.colors import to_rgb
 from pytorch_lightning.callbacks import LearningRateMonitor, ModelCheckpoint
+from torch import Tensor
 from torchvision import transforms
 from torchvision.datasets import MNIST
 from tqdm.notebook import tqdm
@@ -117,8 +118,8 @@ def discretize(sample):
 # %%
 def show_imgs(imgs, title=None, row_size=4):
     # Form a grid of pictures (we use max. 8 columns)
-    num_imgs = imgs.shape[0] if isinstance(imgs, torch.Tensor) else len(imgs)
-    is_int = imgs.dtype == torch.int32 if isinstance(imgs, torch.Tensor) else imgs[0].dtype == torch.int32
+    num_imgs = imgs.shape[0] if isinstance(imgs, Tensor) else len(imgs)
+    is_int = imgs.dtype == torch.int32 if isinstance(imgs, Tensor) else imgs[0].dtype == torch.int32
     nrow = min(num_imgs, row_size)
     ncol = int(math.ceil(num_imgs / nrow))
     imgs = torchvision.utils.make_grid(imgs, nrow=nrow, pad_value=128 if is_int else 0.5)
@@ -1317,7 +1318,7 @@ def interpolate(model, img1, img2, num_steps=8):
 
 
 # %%
-def visualize_dequant_distribution(model: ImageFlow, imgs: torch.Tensor, title: str = None):
+def visualize_dequant_distribution(model: ImageFlow, imgs: Tensor, title: str = None):
     """
     Args:
         model: The flow of which we want to visualize the dequantization distribution

Diff for: course_UvA-DL/10-autoregressive-image-modeling/Autoregressive_Image_Modeling.py

@@ -50,6 +50,7 @@
 from IPython.display import set_matplotlib_formats
 from matplotlib.colors import to_rgb
 from pytorch_lightning.callbacks import LearningRateMonitor, ModelCheckpoint
+from torch import Tensor
 from torchvision import transforms
 from torchvision.datasets import MNIST
 from tqdm.notebook import tqdm
@@ -133,7 +134,7 @@ def discretize(sample):
 
 # %%
 def show_imgs(imgs):
-    num_imgs = imgs.shape[0] if isinstance(imgs, torch.Tensor) else len(imgs)
+    num_imgs = imgs.shape[0] if isinstance(imgs, Tensor) else len(imgs)
     nrow = min(num_imgs, 4)
     ncol = int(math.ceil(num_imgs / nrow))
     imgs = torchvision.utils.make_grid(imgs, nrow=nrow, pad_value=128)
@@ -910,8 +911,8 @@ def autocomplete_image(img):
 # similar likelihoods. We can visualize a discrete logistic below:
 
 # %%
-mu = torch.Tensor([128])
-sigma = torch.Tensor([2.0])
+mu = Tensor([128])
+sigma = Tensor([2.0])
 
 
 def discrete_logistic(x, mu, sigma):

Diff for: course_UvA-DL/11-vision-transformer/Vision_Transformer.py

@@ -155,7 +155,7 @@
 def img_to_patch(x, patch_size, flatten_channels=True):
     """
     Inputs:
-        x - torch.Tensor representing the image of shape [B, C, H, W]
+        x - Tensor representing the image of shape [B, C, H, W]
         patch_size - Number of pixels per dimension of the patches (integer)
         flatten_channels - If True, the patches will be returned in a flattened format
                            as a feature vector instead of a image grid.

Diff for: course_UvA-DL/13-contrastive-learning/.meta.yml

@@ -21,5 +21,5 @@ requirements:
   - matplotlib
   - seaborn
 accelerator:
-  - GPU
   - CPU
+  - GPU

Diff for: lightning_examples/barlow-twins/barlow_twins.py

@@ -17,6 +17,7 @@
 import torchvision.transforms.functional as VisionF
 from pytorch_lightning import Callback, LightningModule, Trainer
 from pytorch_lightning.callbacks import ModelCheckpoint
+from torch import Tensor
 from torch.utils.data import DataLoader
 from torchmetrics.functional import accuracy
 from torchvision.datasets import CIFAR10
@@ -334,7 +335,7 @@ def on_fit_start(self, trainer: pl.Trainer, pl_module: pl.LightningModule) -> No
 
     def extract_online_finetuning_view(
         self, batch: Sequence, device: Union[str, torch.device]
-    ) -> Tuple[torch.Tensor, torch.Tensor]:
+    ) -> Tuple[Tensor, Tensor]:
         (_, _, finetune_view), y = batch
         finetune_view = finetune_view.to(device)
         y = y.to(device)

Diff for: requirements/default.txt

@@ -1,4 +1,5 @@
+setuptools==59.5.0
 ipython[notebook]
+torch>=1.8
 pytorch-lightning>=1.4
-torchmetrics>=0.6
-torch>=1.6, <1.9
+torchmetrics>=0.7