Releases: StallaertLab/cc_mapping
V1.1.0
Random Forest Feature Selection - New Cutoff methods
Added new cutoff method for the Random Forest Feature Selection called "increment" as well as the older implementation which will be called, "jump" which can only be used when using the RF_min_max method
-
Increment: Function checks to see if the random forest accuracy increases by a threshold in certain number of additional features. If the accuracy does increase more than the threshold, the optimal number of features is increased by 1 and this check is done until this condition is not met
-
Jump: Function checks to see if the random forest accuracy increases by a threshold in certain number of additional features. If the accuracy increases more then the threshold, then the optimal number of features is increased to the number of features where the accuracy difference was surpassed. Then the function continues until the threshold is not surpassed
The jump method was the original method and led to large jumps in the number of features. For example, if you have the stable counter set to 20 and the threshold to 10 percent, then if the accuracy surpassed the threshold 16 features in, then the optimal number of features would be 16. Compare this to the increment method, it would slowly increase the optimal number of features until the accuracy does not surpass the threshold in the next 20 features.
An alternative would be to use jump with a low stable counter value to avoid such situations.
GMM Cell Cycle Prediction Revamp
GMM Cell Cycle phase prediction has been redone to make it more user friendly as well as updated some of the plots to be more informative.