{Bug] Hardcoded Absolute Paths in BYOL/Inference Scripts (Transformers Classification DeepLense)

[invo-coder19]

Several Python scripts in the Transformers_Classification_DeepLense_Kartik_Sachdev sub-project contain hardcoded absolute paths to the original developer's local machine. This makes fine-tuning and inference completely non-functional for any other contributor without manually editing source files.

[Shrivadhu]

This is an important fix — hardcoded absolute paths definitely break portability and reproducibility.

Just a small suggestion:

• Should we also consider adding a default config file (e.g., YAML/JSON) for dataset/model paths instead of relying only on argparse?
• It might also help to add a README snippet showing an example CLI command for fine-tuning and inference after this change.
• Maybe we could validate the provided path and raise a clear error message if it doesn't exist.

Happy to help with improving documentation or adding basic path validation if needed.

[vignathi123-vi]

Hey @invo-coder19,
I reviewed the issue and the linked PR. The proposed fix looks good.
I’d be happy to help test the changes on a fresh setup (Windows/Linux/macOS) or assist with adding path validation / documentation improvements if needed.
Please let me know if there’s anything additional I can contribute to strengthen the solution.

[invo-coder19]

@Shrivadhu
Consider adding a default config file (e.g., YAML/JSON) for dataset/model paths instead of relying only on argparse. I thinks It will help a lot !

[invo-coder19]

@vignathi123-vi
Test in fresh setups and please post your insights for the same.

[Shrivadhu]

@invo-coder19
Great! I can work on adding a default YAML-based config system.
I'll open a PR shortly .

[vignathi123-vi]

@invo-coder19
That would be good!
Will open a PR shortly.

[Vikasboura]

Hi, I have opened a PR to address this issue.

PR: #169

Please let me know if any changes are required.

[likitha487]

Hi, I would like to work on this issue as part of my GSoC preparation.
Could you please assign it

[vignathi123-vi]

@invo-coder19
Hey, I tested this on a fresh setup.
Cloned the repo, created a brand new virtual environment with
nothing pre-installed, and tried running the scripts directly.

Here's what I found:

inference.py (lines 16-17) has hardcoded absolute paths:

• log_dir = "/home/kartik/git/DeepLense/..."
• finetune_model_path = "/home/kartik/git/DeepLense/.../Resnet_finetune_Model_II.pt"

Running this on any other machine immediately throws a
FileNotFoundError since these paths only exist on the
original developer's machine.

finetune_byol.py (line 21) same issue:

• saved_model_path = "/home/kartik/git/DeepLense/.../Resnet_finetune_Model_II_2023-07-23-13-30-24.pt"

finetune.py (line 14) same issue:

• saved_model_path = "/home/kartik/git/deepLense_transformer_ssl/output/pretrained_contrastive_embedding.pth"

Also checked main.py, main_byol.py, pretrain.py,
utils/dataset.py and the config/ folder — these all
look fine and handle paths correctly.

I'm thinking the fix would be pretty straightforward — just replacing
these hardcoded paths with argparse arguments so anyone
can pass their own paths when running the scripts.

[invo-coder19]

@vignathi123-vi appreciable work !