An Issue with solve_1d_linesearch_quad Function

[Nuwaisir-1998]

I had this error: TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.

After some observation, I found that inside solve_1d_linesearch_quad function in optim.py, np.divide is used. I think it should be nx.divide where nx = get_backend(a, b, c). And inside TorchBackend class, a new method for divide should be added that uses torch.div function.

[rflamary]

Hello @Nuwaisir-1998 very good point!

We are reworking all the optimization pipeline in PR #431 and I seem to recall that this has been corrected. @cedricvincentcuaz could you confirm that the PR correct this bug?

[cedricvincentcuaz]

Hello @Nuwaisir-1998 and @rflamary.

Indeed this np.divide operation has been removed in PR #431 to prevent this kind of error. The minimum is now computed using only python operators as minimum = min(1., max(0., -b / (2.0 * a))) which inherits the type of a and b, whether there are 1D tensors or floats. Note that the priority system of any backend implies that if a is a tensor and b a float, minimum will be a tensor of the same type than a (and vice-versa).

I do not think that a new method nx.divide is necessary in this function. Please correct me if I am wrong.

[Nuwaisir-1998]

Hi @cedricvincentcuaz,
I think you are correct, thanks!