Competition score issue


#1

when i am training my model on the deep learning studio,it shows the accuracy of 0.98(validation)
but when i submit my model the competion score is just .125?Can someone tell me where i may be wrong??


#2

Jayesh, as mentioned in other threads, there is a bug which affects inference using pre-trained models. We have fixed it in the cloud version and will be releasing fix for desktop version soon.

In the meantime, you can make following change in the desktop version to fix this issue.

mxnet_backend.py is located in the /usr/local/lib/python3.5/dist-packages/keras/backend/ on Linux.
mxnet_backend.py is located in the <DLS_INSTALL_FOLDER>/conda3/Lib/site-packages/keras/backend on Windows.

diff --git a/patches/keras/mxnet_backend.py b/patches/keras/mxnet_backend.py

@@ -203,7 +203,10 @@ class KerasSymbol(object):
     def bind(self, data):
-        self.tensor = data
+        if not hasattr(self, 'tensor'):
+            self.tensor = data
+        else:
+            self.tensor[:] = data
         if self.name in self._bind_values:
             assert self._bind_values[self.name].shape == data.shape, \
                 "Redefinition of variable %s" % self.name

Soda Bottle Identification Challenge
Soda Bottle Identification Challenge