Add fix to the full cache issue (#99)

Co-authored-by: Xinzhuo Jiang <[email protected]>

.gitignore

@@ -7,4 +7,5 @@ config.yml
 docs
 .idea/
 renv.lock
-extras/
+extras/
+.Renviron

R/Estimator.R

@@ -310,7 +310,8 @@ gridCvDeep <- function(mappedData,
 
   fitParams <- names(paramSearch[[1]])[grepl("^estimator", names(paramSearch[[1]]))]
   findLR <- modelSettings$estimatorSettings$findLR
-  for (gridId in trainCache$getLastGridSearchIndex():length(paramSearch)) {
+  if (!trainCache$isFull()) {
+    for (gridId in trainCache$getLastGridSearchIndex():length(paramSearch)) {
     ParallelLogger::logInfo(paste0("Running hyperparameter combination no ", gridId))
     ParallelLogger::logInfo(paste0("HyperParameters: "))
     ParallelLogger::logInfo(paste(names(paramSearch[[gridId]]), paramSearch[[gridId]], collapse = " | "))
@@ -385,7 +386,7 @@ gridCvDeep <- function(mappedData,
     ParallelLogger::logInfo(paste0("Caching all grid search results and prediction for best combination ", indexOfMax))
     trainCache$saveGridSearchPredictions(gridSearchPredictons)
   }
-
+  }
   paramGridSearch <- lapply(gridSearchPredictons, function(x) x$gridPerformance)
   # get best params
   indexOfMax <- which.max(unlist(lapply(gridSearchPredictons, function(x) x$gridPerformance$cvPerformance)))

R/TrainingCache-class.R

@@ -69,6 +69,13 @@ TrainingCache <- R6::R6Class(
       return(private$.paramPersistence$gridSearchPredictions)
     },
 
+    #' @description
+    #' Check if cache is full
+    #' @returns Boolen
+    isFull = function() {
+      return(all(unlist(lapply(private$.paramPersistence$gridSearchPredictions, function(x) !is.null(x$gridPerformance)))))
+    },    
+
     #' @description
     #' Gets the last index from the cached grid search
     #' @returns Last grid search index

extras/example.R

@@ -23,14 +23,14 @@ populationSet <- PatientLevelPrediction::createStudyPopulationSettings(
 #   epochs = 10L
 # ))
 
-modelSettings <- setDefaultResNet(estimatorSettings = setEstimator(
-  learningRate = "auto",
-  weightDecay = 1e-06,
-  device="cuda:0",
-  batchSize=128L,
-  epochs=50L,
-  seed=42
-))
+# modelSettings <- setDefaultResNet(estimatorSettings = setEstimator(
+#   learningRate = "auto",
+#   weightDecay = 1e-06,
+#   device="cuda:0",
+#   batchSize=128L,
+#   epochs=50L,
+#   seed=42
+# ))
 
 modelSettings <- setResNet(numLayers = c(1L, 2L),
                            sizeHidden = 72L,
@@ -45,7 +45,8 @@ modelSettings <- setResNet(numLayers = c(1L, 2L),
                              device = "cpu",
                              seed = 42
                            ),
-                           randomSample = 2)
+                           randomSample = 2,
+                           randomSampleSeed = 1)
 
 res2 <- PatientLevelPrediction::runPlp(
   plpData = plpData,
@@ -67,7 +68,7 @@ res2 <- PatientLevelPrediction::runPlp(
     runModelDevelopment = T,
     runCovariateSummary = F
   ),
-  saveDirectory = '~/test/resnet/'
+  saveDirectory = '~/deep_plp_test/resnet/'
 )