Reimplmenting Keras directly via reticulate

FBerding · Jul 9, 2023 · c5870df · c5870df
1 parent 5ab2580
commit c5870df
Show file tree

Hide file tree

Showing 25 changed files with 59,268 additions and 79,599 deletions.
diff --git a/.Rhistory b/.Rhistory
diff --git a/.github/workflows/R-CMD-check.yaml b/.github/workflows/R-CMD-check.yaml
@@ -71,7 +71,7 @@ jobs:
           #)
 
           conda_install(
-            packages = "tensorflow<2.11",
+            packages = "tensorflow",
             envname = envname,
             conda = "auto",
             pip = TRUE

diff --git a/DESCRIPTION b/DESCRIPTION
@@ -47,7 +47,6 @@ RoxygenNote: 7.2.3
 Encoding: UTF-8
 Depends:
     quanteda,
-    keras,
     foreach,
     R (>= 2.10)
 Suggests: 

diff --git a/NAMESPACE b/NAMESPACE
@@ -29,7 +29,6 @@ import(Rcpp)
 import(bundle)
 import(doParallel)
 import(foreach)
-import(keras)
 import(quanteda)
 import(reshape2)
 import(reticulate)

diff --git a/R/install_and_config.R b/R/install_and_config.R
@@ -20,7 +20,7 @@ install_py_modules<-function(envname="aifeducation"){
   )
 
   reticulate::conda_install(
-    packages = c("tensorflow<2.11"),
+    packages = c("tensorflow"),
     envname = envname,
     conda = "auto",
     pip = TRUE

diff --git a/R/te_classifier_neuralnet_model.R b/R/te_classifier_neuralnet_model.R
diff --git a/R/transformer_bert.R b/R/transformer_bert.R
@@ -328,7 +328,7 @@ train_tune_bert_model=function(output_dir,
     dir.create(paste0(output_dir,"/checkpoints"))
   }
   callback_checkpoint=tf$keras$callbacks$ModelCheckpoint(
-    filepath = paste0(output_dir,"/checkpoints/best_weights.cpkt"),
+    filepath = paste0(output_dir,"/checkpoints/best_weights.h5"),
     monitor="val_loss",
     verbose=1L,
     mode="auto",
@@ -352,7 +352,7 @@ train_tune_bert_model=function(output_dir,
                 callbacks=list(callback_checkpoint))
 
   cat(paste(date(),"Load Weights From Best Checkpoint"))
-  mlm_model$load_weights(paste0(output_dir,"/checkpoints/best_weights.cpkt"))
+  mlm_model$load_weights(paste0(output_dir,"/checkpoints/best_weights.h5"))
 
   cat(paste(date(),"Saving Bert Model"))
   mlm_model$save_pretrained(save_directory=output_dir)

diff --git a/R/transformer_longformer.R b/R/transformer_longformer.R
@@ -276,7 +276,7 @@ train_tune_longformer_model=function(output_dir,
   }
 
   callback_checkpoint=tf$keras$callbacks$ModelCheckpoint(
-    filepath = paste0(output_dir,"/checkpoints/best_weights.cpkt"),
+    filepath = paste0(output_dir,"/checkpoints/best_weights.h5"),
     monitor="val_loss",
     verbose=1L,
     mode="auto",
@@ -299,7 +299,7 @@ train_tune_longformer_model=function(output_dir,
                 callbacks=list(callback_checkpoint))
 
   cat(paste(date(),"Load Weights From Best Checkpoint"))
-  mlm_model$load_weights(paste0(output_dir,"/checkpoints/best_weights.cpkt"))
+  mlm_model$load_weights(paste0(output_dir,"/checkpoints/best_weights.h5"))
 
   cat(paste(date(),"Saving Longformer Model"))
   mlm_model$save_pretrained(save_directory=output_dir)

diff --git a/R/transformer_roberta.R b/R/transformer_roberta.R
@@ -281,7 +281,7 @@ train_tune_roberta_model=function(output_dir,
   }
 
   callback_checkpoint=tf$keras$callbacks$ModelCheckpoint(
-    filepath = paste0(output_dir,"/checkpoints/best_weights.cpkt"),
+    filepath = paste0(output_dir,"/checkpoints/best_weights.h5t"),
     monitor="val_loss",
     verbose=1L,
     mode="auto",
@@ -304,7 +304,7 @@ train_tune_roberta_model=function(output_dir,
                 callbacks=list(callback_checkpoint))
 
   print(paste(date(),"Load Weights From Best Checkpoint"))
-  mlm_model$load_weights(paste0(output_dir,"/checkpoints/best_weights.cpkt"))
+  mlm_model$load_weights(paste0(output_dir,"/checkpoints/best_weights.h5"))
 
   print(paste(date(),"Saving RoBERTa Model"))
   mlm_model$save_pretrained(save_directory=output_dir)

diff --git a/docs/articles/classification_tasks.html b/docs/articles/classification_tasks.html
diff --git a/docs/deps/bootstrap-5.2.2/bootstrap.min.css b/docs/deps/bootstrap-5.2.2/bootstrap.min.css
diff --git a/docs/index.html b/docs/index.html
diff --git a/docs/pkgdown.yml b/docs/pkgdown.yml
@@ -5,5 +5,5 @@ articles:
   aifeducation: aifeducation.html
   classification_tasks: classification_tasks.html
   sharing_and_publishing: sharing_and_publishing.html
-last_built: 2023-07-07T07:43Z
+last_built: 2023-07-09T19:11Z
 
diff --git a/docs/reference/TextEmbeddingClassifierNeuralNet.html b/docs/reference/TextEmbeddingClassifierNeuralNet.html
diff --git a/docs/search.json b/docs/search.json
diff --git a/man/TextEmbeddingClassifierNeuralNet.Rd b/man/TextEmbeddingClassifierNeuralNet.Rd
diff --git a/tests/testthat/test-classifier_neural_net.R b/tests/testthat/test-classifier_neural_net.R
@@ -156,7 +156,6 @@ test_that("training_baseline_only", {
       batch_size=32,
       dir_checkpoint=testthat::test_path("test_data/tmp"),
       trace=FALSE,
-      view_metrics=FALSE,
       keras_trace=0,
       n_cores=1)
   )
@@ -189,7 +188,6 @@ test_that("training_bsc_only", {
       batch_size=32,
       dir_checkpoint=testthat::test_path("test_data/tmp"),
       trace=FALSE,
-      view_metrics=FALSE,
       keras_trace=0,
       n_cores=1)
   )
@@ -222,7 +220,6 @@ test_that("training_pbl_baseline", {
       batch_size=32,
       dir_checkpoint=testthat::test_path("test_data/tmp"),
       trace=FALSE,
-      view_metrics=FALSE,
       keras_trace=0,
       n_cores=1)
   )
@@ -255,7 +252,6 @@ test_that("training_pbl_bsc", {
       batch_size=32,
       dir_checkpoint=testthat::test_path("test_data/tmp"),
       trace=FALSE,
-      view_metrics=FALSE,
       keras_trace=0,
       n_cores=1)
   )

diff --git a/tests/testthat/test_data/bert/config.json b/tests/testthat/test_data/bert/config.json
@@ -1,7 +1,6 @@
 {
-  "_name_or_path": "test_data/bert",
   "architectures": [
-    "BertForMaskedLM"
+    "BertModel"
   ],
   "attention_probs_dropout_prob": 0.1,
   "classifier_dropout": null,
@@ -21,5 +20,5 @@
   "transformers_version": "4.30.2",
   "type_vocab_size": 2,
   "use_cache": true,
-  "vocab_size": 29729
+  "vocab_size": 30522
 }