pxiangwu/topofilter

NeurIPS 2020, "A Topological Filter for Learning with Label Noise".

topofilter's Issues

some question about function "calc_topo_weights_with_components_idx"

Hi, thanks for the nice code.
In the train.py, some lines confuse me as follows:

lines 277,283 and 284, define some vectors

     prob_all[indices] = softmax_outputs.detach().cpu()
     train_gt_labels[indices] = labels.cpu().numpy().astype(np.int64)
     train_pred_labels[indices] = labels.cpu().numpy().astype(np.int64)

why the "train_gt_labels" is the same as "train_pred_labels" both of which are taken as input to the function calc_topo_weights_with_components_idx? and the "prob_all" is defined but not used.

issues with the filling the variables.

Hi,
I'm wondering why in the train.py file you are feeding the same values to the ground truth label variable and the predicted label variable:

train_gt_labels[indices] = labels.cpu().numpy().astype(np.int64)
train_pred_labels[indices] = labels.cpu().numpy().astype(np.int64)

Thanks,
Naghmeh

Can you share the bash script used to reproduce your results?

cannot pip package named PythonGraphPers_withCompInfo

hi，sorry to bother you.
i cannot pip the package named PythonGraphPers_withCompInfo in knn_untils.py.
where to download this package？
thanks for answering my question.

tabular data/ noisy instances

Hi,
thanks for sharing your implementation. I have two questions about it:

1. Does it also work on tabular data?
2. Is it possible to identify the noisy instances (return the noisy IDs or the clean set)?

Thanks!