anthonymrios/multi-label-zero-shot

Few- and Zero-shot Multi-Label Learning for Structured Label Spaces

multi-label-zero-shot's Issues

Confirming the fraction of non-leaf and leaf nodes

Hi, thanks for this great work. I have a question about the fraction of the labels. The paper says that for MIMIC3 data, "level 2 (leaf level) makes up about 33%". However, non-leaf ICD9 labels are not billable and so, the billable leaf ICD9 are strongly encouraged to be used instead.

I converted all non-decimal labels back to decimals (i.e. code 40301 into 403.01). Then, I counted the fraction of ICD9 labels which were used, and found that level 2 (leaf level) makes up 98% of all the labels found for the patients. Would you be able to tell me how did you count the fraction of leaf nodes?

Thanks.

Small question about the construction of adjacency matrix and training code

Hello, thank you for doing this great job. I have a few questions about this open source code:

1. How to construct mimic3_adj_matrix.pkl matrix
2. What are the specific meanings of lines 93 to 98 in train.py?

    label_cooc = Y_train.T.dot(Y_train)
    topk = label_cooc.argsort(axis=1)[:, -10:]
    label_cooc[label_cooc > 0] = 0.
    label_cooc[np.repeat(np.arange(label_cooc.shape[0]), 10), topk.flatten()] = 1.
    label_cooc = label_cooc + np.eye(label_cooc.shape[0])
    label_cooc[label_cooc > 0] = 1.

Thanks.

Why 7042

Detail about baseline ESZSL

In the paper you describe the comparison with ESZSL, you describe how you set the label representation. But I wonder how you get the text representation? How do you encode the text?

Question: size of the prediction label set

Hi!

It is a really interesting paper. My question is that when you do prediction, are you predicting on the whole ICD9 label set or just on the labels that have been appeared in the MIMIC dataset?

Thanks,
Xindi

sigmoid on prediction of each label

In the paper, when generate the sigmoid prediction of each label, in the formula, it is not clear what kind of multiplication it refers to in yˆi = sigmoid(eTi vi3), i = 1,...,L.. (For instance, dimension of eT is [batch_size, q+d, num_labels] and v3 is [q+d, num_labels], I am not quite get how the multiplication is expected to be like. ) I am wondering could you please explain and refer to the codes where this formula is implemented if possible? Thank you!