aritraghsh09 / gampen Goto Github PK

Trying to download the packages necessary through r-requirements.txt for autocropping.ipynb in Google Colab runs into issues downloading PANDAS on my machine. When the programs attempts to download PANDAS, it pauses for several minutes and then throws the following error:

`error: subprocess-exited-with-error

× pip subprocess to install build dependencies did not run successfully.
│ exit code: 1
╰─> See above for output.

note: This error originates from a subprocess, and is likely not a problem with pip.
Installing build dependencies ... error
error: subprocess-exited-with-error

× pip subprocess to install build dependencies did not run successfully.
│ exit code: 1
╰─> See above for output.

note: This error originates from a subprocess, and is likely not a problem with pip.`

In the GaMPEN Getting Started documentation (https://gampen.readthedocs.io/en/latest/Getting_Started.html), it is suggested to run the trainer with the following code:

python ggt/train/train.py \ --experiment_name='demo' \ --data_dir='./../hsc/' \ --split_slug='balanced-dev2' \ --batch_size=16 \ --epochs=2 \ --lr=5e-7 \ --momentum=0.99 \ --crop \ --cutout_size=239 \ --target_metrics='custom_logit_bt,log_R_e,log_total_flux' \ --repeat_dims \ --no-nesterov \ --label_scaling='std' \ --dropout_rate=0.0004 \ --loss='aleatoric_cov' \ --weight_decay=0.0001 \ --parallel

Upon investigation of the target_metrics in the info.csv file within the hsc directory, the correct target_metrics should be custom_logit_bt,ln_R_e_asec,ln_total_flux_adus rather than custom_logit_bt,log_R_e,log_total_flux.

In /GaMPEN/ggt/docs/conf.py, the Sphinx mathjax extension is specified; but math equations are not being rendered properly on RTD.

For example, math equations on this page https://gampen.readthedocs.io/en/latest/Using_GaMPEN.html#running-the-trainer

1. Dockerize GaMPEN
2. Test on AWS
3. Write unit tests (if possible)

Right now, requirements.txt is pretty bloated, especially with old requirements that no longer are a primary feature of GaMPEN (e.g., the group convolutional neural networks.)

Do a thorough code review and get rid of both old features and their corresponding dependencies in requirements.txt

Aritra Ghosh created the following fix to the source-code of autocrop.py to allow it to run on machines without a GPU!

Code: model.load_state_dict(torch.load(model_path,map_location=torch.device(device)))