Merge branch 'travis'

.gitignore

@@ -1,2 +1,3 @@
 .idea
+*.pyc
 datasets/mnist

.travis.yml

@@ -6,18 +6,15 @@ python:
   - 3.6
 cache: bundler
 install:
-  - if [[ "$TRAVIS_PYTHON_VERSION" == "2.7" ]]; then
-      pip install -q http://download.pytorch.org/whl/cu75/torch-0.1.12.post2-cp27-none-linux_x86_64.whl;
-    else
-      pip install -q http://download.pytorch.org/whl/cu75/torch-0.1.12.post2-cp36-cp36m-linux_x86_64.whl;
-    fi
-  - pip install -q torchvision
-  - pip install -q numpy==1.13.0
+  - pip install --no-cache-dir -q torch
+
+# We use timeout to prevent the Python process from running too long since the long running process will be
+# auto-killed by Travis. When the timeout happens, we override its exit code 143 by the success exit code 0.
 script:
-  - python 0_multiply.py
-  - python 1_linear_regression.py
-  - python 2_logistic_regression.py
-  - python 3_neural_net.py
-  - python 4_modern_neural_net.py
-  - python 5_convolutional_net.py
-  - python 6_lstm.py
+  - timeout --preserve-status 2m python 0_multiply.py            || if [ $? -eq 0 ] || [ $? -eq 143 ]; then exit 0; fi
+  - timeout --preserve-status 2m python 1_linear_regression.py   || if [ $? -eq 0 ] || [ $? -eq 143 ]; then exit 0; fi
+  - timeout --preserve-status 5m python 2_logistic_regression.py || if [ $? -eq 0 ] || [ $? -eq 143 ]; then exit 0; fi
+  - timeout --preserve-status 5m python 3_neural_net.py          || if [ $? -eq 0 ] || [ $? -eq 143 ]; then exit 0; fi
+  - timeout --preserve-status 5m python 4_modern_neural_net.py   || if [ $? -eq 0 ] || [ $? -eq 143 ]; then exit 0; fi
+  - timeout --preserve-status 10m python 5_convolutional_net.py  || if [ $? -eq 0 ] || [ $? -eq 143 ]; then exit 0; fi
+  - timeout --preserve-status 10m python 6_lstm.py               || if [ $? -eq 0 ] || [ $? -eq 143 ]; then exit 0; fi

1_linear_regression.py

@@ -17,7 +17,7 @@ def train(model, loss, optimizer, x, y):
     optimizer.zero_grad()
 
     # Forward
-    fx = model.forward(x.view(len(x), 1))
+    fx = model.forward(x.view(len(x), 1)).squeeze()
     output = loss.forward(fx, y)
 
     # Backward
@@ -26,7 +26,7 @@ def train(model, loss, optimizer, x, y):
     # Update parameters
     optimizer.step()
 
-    return output.data[0]
+    return output.item()
 
 
 def main():
@@ -35,7 +35,7 @@ def main():
     Y = 2 * X + torch.randn(X.size()) * 0.33
 
     model = build_model()
-    loss = torch.nn.MSELoss(size_average=True)
+    loss = torch.nn.MSELoss(reduction='elementwise_mean')
     optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.9)
     batch_size = 10
 

2_logistic_regression.py

@@ -33,7 +33,7 @@ def train(model, loss, optimizer, x_val, y_val):
     # Update parameters
     optimizer.step()
 
-    return output.data[0]
+    return output.item()
 
 
 def predict(model, x_val):
@@ -52,7 +52,7 @@ def main():
     n_examples, n_features = trX.size()
     n_classes = 10
     model = build_model(n_features, n_classes)
-    loss = torch.nn.CrossEntropyLoss(size_average=True)
+    loss = torch.nn.CrossEntropyLoss(reduction='elementwise_mean')
     optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.9)
     batch_size = 100
 

3_neural_net.py

@@ -32,7 +32,7 @@ def train(model, loss, optimizer, x_val, y_val):
     # Update parameters
     optimizer.step()
 
-    return output.data[0]
+    return output.item()
 
 
 def predict(model, x_val):
@@ -51,7 +51,7 @@ def main():
     n_examples, n_features = trX.size()
     n_classes = 10
     model = build_model(n_features, n_classes)
-    loss = torch.nn.CrossEntropyLoss(size_average=True)
+    loss = torch.nn.CrossEntropyLoss(reduction='elementwise_mean')
     optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.9)
     batch_size = 100
 

4_modern_neural_net.py

@@ -36,7 +36,7 @@ def train(model, loss, optimizer, x_val, y_val):
     # Update parameters
     optimizer.step()
 
-    return output.data[0]
+    return output.item()
 
 
 def predict(model, x_val):
@@ -55,7 +55,7 @@ def main():
     n_examples, n_features = trX.size()
     n_classes = 10
     model = build_model(n_features, n_classes)
-    loss = torch.nn.CrossEntropyLoss(size_average=True)
+    loss = torch.nn.CrossEntropyLoss(reduction='elementwise_mean')
     optimizer = optim.Adam(model.parameters())
     batch_size = 100
 

5_convolutional_net.py

@@ -50,7 +50,7 @@ def train(model, loss, optimizer, x_val, y_val):
     # Update parameters
     optimizer.step()
 
-    return output.data[0]
+    return output.item()
 
 
 def predict(model, x_val):
@@ -72,7 +72,7 @@ def main():
     n_examples = len(trX)
     n_classes = 10
     model = ConvNet(output_dim=n_classes)
-    loss = torch.nn.CrossEntropyLoss(size_average=True)
+    loss = torch.nn.CrossEntropyLoss(reduction='elementwise_mean')
     optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.9)
     batch_size = 100
 

6_lstm.py

@@ -40,7 +40,7 @@ def train(model, loss, optimizer, x_val, y_val):
     # Update parameters
     optimizer.step()
 
-    return output.data[0]
+    return output.item()
 
 
 def predict(model, x_val):
@@ -73,7 +73,7 @@ def main():
     trY = torch.from_numpy(trY).long()
 
     model = LSTMNet(input_dim, hidden_dim, n_classes)
-    loss = torch.nn.CrossEntropyLoss(size_average=True)
+    loss = torch.nn.CrossEntropyLoss(reduction='elementwise_mean')
     optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.9)
 
     for i in range(epochs):

README.md

@@ -1,4 +1,4 @@
-<!-- [![Build Status](https://travis-ci.org/vinhkhuc/PyTorch-Mini-Tutorials.svg?branch=master)](https://travis-ci.org/vinhkhuc/PyTorch-Mini-Tutorials) -->
+[![Build Status](https://travis-ci.org/vinhkhuc/PyTorch-Mini-Tutorials.svg?branch=master)](https://travis-ci.org/vinhkhuc/PyTorch-Mini-Tutorials)
 
 Minimal tutorials for [PyTorch](https://github.com/pytorch/pytorch) adapted
 from Alec Radford's [Theano tutorials](https://github.com/Newmu/Theano-Tutorials).