[GitHub] sandeep-krishnamurthy commented on a change in pull request #13717: ONNX import: Hardmax

[GitBox <gi...@apache.org>]

sandeep-krishnamurthy commented on a change in pull request #13717: ONNX import: Hardmax
URL: https://github.com/apache/incubator-mxnet/pull/13717#discussion_r244046775
 
 

 ##########
 File path: python/mxnet/contrib/onnx/onnx2mx/_op_translations.py
 ##########
 @@ -714,3 +714,28 @@ def spacetodepth(attrs, inputs, proto_obj):
     new_attrs = translation_utils._fix_attribute_names(attrs, {'blocksize':'block_size'})
 
     return "space_to_depth", new_attrs, inputs
+
+
+def hardmax(attrs, inputs, proto_obj):
+    """Returns batched one-hot vectors."""
+    input_tensor_data = proto_obj.model_metadata.get('input_tensor_data')[0]
+    input_shape = input_tensor_data[1]
+
+    axis = int(attrs.get('axis', 1))
+    axis = axis if axis >= 0 else len(input_shape) + axis
+
+    if axis == len(input_shape) - 1:
+        amax = symbol.argmax(inputs[0], axis=-1)
+        one_hot = symbol.one_hot(amax, depth=input_shape[-1])
+        return one_hot, attrs, inputs
+
+    # since reshape doesn't take a tensor for shape,
+    # computing with np.prod. This needs to be changed to
+    # to use mx.sym.prod() when mx.sym.reshape() is fixed.
 
 Review comment:
   github issue link?

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services