v0.3.6

给RNN中加入了dropout, L1-L2正则化

Author: yuki-961004

R/document_control.R

@@ -301,10 +301,39 @@
 #'    \item \code{units [int]}
 #'
 #'          The number of neurons (or units) in the Recurrent Layer
-#'          (GRU or LSTM). Conceptually, this parameter represents the memory
-#'          capacity and complexity of the network; it dictates how much
+#'          (RNN, GRU or LSTM). Conceptually, this parameter represents the 
+#'          memory capacity and complexity of the network; it dictates how much
 #'          information about the sequential trials the model can store and
 #'          process.
+#'          
+#'    \item \code{dropout [double]}
+#'    
+#'          Dropout is a powerful regularization technique used to prevent 
+#'          overfitting in RNNs. During each training iteration, a predefined 
+#'          percentage of neurons (units) are randomly "dropped" or deactivated 
+#'          by setting their activations to zero.
+#' 
+#'    \item \code{L [int]}
+#'
+#'          This parameter determines the type of regularization applied to the
+#'          log-likelihood to penalize model complexity, which helps prevent
+#'          overfitting. The default is \code{0}.
+#'          \itemize{
+#'            \item \code{L = 0}: No regularization.
+#'            \item \code{L = 1}: L1 regularization (Lasso), which adds a
+#'                  penalty proportional to the sum of the absolute values of 
+#'                  the free parameters.
+#'            \item \code{L = 2}: L2 regularization (Ridge), which adds a
+#'                  penalty proportional to the sum of the squared values of 
+#'                  the free parameters.
+#'          }
+#'
+#'    \item \code{penalty [double]}
+#'
+#'          This parameter specifies the strength of the regularization, acting
+#'          as a multiplier for the penalty term defined by \code{L}. A larger
+#'          value imposes a stronger penalty on the free parameters. The
+#'          default value is \code{1e-5}.
 #'
 #'    \item \code{batch_size [int]}
 #'
@@ -363,6 +392,9 @@
 #'    loss = "MSE",
 #'    info = c(colnames$object, colnames$action),
 #'    units = 128,
+#'    dropout = 0
+#'    L = 0,
+#'    penalty = 1e-5,
 #'    batch_size = 10,
 #'    epochs = 100,
 #'    check = TRUE

R/engine_RNN.R

@@ -197,7 +197,7 @@ engine_RNN <- function(
   } else if (loss == "MAE") {
     units_out <- n_params
     loss_func <- "mean_absolute_error"
-  } else if (loss == "Huber") {
+  } else if (loss == "HBR") {
     units_out <- n_params
     loss_func <- "huber_loss"
   } else {
@@ -216,55 +216,79 @@ engine_RNN <- function(
         object = RNN,
         units = units,
         input_shape = c(n_trials, n_info),
-        return_sequences = FALSE
       )
     },
     "GRU" = {
       RNN <- keras::layer_gru(
         object = RNN,
         units = units,
         input_shape = c(n_trials, n_info),
-        return_sequences = FALSE,
       )
     },
     "LSTM" = {
       RNN <- keras::layer_lstm(
         object = RNN,
         units = units,
         input_shape = c(n_trials, n_info),
-        return_sequences = FALSE,
       )
     },
     "BiRNN" = {
       RNN <- keras::bidirectional(
         object = RNN,
-        layer = keras::layer_simple_rnn(
-          units = units,
-          return_sequences = FALSE
-        ),
+        layer = keras::layer_simple_rnn(units = units),
         input_shape = c(n_trials, n_info)
       )
     },
     "BiGRU" = {
       RNN <- keras::bidirectional(
         object = RNN,
-        layer = keras::layer_gru(units = units, return_sequences = FALSE),
+          layer = keras::layer_gru(units = units),
         input_shape = c(n_trials, n_info)
       )
     },
     "BiLSTM" = {
       RNN <- keras::bidirectional(
         object = RNN,
-        layer = keras::layer_lstm(units = units, return_sequences = FALSE),
+          layer = keras::layer_lstm(units = units),
         input_shape = c(n_trials, n_info)
       )
     },
-  ) |>
-    # Hidden Layer
-    keras::layer_dense(
-      units = units / 2,
-      activation = "relu"
-    ) |>
+  )
+
+  # Hidden Layer
+  switch(
+    EXPR = as.character(L),
+    "0" = {
+      RNN <- keras::layer_dense(
+        object = RNN,
+        units = units / 2,
+        activation = "relu",
+        kernel_initializer = keras::initializer_he_normal()
+      )
+    },
+    "1" = {
+      RNN <- keras::layer_dense(
+        object = RNN,
+        units = units / 2,
+        activation = "relu",
+        kernel_initializer = keras::initializer_he_normal(),
+        kernel_regularizer = keras::regularizer_l1(penalty)
+      )
+    },
+    "2" = {
+      RNN <- keras::layer_dense(
+        object = RNN,
+        units = units / 2,
+        activation = "relu",
+        kernel_initializer = keras::initializer_he_normal(),
+        kernel_regularizer = keras::regularizer_l2(penalty)
+      )
+    }
+  )
+
+  RNN <- RNN |>
+    # Dropout Layer
+    keras::layer_dropout(rate = dropout) |>
     # Output Layer
     keras::layer_dense(
       units = units_out,

