Closed siboehm closed 2 years ago
Class imbalance should be incorporated into the BCE loss. Example on Trapnell:
> adata.obs["condition"].value_counts()[:10] control 6464 GSK-LSD1 1868 BRD4770 1868 Baricitinib 1862 Entacapone 1853 RG108 1852 WP1066 1851 Curcumin 1850 Capecitabine 1849 Mesna 1847 Name: condition, dtype: int64 > adata.obs["condition"].value_counts()[-10:] Rigosertib 984 Luminespib 980 Tozasertib 975 Mocetinostat 949 Alvespimycin 930 AT9283 910 Patupilone 757 Flavopiridol 693 Epothilone 583 YM155 394 Name: condition, dtype: int64
This is not super important, as the dataset is not actually very imbalanced. But it will make the adversarial loss more meaningful.
Class imbalance should be incorporated into the BCE loss. Example on Trapnell:
This is not super important, as the dataset is not actually very imbalanced. But it will make the adversarial loss more meaningful.