tutorial_inference_Norman.ipynb is out-of-date

[jzqin]

When trying to run the tutorial_inference_Norman.ipynb file, loading the pretrained model generates the following error:

TypeError                                 Traceback (most recent call last)
Cell In[111], line 6
      1 gears_model = GEARS(pert_data, 
      2                     device = 'cpu',
      3                     weight_bias_track = False,
      4                     proj_name = 'gears',
      5                     exp_name = 'test')
----> 6 gears_model.load_pretrained('../model/model_ckpt')

File ~/miniconda3/envs/gears/lib/python3.10/site-packages/gears/gears.py:257, in GEARS.load_pretrained(self, path)
    254     config = pickle.load(f)
    256 del config['device'], config['num_genes'], config['num_perts']
--> 257 self.model_initialize(**config)
    258 self.config = config
    260 state_dict = torch.load(os.path.join(path, 'model.pt'), map_location = torch.device('cpu'))

TypeError: GEARS.model_initialize() got an unexpected keyword argument 'cell_fitness_pred'

This is because the model.py file was updated with the commit 8da5e36, which removed the cell_fitness_pred argument, but the model checkpoint provided via
dataverse_download('https://dataverse.harvard.edu/api/access/datafile/6979956', './model.zip') does not use the updated model.

[yhr91]

Sorry for the dealyed response! I've updated the tutorial files and this should now work out of the box.

[KyleFerchen]

I am still getting this error as of 2024/10/07:

>>> gears_model.load_pretrained('./model_ckpt')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/users/fero3l/env/gears/lib/python3.10/site-packages/gears/gears.py", line 257, in load_pretrained
    self.model_initialize(**config)
TypeError: GEARS.model_initialize() got an unexpected keyword argument 'cell_fitness_pred'

[ManuelMoradiellos]

Hi! I want to join in this conversation as well, because I was trying to replicate Fig. 4 and Fig. 5b from the paper as I wanted to create the predictions' metrics heatmap for my own data, and I got the exact same error when loading the Norman pretrained model.

I will try with another dataset but maybe it's something related to the Norman one.
Thanks in advance for the package and help provided~

[jzqin]

@ManuelMoradiellos the GEARS_misc package uses a different (older) version of GEARS compared to the main repo. I suggest cloning the GEARS_misc repo and using it by adding the directory to your local path, through something like

sys.path.append('/path/to/GEARS_misc/')
from gears import PertData, GEARS

[ManuelMoradiellos]

@jzqin thanks for your comment and for you help!

Even though I was able to install all packages needed to run the GEARS_misc repo, but when I use PertData() with the Norman dataset from the 6979957 I later have some problem initializing the pretrained model obtained form 6979956, there's size mismatch in the information loaded.

Are those the correct link to run the same code or were they updated in some regard?

Bests

Edit: I leave my mistake here, as I just noticed that it was clearly stated in the notebooks/code that in order to replicate the results we first need to train the model and then run the predictions, there's no pretrained model for the GEARS_misc code. Sorry for my näive comment