transferLearning: [Under development] Transfer learning

In hxia/plp-git-demo: Package for patient level prediction using data in the OMOP Common Data Model

Description

[Under development] Transfer learning

Usage

transferLearning(
  plpResult,
  plpData,
  population,
  fixLayers = T,
  includeTop = F,
  addLayers = c(100, 10),
  layerDropout = c(T, T),
  layerActivation = c("relu", "softmax"),
  outcomeWeight = 1,
  batchSize = 10000,
  epochs = 20
)

Arguments

plpResult	The plp result when training a kersa deep learning model on big data
plpData	The new data to fine tune the model on
population	The population for the new data
fixLayers	boolean specificying whether to fix weights in model being transferred
includeTop	If TRUE the final layer of the model being transferred is removed
addLayers	vector specifying nodes in each layer to add e.g. c(100,10) will add another layer with 100 nodels and then a final layer with 10
layerDropout	Add dropout to each new layer (binary vector length of addLayers)
layerActivation	Activation function for each new layer (string vector length of addLayers)
outcomeWeight	The weight to assign the class 1 when training the model
batchSize	Size of each batch for updating layers
epochs	Number of epoches to run

Examples

## Not run: 
modelSet <- setDeepNN()
plpResult <- runPlp(plpData, population, modelSettings = modelSet, ...)

transferLearning(...)

## End(Not run)

hxia/plp-git-demo documentation built on March 19, 2021, 1:54 a.m.