Add model architecture diagrams

WSU-SEAL · Feb 26, 2022 · e75868b · e75868b
1 parent 1486cb7
commit e75868b
Show file tree

Hide file tree

Showing 23 changed files with 20 additions and 9 deletions.
diff --git a/DNNModels.py b/DNNModels.py
@@ -79,11 +79,12 @@ def get_coefs(word, *arr): return word, np.asarray(arr, dtype='float32')
 class DNNModel:
     def __init__(self, X_train=None, Y_train=None, algo="CNN", embedding="fasttext",
                  max_features=5000, maxlen=500,
-                 embedding_size=300, load_from_file=None):
+                 embedding_size=300, load_from_file=None, plot_file_name =None):
         self.max_features = max_features
         self.maxlen = maxlen
         self.embed_size = embedding_size
         self.embedding_type = embedding
+        self.plot=plot_file_name
 
 
         session_conf = tf.compat.v1.ConfigProto(intra_op_parallelism_threads=4, inter_op_parallelism_threads=4)
@@ -347,9 +348,8 @@ def _train(self, X_train, Y_train):
         X_train_vector = self._prepare_df(X_train)
 
         if model is not None:
-            #print("Type is:")
-            #print(type(X_train_vector), type(Y_train))
-
+            if self.plot is not None:
+                tf.keras.utils.plot_model(model, to_file=self.plot, show_shapes=True)
             ##convert Y_train to np.ndaaray
             Y_train = Y_train.values
 

diff --git a/ToxiCR.py b/ToxiCR.py
@@ -151,17 +151,22 @@ def load_pretrained_model(self, filename):
 
     def get_model(self, X_train, Y_train, tuning=False):
         ALGO = self.ALGO
+        plot_file_name = "./architecture/" + ALGO + ".png"
         if (ALGO == "RF") | (ALGO == "GBT") | (ALGO == "SVM") | (ALGO == "DT") | (ALGO == "LR"):
             self.classifier_model = CLEModel(X_train=X_train, Y_train=Y_train, algo=self.ALGO, tuning=tuning)
         elif ALGO == "BERT":
             from TransformerModel import TransformerModel
-            self.classifier_model = TransformerModel(X_train=X_train, Y_train=Y_train)
+            self.classifier_model = TransformerModel(X_train=X_train, Y_train=Y_train, plot_file_name=plot_file_name)
         elif (ALGO == "CNN") | (ALGO == "LSTM") | (ALGO == "GRU") | (ALGO == "biLSTM") :
             import DNNModels
+
             self.classifier_model = DNNModels.DNNModel(X_train=X_train,
                                                        Y_train=Y_train,
-                                                       algo=ALGO,
+                                                       algo=ALGO, plot_file_name=plot_file_name,
                                                        embedding=self.embedding)
+        else:
+            print("Unknown algorithm: "+ALGO)
+            exit(1)
 
         return self.classifier_model
 
@@ -257,7 +262,7 @@ def ten_fold_cross_validation(toxicClassifier, rand_state):
     args = parser.parse_args()
 
     print(args)
-    ALGO = args.algo
+    ALGO = str(args.algo).upper()
     REPEAT = args.repeat
     embedding = args.embed
     mode = args.mode

diff --git a/TransformerModel.py b/TransformerModel.py
@@ -38,7 +38,8 @@ def df_to_dataset(dataframe, batch_size=16):
 
 class TransformerModel:
     def __init__(self, X_train=None, Y_train=None,
-                 bert_model_name="bert_en_uncased_L-12_H-768_A-12", load_from_file=None):
+                 bert_model_name="bert_en_uncased_L-12_H-768_A-12", load_from_file=None,
+                 plot_file_name=None):
         self.tfhub_handle_encoder = BertLocator.getBERTEncoderURL(bert_model_name)
         self.tfhub_handle_preprocess = BertLocator.getPreprocessURL(bert_model_name)
 
