In linxihui/lazyML: Automatic machine learning algorithms and hyper-parameters selection

More features

• About mpTune:

  • track on start time, running time, memory usage, warning and error messages in the tuning loop and summary procedure.
  • trace
  • retained predictions, and used for ensemble

• survival support and test (remember that some prediciton gives risk instead of survival time!)

  • survival support and test (DONE!)
  • need more survival models (coxph, RandomForestSRC, cforest, earth, mboost, ...)
  • insistant prediction(some are survival times some are risk score): which is mostly the case? if both, how to solve this issue? (CURRENTLY just reverse the result if CI < 0.5, corr < 0)
• test on regression
• a script for SGE support, run in background and wait to collect results
• a fit function to automatically fit the best models
• (DONE) CV or bootstrap for predictability estimate (support 2 levels of parallel)
• print fitting progress? Add logging: start time of the whole fiting, running time/memory of each model
• resample: trace
• requireMetric: partial match, built-in individule metric visible

Individual model

• gbm: verbose disable by default
• glmnet2: fix problem, check if a level has only two unique value
• rf: mtry range

feature interface

```r mpTune.default( x, y, weights, modelList = list(RF = 'rf', balancedRF = 'rf', 'gbm'), modelControl = list( balancedRF = list(sampsize = quote(rep(min(table(y)),2))), gbm = list(verbose = TRUE)), preProcessControl = list(FUN = PCA, nPC = c(2,3,4), ...), tuningControl = list( tuneGrid = list(rf = ....), FUN = makeTuneGrid, gridLength = 5, randomizedLength = 20, # other parameter for makTuneGrid ...), samplingControl = list( FUN = createCVFolds, args = ..., trainIndex = NULL, testIndex = NULL, returnSample = TRUE), performanceControl = list( FUN = defaultSummary, args = ..., metric = NA, targetMetric = 'AUC'), parallelControl = list(backend = 'doMC', batchSize = 10), classProbs = TRUE, survivalPrediction = 'risk', returnData = TRUE, returnPrediction = TRUE, seeds = NA ); mpTune.data(data, ...)

save all these controls as mpTuneControl

a flow like interface

lazyML(data) %>% preProcess() %>% sampling() %>% modelBlender(preProcess(), modelList, modelControl) %>% tuning(, performance()) %>% bestFit %>% run();

```

linxihui/lazyML documentation built on May 21, 2019, 6:39 a.m.