Vikas89 edited a comment on issue #14340: [bug] Bug in Gradient flow with backward(retain_graph=True) and split() URL: https://github.com/apache/incubator-mxnet/issues/14340#issuecomment-483452493 @ThomasDelteil I tried running code on my mac , ``` import mxnet as mx from mxnet import gluon, autograd, nd for i in range(1): ctx = mx.cpu() data = mx.nd.ones((1,10,), ctx=ctx) param = mx.nd.ones((1,10,), ctx=ctx) data.attach_grad(grad_req='write') param.attach_grad(grad_req='write') with autograd.record(): z = data*param # print(z.shape()) z1, z2 = z.split(2,1) #z1, z2 = z[:,:5], z[:,5:] z1.backward(retain_graph=True) print(data.grad.asnumpy()) z2.backward(retain_graph=True) print(data.grad.asnumpy()) print() break ``` The results I get is : ``` [[1. 1. 1. 1. 1. 0. 0. 0. 0. 0.]] [[ 0.0000000e+00 2.0000000e+00 2.5715471e+36 -4.6577453e-10 1.6986417e-14 3.6433760e-44 4.8807685e+13 6.0397474e+26 4.3427447e-20 1.4012985e-45]] ``` I guess, the expected output of print(data.grad.asnumpy()) is : `[[0. 0. 0. 0. 0. 1. 1. 1. 1. 1.]] ` Can you confirm my understanding is correct ? I also verified that split_v2 is giving correct results, ie, `[[0. 0. 0. 0. 0. 1. 1. 1. 1. 1.]] `
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to 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
