mnist one layer to row major

contrib/Python/cntk/examples/LogReg/logreg.py

@@ -49,7 +49,7 @@ def train_eval_logistic_regression_from_file(criterion_name=None,
     my_sgd = C.SGDParams(
         epoch_size=0, minibatch_size=25, learning_rates_per_mb=0.1, max_epochs=3)
 
-    with C.LocalExecutionContext('logreg', device_id=device_id, clean_up=True) as ctx:
+    with C.LocalExecutionContext('logreg', device_id=device_id, clean_up=False) as ctx:
         ctx.train(
             root_nodes=[ce, eval], 
             training_params=my_sgd,

contrib/Python/cntk/examples/MNIST/mnist_one_layer.py

@@ -22,17 +22,17 @@ import cntk as C
 
 
 def add_dnn_sigmoid_layer(in_dim, out_dim, x, param_scale):
-    W = C.parameter((out_dim, in_dim)) * param_scale
-    b = C.parameter((out_dim, 1)) * param_scale
-    t = C.times(W, x)
+    W = C.parameter((in_dim, out_dim)) * param_scale
+    b = C.parameter((1, out_dim)) * param_scale
+    t = C.times(x, W)
     z = C.plus(t, b)
     return C.sigmoid(z)
 
 
 def add_dnn_layer(in_dim, out_dim, x, param_scale):
-    W = C.parameter((out_dim, in_dim)) * param_scale
-    b = C.parameter((out_dim, 1)) * param_scale
-    t = C.times(W, x)
+    W = C.parameter((in_dim, out_dim)) * param_scale
+    b = C.parameter((1, out_dim)) * param_scale
+    t = C.times(x, W)
     return C.plus(t, b)
 
 def train_eval_mnist_onelayer_from_file(criterion_name=None, eval_name=None):
@@ -73,11 +73,11 @@ def train_eval_mnist_onelayer_from_file(criterion_name=None, eval_name=None):
     eval.tag = 'eval'
     
     # Specify the training parameters (settings are scaled down)
-    my_sgd = C.SGDParams(epoch_size=600, minibatch_size=32,
-                       learning_rates_per_mb=0.1, max_epochs=5, momentum_per_mb=0)
+    my_sgd = C.SGDParams(epoch_size=60000, minibatch_size=32,
+                       learning_rates_per_mb=0.1, max_epochs=30, momentum_per_mb=0)
 
     # Create a context or re-use if already there
-    with C.LocalExecutionContext('mnist_one_layer', clean_up=True) as ctx:
+    with C.LocalExecutionContext('mnist_one_layer', clean_up=False) as ctx:
         # CNTK actions
          ctx.train(
             root_nodes=[ec, eval],

contrib/Python/cntk/ops/__init__.py

@@ -394,7 +394,8 @@ def times(left, right, output_rank=1, name=None):
     # CNTK uses column vectors and column major representation, thus we reverse
     # params and adjust output_rank    
     output_rank = left.rank + 1 - output_rank
-    op = Times(right, left, outputRank=output_rank, name=name)
+    #TODO: output_rank is not correct fix it and repalce the hardcoded (1)
+    op = Times(right, left, outputRank=1, name=name)
     op.rank = left.rank + right.rank -2
     return op