servicenow/beyond-trivial-explanations

Beyond Trivial Counterfactual Explanations with Diverse Valuable Explanations is a ServiceNow Research project that was started at Element AI.

beyond-trivial-explanations's Issues

training error

hello

I got an error on command(python trainval.py -e random_search -sb ./logs/bte -d ./data -r 1)

my error:
Traceback (most recent call last):
File "trainval.py", line 155, in
results_fname='results.ipynb')
File "/hd12/anaconda3/envs/dive_3.6/lib/python3.6/site-packages/haven/haven_wizard.py", line 145, in run_wizard
func(exp_dict=exp_dict, savedir=savedir, args=args)
File "trainval.py", line 76, in trainval
train_dict = model.train_on_loader(epoch, train_loader)
File "/hd12/anaconda3/envs/dive_3.6/lib/python3.6/site-packages/torch/nn/modules/module.py", line 1178, in getattr
type(self).name, name))
AttributeError: 'GradientAttack' object has no attribute 'train_on_loader'

and I also checked that the GradientAttack class does not have the train_on_loader attribute.

I am not sure what is the cause of the problem, looking forward to your help.

training problem

hello,
I want to train it to reproduce the results again(just to learn) but I got an error on the third command(python trainval.py -e oracle -sb ./logs/oracle -d ./data -r 1)
my error:

Traceback (most recent call last):
File "/content/beyond-trivial-explanations/trainval.py", line 140, in
hw.run_wizard(func=trainval,
File "/usr/local/lib/python3.10/dist-packages/haven/haven_wizard.py", line 99, in run_wizard
exp_list += exp_groups[exp_group_name]
KeyError: 'oracle'

to solve this problem although I don't know if it is true or not... I replaced "oracle" with "cls_oracle" and I run the command below:
python trainval.py -e cls_oracle -sb ./logs/oracle -d ./data -r 1
again this time I got the below error:
FileNotFoundError: [Errno 2] No such file or directory: './data/pretrained_models/resnet50_128.pth'

can you please help me...

Results

Hi,

I've been playing with this code. I don't know if you can solve me some questions about it.

1. Do you know the success rate of your method? Specially DiVE with and without Fisher Spectral?
2. Can you tell me something about the diversity of the counterfactuals? The generated results for DiVE do not have any diversity, practically they are all the same for the 8 generated CF. For DiVE Fisher_spectal I see a little bit of diversity, but the success of the CF is ~22%.
3. You computed the FID with the Pytorch or TensorFlow version? I don't manage to get the same FID from your paper.

haven file can not get

thank you for you