IRV update #1054
IRV update #1054
Conversation
| out_tensor = tf.nn.l2_loss(self.IRVLayer.W) + \ | ||
| tf.nn.l2_loss(self.IRVLayer.V) + tf.nn.l2_loss(self.IRVLayer.b) + \ | ||
| tf.nn.l2_loss(self.IRVLayer.b2) | ||
| out_tensor = out_tensor * self.penalty |
There was a problem hiding this comment.
IRV has l2 regularization, is there a better way to handle these extra loss values?
| out_tensor = tf.nn.l2_loss(self.IRVLayer.W) + \ | ||
| tf.nn.l2_loss(self.IRVLayer.V) + tf.nn.l2_loss(self.IRVLayer.b) + \ | ||
| tf.nn.l2_loss(self.IRVLayer.b2) | ||
| out_tensor = out_tensor * self.penalty |
| @@ -1184,6 +1184,26 @@ def create_tensor(self, in_layers=None, set_tensors=True, **kwargs): | |||
| return out_tensor | |||
|
|
|||
|
|
|||
| class Sigmoid(Layer): | |||
There was a problem hiding this comment.
Could you add a docstring for this class along with a usage example?
|
|
||
|
|
||
| class Slice(Layer): | ||
| """ Choose a slice of input given axis and order |
There was a problem hiding this comment.
Could you add a usage example for this class in the docstring?
| @@ -1531,6 +1551,27 @@ def create_tensor(self, in_layers=None, set_tensors=True, **kwargs): | |||
| return out_tensor | |||
|
|
|||
|
|
|||
| class SigmoidCrossEntropy(Layer): | |||
There was a problem hiding this comment.
Could you add a usage example and docstring for this class?
|
|
||
|
|
||
| class TensorflowMultiTaskIRVClassifier(TensorflowLogisticRegression): | ||
| class IRVLayer(Layer): |
There was a problem hiding this comment.
Could you add a unit test for these and the other layers added?
|
@miaecle Thanks for porting! Made a few comments with requests for more docs and tests. |
|
@rbharath Thanks for the review! Just added some more docstrings and unit tests, let me know if anything else could be improved. |
|
@rbharath No problem, I will handle it. |
|
Thanks for the fixes! LGTM |
This PR adds in:
Tensorgraph version of IRV with identical structure
TensorflowLogisticRegression is removed and standard sklearn logistic regression is used in examples and benchmark