Error when calculating L2 distances

[MisaOgura]

Hello and thank you for making the code to run cell painting VAE available!

I've encountered a problem executing the notebook cell-painting/3.application/experiment_level5.ipynb and wanted to get your thoughts on whether the modification I applied is a correct one.

The function calculate_L2_distances defined in code block no.11 and executed in code block no.44 in throws below error:-

TypeError: Could not convert DNA inhibitornorepinephrine reputake inhibitormaternal embryonic leucine zipper kinase inhibitordopamine receptor antagonist|serotonin receptor agonistdopamine reuptake inhibitorDMSO to numeric

This is presumably due to the fact that the first line of the function is trying to take a mean over feature columns of a dataframe, but this dataframe contains non-feature columns with non-numeric values such as moa.

Below is my best guess at achieving the desired functionality:-

# BEFORE
...
def calculate_L2_distances(predictions, predictions_random):
    mean_of_moas = meta_moa_complete_df.groupby(['moa']).mean().loc[:,'Cells_AreaShape_FormFactor':]
    ...

# AFTER
...
def calculate_L2_distances(predictions, predictions_random):
    mean_of_moas = (
        meta_moa_complete_df
        .groupby(['moa'])[complete_features_df.columns]
        .mean()
        .loc[:,'Cells_AreaShape_FormFactor':]
    )
...

Retaining only the feature columns enables to calculate the mean over these columns. complete_features_df is defined the code block no.3 of the same notebook to be a dataframe that only includes feature columns:-

# code block #3
data_dict = load_data(["complete"])
meta_features = infer_cp_features(data_dict["complete"], metadata=True)
cp_features = infer_cp_features(data_dict["complete"])

complete_features_df = data_dict["complete"].reindex(cp_features, axis="columns")
complete_meta_df = data_dict["complete"].reindex(meta_features, axis="columns")

Look forward to your input.

Thanks so much,
Misa