larroy edited a comment on issue #13409: [MXNET-1234] Fix shape inference problems in Activation backward URL: https://github.com/apache/incubator-mxnet/pull/13409#issuecomment-443760161 This never executes on MKL: Only if you bind the relu to cpu() context it does. ``` import mxnet as mx import numpy as np np.random.seed(12345) num_filter = 256 num_group = 1 kernel = (3, 3) pad = (1, 1) shape = (1, 256, 200, 233) x = mx.sym.Variable('x') w = mx.sym.Variable('w') conv = mx.sym.Convolution(data=x, weight=w, num_filter=num_filter, num_group=num_group, kernel=kernel, no_bias=True, pad=pad) relu = mx.sym.Activation(data=conv, act_type='relu', name='relu') exe = relu.simple_bind(ctx=mx.gpu(), x=shape) exe.arg_arrays[0][:] = mx.nd.array(np.random.normal(size=exe.arg_arrays[0].shape), ctx=mx.cpu()) exe.arg_arrays[1][:] = mx.nd.array(np.random.normal(size=exe.arg_arrays[1].shape), ctx=mx.cpu()) for i in range(10): exe.forward(is_train=True) exe.backward(exe.outputs[0]) o = exe.grad_arrays[0] t = o.asnumpy() ``` Output is: ``` ActivationGradCompute ActivationGradComputeImpl ActivationBackward ActivationCompute ActivationComputeImpl ActivationForward ActivationGradCompute ActivationGradComputeImpl ActivationBackward ```
---------------------------------------------------------------- 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: [email protected] With regards, Apache Git Services
