Merge pull request lisa-lab#19 from lisa-lab/fix_iter_index

[WIP] Fix off-by-one error in numbering of epochs in most tutorials

Author: pascanur

code/DBN.py

@@ -359,7 +359,7 @@ def test_DBN(finetune_lr=0.1, pretraining_epochs=100,
         for minibatch_index in xrange(n_train_batches):
 
             minibatch_avg_cost = train_fn(minibatch_index)
-            iter = epoch * n_train_batches + minibatch_index
+            iter = (epoch - 1) * n_train_batches + minibatch_index
 
             if (iter + 1) % validation_frequency == 0:
 

code/SdA.py

@@ -395,9 +395,10 @@ def test_SdA(finetune_lr=0.1, pretraining_epochs=15,
     epoch = 0
 
     while (epoch < training_epochs) and (not done_looping):
+        epoch = epoch + 1
         for minibatch_index in xrange(n_train_batches):
             minibatch_avg_cost = train_fn(minibatch_index)
-            iter = epoch * n_train_batches + minibatch_index
+            iter = (epoch - 1) * n_train_batches + minibatch_index
 
             if (iter + 1) % validation_frequency == 0:
                 validation_losses = validate_model()
@@ -429,7 +430,6 @@ def test_SdA(finetune_lr=0.1, pretraining_epochs=15,
             if patience <= iter:
                 done_looping = True
                 break
-        epoch = epoch + 1
 
     end_time = time.clock()
     print(('Optimization complete with best validation score of %f %%,'

code/convolutional_mlp.py

@@ -246,7 +246,7 @@ def evaluate_lenet5(learning_rate=0.1, n_epochs=200,
         epoch = epoch + 1
         for minibatch_index in xrange(n_train_batches):
 
-            iter = epoch * n_train_batches + minibatch_index
+            iter = (epoch - 1) * n_train_batches + minibatch_index
 
             if iter % 100 == 0:
                 print 'training @ iter = ', iter
@@ -296,7 +296,7 @@ def evaluate_lenet5(learning_rate=0.1, n_epochs=200,
                           ' ran for %.2fm' % ((end_time - start_time) / 60.))
 
 if __name__ == '__main__':
-    evaluate_lenet5()
+    evaluate_lenet5(n_epochs=1, batch_size=1)
 
 
 def experiment(state, channel):

code/logistic_sgd.py

@@ -323,7 +323,7 @@ def sgd_optimization_mnist(learning_rate=0.13, n_epochs=1000,
 
             minibatch_avg_cost = train_model(minibatch_index)
             # iteration number
-            iter = epoch * n_train_batches + minibatch_index
+            iter = (epoch - 1) * n_train_batches + minibatch_index
 
             if (iter + 1) % validation_frequency == 0:
                 # compute zero-one loss on validation set

code/mlp.py

@@ -309,7 +309,7 @@ def test_mlp(learning_rate=0.01, L1_reg=0.00, L2_reg=0.0001, n_epochs=1000,
 
             minibatch_avg_cost = train_model(minibatch_index)
             # iteration number
-            iter = epoch * n_train_batches + minibatch_index
+            iter = (epoch - 1) * n_train_batches + minibatch_index
 
             if (iter + 1) % validation_frequency == 0:
                 # compute zero-one loss on validation set

doc/gettingstarted.txt

@@ -576,17 +576,19 @@ of a strategy based on a geometrically increasing amount of patience.
     done_looping = False
     epoch = 0
     while (epoch < n_epochs) and (not done_looping):
+        # Report "1" for first epoch, "n_epochs" for last epoch
         epoch = epoch + 1
         for minibatch_index in xrange(n_train_batches):
 
             d_loss_wrt_params = ... # compute gradient
             params -= learning_rate * d_loss_wrt_params # gradient descent
 
-            # iteration number
-            iter = epoch * n_train_batches + minibatch_index
+            # iteration number. We want it to start at 0.
+            iter = (epoch - 1) * n_train_batches + minibatch_index
             # note that if we do `iter % validation_frequency` it will be
-            # true for iter = 0 which we do not want
-            if iter and iter % validation_frequency == 0:
+            # true for iter = 0 which we do not want. We want it true for
+            # iter = validation_frequency - 1.
+            if (iter + 1) % validation_frequency == 0:
 
                 this_validation_loss = ... # compute zero-one loss on validation set
 

doc/logreg.txt

@@ -396,12 +396,12 @@ The output one should expect is of the form :
 .. code-block:: bash
 
     ...
+    epoch 72, minibatch 83/83, validation error 7.510417 %
+         epoch 72, minibatch 83/83, test error of best model 7.510417 %
     epoch 73, minibatch 83/83, validation error 7.500000 %
-        epoch 73, minibatch 83/83, test error of best model 7.489583 %
-    epoch 74, minibatch 83/83, validation error 7.479167 %
-        epoch 74, minibatch 83/83, test error of best model 7.489583 %
-    Optimization complete with best validation score of 7.479167 %,with test performance 7.489583 %
-    The code run for 75 epochs, with 1.936983 epochs/sec
+         epoch 73, minibatch 83/83, test error of best model 7.489583 %
+    Optimization complete with best validation score of 7.500000 %,with test performance 7.489583 %
+    The code run for 74 epochs, with 1.936983 epochs/sec
 
 
 On an Intel(R) Core(TM)2 Duo CPU E8400 @ 3.00 Ghz  the code runs with