TCAV score computation doesn't preserve experimental set order

[pris0]

🐛 Bug

After creating a TCAV instance, when one calls the .interpret() method, the yielded dictionary of scores don't follow the order of the experimental sets given in input. This creates confusion.

To Reproduce

Steps to reproduce the behavior:

1. Define some experimental sets. In my case I had 25 pairs of "base concept - random concept j" for j = 0, .. 24 for every base concept. The order is: base concept 0 - random concept 0, base concept 0 - random concept 1, ... base concept 0 - random concept 24, base concept 1 - random concept 0, base concept 1 - random concept 1, .. etc
2. Create a TCAV instnce mytcav = TCAV( model=model, layers=["conv1", "conv2"], layer_attr_method=LayerIntegratedGradients(model, None, multiply_by_inputs=False),)
3. Compute scores using the interpret function:
   scores = mytcav.interpret(inputs=input_class, experimental_sets=experimental_set_rand, target=target_id,)
4. Compare the order of the experimental sets and of the produced scores.

Expected behavior

The keys of the scores' dictionary should follow the order of the experimental sets given in input.

• Captum / PyTorch Version (e.g., 1.0 / 0.4.0): 0.7.0
• OS (e.g., Linux): Linux
• How you installed Captum / PyTorch (conda, pip, source): conda
• Python version: 3.12.2