186056b878a0a9dd576b684106fb4ddaa4af246c,theano/gpuarray/tests/test_dnn.py,,test_dnn_batchnorm_train_without_running_averages,#,1480

Before Change

    param_shape = (1, 10, 30, 25)

    // forward pass
    out, x_mean, x_invstd = bn.batch_normalization_train(x, scale, bias, "per-activation")
    // backward pass
    grads = T.grad(None, wrt=[x, scale, bias], known_grads={out: dy})
    // compile

After Change

        bn.batch_normalization_train(x, scale, bias, "per-activation")
    // backward pass
    grads_gpu = T.grad(None, wrt=[x, scale, bias], known_grads={out_gpu: dy})
    grads_abstract = T.grad(None, wrt=[x, scale, bias], known_grads={out_gpu: dy})
    // compile
    f_gpu = theano.function([x, scale, bias, dy],
                            [out_gpu, x_mean_gpu, x_invstd_gpu] +
                            grads_gpu,
                            mode=mode_with_gpu)
    f_abstract = theano.function([x, scale, bias, dy],
                                 [out_abstract, x_mean_abstract, x_invstd_abstract] +
                                 grads_abstract,
                                 mode=mode_with_gpu)
    // check if the abstract Ops have been replaced
    assert any([isinstance(n.op, dnn.GpuDnnBatchNorm)
                for n in f_abstract.maker.fgraph.toposort()])
    assert any([isinstance(n.op, dnn.GpuDnnBatchNormGrad)

Instances