R/estimate_2_RNN.R

@@ -113,6 +113,9 @@ estimate_2_RNN <- function(
     loss = "MSE",
     info = c(colnames$object, colnames$action),
     units = 128,
+    dropout = 0,
+    L = 0,
+    penalty = 1e-5,
     batch_size = 10,
     epochs = 100,
     check = TRUE

R/step_2_rcv_d.R

@@ -204,6 +204,9 @@ rcv_d <- function(
     loss = "MSE",
     info = c(colnames$object, colnames$action),
     units = 128,
+    dropout = 0,
+    L = 0,
+    penalty = 1e-5,
     batch_size = 10,
     epochs = 100,
     check = TRUE

R/step_3_fit_p.R

@@ -167,6 +167,9 @@ fit_p <- function(
     loss = "MSE",
     info = c(colnames$object, colnames$action),
     units = 128,
+    dropout = 0,
+    L = 0,
+    penalty = 1e-5,
     batch_size = 10,
     epochs = 100,
     check = TRUE

R/zzz.R

@@ -15,5 +15,6 @@ utils::globalVariables(c(
   # ABC
   "tol", "reduction", "ncomp", "metric",
   # RNN,
-  "info", "layer", "loss", "units", "batch_size", "epochs", "check"
+  "layer", "loss", "info", "units", "dropout", "L", "penalty", 
+  "batch_size", "epochs", "check"
 ))

dev/CODE/3_Estimate_SBI.Rmd

@@ -245,9 +245,9 @@ recovery.RNN <- multiRL::rcv_d(
   lowers = list(c(0, 0)),
   uppers = list(c(1, 10)),
   control = list(
-    core = 10, sample = 100, train = 100,
+    core = 10, sample = 100, train = 1000,
     #tol = 0.5, reduction = "PCA", 
-    layer = "BiLSTM", loss = "MDN", check = FALSE
+    layer = "GRU", loss = "MSE", check = FALSE, dropout = 0, L = 2
   )
 )
 ```

docs/pkgdown.yml

@@ -8,7 +8,7 @@ articles:
   Step_2_rcv_d: Step_2_rcv_d.html
   Step_3_fit_p: Step_3_fit_p.html
   Step_4_rpl_e: Step_4_rpl_e.html
-last_built: 2026-03-28T09:04Z
+last_built: 2026-03-28T16:19Z
 urls:
   reference: https://yuki-961004.github.io/multiRL/reference
   article: https://yuki-961004.github.io/multiRL/articles