[Doc] specify backend as pytorch for tutorials (#5198)

tutorials/blitz/1_introduction.py

@@ -21,6 +21,8 @@ networks with PyTorch.
 
 """
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import torch
 import torch.nn as nn
 import torch.nn.functional as F

tutorials/blitz/2_dglgraph.py

@@ -30,6 +30,8 @@ By the end of this tutorial you will be able to:
 # center node to the leaves.
 #
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import numpy as np
 import torch
 

tutorials/blitz/3_message_passing.py

@@ -18,6 +18,8 @@ GNN for node classification <1_introduction>`.
 
 """
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import torch
 import torch.nn as nn
 import torch.nn.functional as F

tutorials/blitz/5_graph_classification.py

@@ -13,6 +13,8 @@ By the end of this tutorial, you will be able to
 (Time estimate: 18 minutes)
 """
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import torch
 import torch.nn as nn
 import torch.nn.functional as F

tutorials/blitz/6_load_data.py

@@ -88,7 +88,7 @@ interactions.head()
 #
 
 import os
-
+os.environ['DGLBACKEND'] = 'pytorch'
 import torch
 
 import dgl

tutorials/dist/1_node_classification.py

@@ -22,6 +22,8 @@ Here we store the node labels as node data in the DGL Graph.
 .. code-block:: python
 
 
+    import os
+    os.environ['DGLBACKEND'] = 'pytorch'
     import dgl
     import torch as th
     from ogb.nodeproppred import DglNodePropPredDataset

tutorials/dist/2_link_prediction.py

@@ -19,6 +19,8 @@ into a training graph, validation edges and test edges with :class:`~dgl.data.As
 .. code-block:: python
 
 
+    import os
+    os.environ['DGLBACKEND'] = 'pytorch'
     import dgl
     import torch as th
     from ogb.linkproppred import DglLinkPropPredDataset

tutorials/large/L1_large_node_classification.py

@@ -25,6 +25,8 @@ Sampling for GNN Training <L0_neighbor_sampling_overview>`.
 # OGB already prepared the data as DGL graph.
 #
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import dgl
 import torch
 import numpy as np

tutorials/large/L2_large_link_prediction.py

@@ -52,6 +52,8 @@ Sampling for Node Classification <L1_large_node_classification>`.
 # :doc:`previous tutorial <L1_large_node_classification>`.
 #
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import dgl
 import torch
 import numpy as np

tutorials/large/L4_message_passing.py

@@ -13,6 +13,8 @@ for stochastic GNN training. It assumes that
 
 """
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import dgl
 import torch
 import numpy as np

tutorials/models/1_gnn/1_gcn.py

@@ -45,6 +45,8 @@ message passing APIs.
 # aggregation on a node :math:`u` only involves summing over the neighbors'
 # representations :math:`h_v`, we can simply use builtin functions:
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import torch as th
 import torch.nn as nn
 import torch.nn.functional as F

tutorials/models/1_gnn/4_rgcn.py

@@ -136,6 +136,8 @@ multiple edges among any given pair.
 #    efficient :class:`builtin R-GCN layer module <dgl.nn.pytorch.conv.RelGraphConv>`.
 #
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import dgl
 import torch
 import torch.nn as nn

tutorials/models/1_gnn/6_line_graph.py

@@ -86,6 +86,8 @@ Line Graph Neural Network
 # The following code snippet verifies that there are more intra-class edges
 # than inter-class.
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import torch
 import torch as th
 import torch.nn as nn

tutorials/models/1_gnn/9_gat.py

@@ -104,6 +104,8 @@ structure-free normalization, in the style of attention.
 # DGL provides an off-the-shelf implementation of the GAT layer under the ``dgl.nn.<backend>``
 # subpackage. Simply import the ``GATConv`` as the follows.
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 from dgl.nn.pytorch import GATConv
 
 ###############################################################

tutorials/models/2_small_graph/3_tree-lstm.py

@@ -58,6 +58,8 @@ Tree-LSTM in DGL
 
 from collections import namedtuple
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import dgl
 from dgl.data.tree import SSTDataset
 

tutorials/models/3_generative_model/5_dgmg.py

@@ -50,6 +50,8 @@ Generative Models of Graphs
 # you can write the code as follows.
 #
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import dgl
 
 g = dgl.DGLGraph()

tutorials/models/4_old_wines/2_capsule.py

@@ -67,6 +67,8 @@ offers a different perspective. The tutorial describes how to implement a Capsul
 #
 # Here's how we set up the graph and initialize node and edge features.
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import matplotlib.pyplot as plt
 import numpy as np
 import torch as th

tutorials/multi/1_graph_classification.py

@@ -70,6 +70,8 @@ to start the distributed backend at the beginning of each process. We use
 process ID, which should be an integer from `0` to `world_size - 1`.
 """
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import torch.distributed as dist
 
 

tutorials/multi/2_node_classification.py

@@ -33,6 +33,8 @@ models with multi-GPU with ``DistributedDataParallel``.
 # tutorial.
 # 
 
+import os
+os.environ['DGLBACKEND'] = 'pytorch'
 import dgl
 import torch
 import numpy as np