@@ -48,6 +49,7 @@ def __init__(self, X_train=None, Y_train=None,
         self.bert_preprocess_model = hub.KerasLayer(self.tfhub_handle_preprocess)
         self.bert_model = hub.KerasLayer(self.tfhub_handle_encoder)
         self.epochs = 20
+        self.plot=plot_file_name
 
         if load_from_file is not None:
             self.steps_per_epoch = 19571  # size of our dataset
@@ -77,6 +79,8 @@ def build_classifier_model(self):
         classifier_model.compile(optimizer=optimizer,
                                  loss=loss,
                                  metrics=metrics)
+        if self.plot is not None:
+            tf.keras.utils.plot_model(classifier_model, to_file=self.plot, show_shapes=True)
         return classifier_model
 
     def get_optimizer(self):

diff --git a/...s/comparisions/Cross-validations-CNN.xlsx → ...ns/comparisons/Cross-validations-CNN.xlsx b/...s/comparisions/Cross-validations-CNN.xlsx → ...ns/comparisons/Cross-validations-CNN.xlsx
diff --git a/...ns/comparisions/Cross-validations-DT.xlsx → ...ons/comparisons/Cross-validations-DT.xlsx b/...ns/comparisions/Cross-validations-DT.xlsx → ...ons/comparisons/Cross-validations-DT.xlsx
diff --git a/...s/comparisions/Cross-validations-GBC.xlsx → ...ns/comparisons/Cross-validations-GBC.xlsx b/...s/comparisions/Cross-validations-GBC.xlsx → ...ns/comparisons/Cross-validations-GBC.xlsx
diff --git a/...s/comparisions/Cross-validations-GRU.xlsx → ...ns/comparisons/Cross-validations-GRU.xlsx b/...s/comparisions/Cross-validations-GRU.xlsx → ...ns/comparisons/Cross-validations-GRU.xlsx
diff --git a/...ns/comparisions/Cross-validations-LR.xlsx → ...ons/comparisons/Cross-validations-LR.xlsx b/...ns/comparisions/Cross-validations-LR.xlsx → ...ons/comparisons/Cross-validations-LR.xlsx
diff --git a/.../comparisions/Cross-validations-LSTM.xlsx → ...s/comparisons/Cross-validations-LSTM.xlsx b/.../comparisions/Cross-validations-LSTM.xlsx → ...s/comparisons/Cross-validations-LSTM.xlsx
diff --git a/...ns/comparisions/Cross-validations-RF.xlsx → ...ons/comparisons/Cross-validations-RF.xlsx b/...ns/comparisions/Cross-validations-RF.xlsx → ...ons/comparisons/Cross-validations-RF.xlsx
diff --git a/...s/comparisions/Cross-validations-SVM.xlsx → ...ns/comparisons/Cross-validations-SVM.xlsx b/...s/comparisions/Cross-validations-SVM.xlsx → ...ns/comparisons/Cross-validations-SVM.xlsx
diff --git a/...omparisions/Cross-validations-biLSTM.xlsx → ...comparisons/Cross-validations-biLSTM.xlsx b/...omparisions/Cross-validations-biLSTM.xlsx → ...comparisons/Cross-validations-biLSTM.xlsx
diff --git a/.../comparisions/cross-validations-BERT.xlsx → ...s/comparisons/cross-validations-BERT.xlsx b/.../comparisions/cross-validations-BERT.xlsx → ...s/comparisons/cross-validations-BERT.xlsx
diff --git a/...omparisions/embedding_comparison-CNN.xlsx → ...comparisons/embedding_comparison-CNN.xlsx b/...omparisions/embedding_comparison-CNN.xlsx → ...comparisons/embedding_comparison-CNN.xlsx
diff --git a/...omparisions/embedding_comparison-GRU.xlsx → ...comparisons/embedding_comparison-GRU.xlsx b/...omparisions/embedding_comparison-GRU.xlsx → ...comparisons/embedding_comparison-GRU.xlsx
diff --git a/...mparisions/embedding_comparison-LSTM.xlsx → ...omparisons/embedding_comparison-LSTM.xlsx b/...mparisions/embedding_comparison-LSTM.xlsx → ...omparisons/embedding_comparison-LSTM.xlsx
diff --git a/...risions/embedding_comparison-biiLSTM.xlsx → ...arisons/embedding_comparison-biiLSTM.xlsx b/...risions/embedding_comparison-biiLSTM.xlsx → ...arisons/embedding_comparison-biiLSTM.xlsx
diff --git a/model-architecture/BERT.png b/model-architecture/BERT.png
diff --git a/model-architecture/CNN.png b/model-architecture/CNN.png
diff --git a/model-architecture/GRU.png b/model-architecture/GRU.png
diff --git a/model-architecture/LSTM.png b/model-architecture/LSTM.png
diff --git a/model-architecture/biLSTM.png b/model-architecture/biLSTM.png
diff --git a/requirements.txt b/requirements.txt
@@ -9,4 +9,6 @@ tensorflow-text == 2.5.0
 tensorflow-hub>=0.12.0
 nltk>=3.5
 spacy>=2.3
-openpyxl>=3.0.9
+openpyxl>=3.0.9
+pydot
+graphviz