Conversation
1) added interpretation.utils.get_group_index(): function returns group_index parameter required for training SGL probe. 2) added interpretation.linear_probe._train_sgl_probe() : function to train a logistic regression probe with SGL regularization using GroupLasso library object. 3) imported libraries -time and -LogisticGroupLasso in interpretation.linear_probe
…_probe() to include options for regularization method Edited function interpretation.linear_probe.train_logistic_regression_probe() to accomodate options for Sparse Group Lasso and Group Lasso Regularization. Function will choose ElasticNet Regularization by default
…be parameter initialization this function is called within the function interpretation.linear_probe.train_logistic_regression_probe()
fdalvi
left a comment
fdalvi
left a comment
There was a problem hiding this comment.
Looks pretty good to me, if you can just generally add comments to the new functions you've added, I think we will be good to merge!
| plt.title('Layer '+str(layer)+' weights') | ||
| plt.show() | ||
|
|
||
| def sparsity(model_name, data, **kwargs) : |
There was a problem hiding this comment.
This and the following functions seem "internal" to the lib, can you prepend them with an underscore so the name becomes _sparsity? Also it might be better to name it to something a bit more clear, is this plotting sparsity? Computing it?
| import torch | ||
| import torch.nn as nn | ||
| from torch.autograd import Variable | ||
| from group_lasso import LogisticGroupLasso |
There was a problem hiding this comment.
Please add this lib to requirements.txt along with a version that you know works well
| regularization = "elastic_net", | ||
| **kwargs | ||
| ): | ||
|
|
There was a problem hiding this comment.
can you update the doc string below with the new "regularization" parameter?
|
Hello @sherinrasheed, hope you are well! I know you've been busy lately, but it would be awesome if we can push these changes (its mostly documentation related). We want to do some major changes to the codebase and merging this after those changes would be a bit painful. Let me know if you need any assistance! |
2 participants
No description provided.