Merge pull request #150 from david-thrower/148-tensorflow-upgrades

Author: sashakolpakov

.github/workflows/automerge.yml

@@ -5,20 +5,22 @@ name: Python application
 
 on:
   push:
-    branches: [ "main", "125-add-use-cases-gpt2-encoder-for-phishing-email-detection" ]
+    branches: [ "main", "148-tensorflow-upgrades" ]
 
 permissions:
   contents: read
 
 jobs:
   build:
     runs-on: ubuntu-latest
+    # container:
+    #   image: python:33.11.9
     steps:
     - uses: actions/checkout@v3
     - name: Test Cerebros
       uses: actions/setup-python@v3
       with:
-        python-version: "3.10.4"
+        python-version: "3.11"
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip

cerebros/simplecerebrosrandomsearch/simple_cerebros_random_search.py

@@ -502,7 +502,7 @@ def run_moity_permutations(self, spec, subtrial_number, lock):
         with open(neural_network_spec_file, 'w') as f:
             f.write(str(spec))
         next_model_name =\
-            f"{self.project_name}/models/tr_{str(self.trial_number).zfill(16)}_subtrial_{str(subtrial_number).zfill(16)}"\
+            f"{self.project_name}/models/tr_{str(self.trial_number).zfill(16)}_subtrial_{str(subtrial_number).zfill(16)}.keras"\
             .lower()
         neural_network.save(next_model_name)
         oracle_0['trial_number'] = self.trial_number

cerebros/units/units.py

@@ -83,8 +83,13 @@ def __init__(self,
                  train_data_dtype=tf.float32,
                  *args,
                  **kwargs):
-
-        self.input_shape = input_shape
+        if isinstance(input_shape, int):
+            self.input_shape = (input_shape,)
+        elif isinstance(input_shape, str):
+            self.input_shape = (int(input_shape),)
+        else:
+            _input_shape = [int(ax) for ax in input_shape]
+            self.input_shape = tuple(_input_shape)
         self.neural_network_layer = []
         self.base_models = base_models
         self.train_data_dtype = train_data_dtype

cicd-requirements.txt

@@ -1,4 +1,5 @@
-matplotlib==3.6.2
-tensorflow-text==2.12.0
-keras-nlp==0.6.2
-scikit-learn==1.3.2
+matplotlib==3.8.4
+tensorflow-text==2.15.0
+keras-nlp==0.9.1
+scikit-learn==1.4.1.post1
+tensorflow-hub==0.16.1

regression-example-ames-no-preproc-val-set.py

@@ -122,26 +122,26 @@ def hash_based_split(df,  # Pandas dataframe
 print(f"Shape of train data: {train_data_np.shape}")
 
 tensor_x =\
-    tf.constant(train_df.values)
+    tf.constant(train_df.values, dtype=tf.float32)
 
 
 training_x = [tensor_x]
 
 INPUT_SHAPES = [training_x[i].shape[1] for i in np.arange(len(training_x))]
 
-train_labels = [train_labels_pd.values]
+train_labels = [tf.constant(train_labels_pd.values.astype(float), dtype=tf.float32)]
 print(f"Shape of train labels: {train_labels_pd.shape}")
 
 OUTPUT_SHAPES = [1]  # [train_labels[i].shape[1]
 
 ## Val set:
 
 print(f"Shape of val data: {val_df.shape}")
-val_tensor_x = tf.constant(val_df.values)
+val_tensor_x = tf.constant(val_df.values, dtype=tf.float32)
 val_x = [val_tensor_x]
 
 
-val_labels = [val_labels_pd.values]
+val_labels = [tf.constant(val_labels_pd.values.astype(float), dtype=tf.float32)]
 print(f"Shape of val labels: {val_labels_pd.shape}")
 
 # Params for a training function (Approximately the oprma

regression-example-ames-no-preproc.py

@@ -43,7 +43,7 @@
 
 INPUT_SHAPES = [training_x[i].shape[1] for i in np.arange(len(training_x))]
 
-train_labels = [label.values]
+train_labels = [tf.constant(label.values.astype(float))]
 
 OUTPUT_SHAPES = [1]  # [train_labels[i].shape[1]
 

requirements.txt

@@ -1,11 +1,10 @@
-jax==0.4.1
-jaxlib==0.4.1
-pendulum==2.1.2
-tensorflow==2.12.0
-numpy==1.23.5
-pandas==2.0.3
-pyvis==0.3.1
-plotly==5.11.0
-matplotlib==3.6.2
-tensorflow-text==2.12.0
-imageio==2.25.0
+jax==0.4.26
+jaxlib==0.4.26
+pendulum==3.0.0
+tensorflow==2.15.0
+numpy==1.26.4
+pandas==2.2.1
+pyvis==0.3.2
+plotly==5.20.0
+matplotlib==3.8.4
+imageio==2.34.0

text-class-ham-or-spam.py

@@ -45,17 +45,52 @@
 
 
 # Build BERT base model
+# text_input = tf.keras.layers.Input(shape=(), dtype=tf.string)
+# preprocessor = hub.KerasLayer(
+#     "https://tfhub.dev/tensorflow/bert_en_uncased_preprocess/3")
+# encoder_inputs = preprocessor(text_input)
+
+###
+# preprocessor = hub.load(
+#     "https://www.kaggle.com/models/tensorflow/bert/TensorFlow2/en-uncased-preprocess/3")
+# inp = tf.keras.layers.Input(shape=(), dtype=tf.string)
+# text_inputs = [inp]
+# tokenize = hub.KerasLayer(preprocessor.tokenize)
+# tokenized_inputs = [tokenize(segment) for segment in text_inputs]
+
+# seq_length = 128  # Your choice here.
+# bert_pack_inputs = hub.KerasLayer(
+#     preprocessor.bert_pack_inputs,
+#     arguments=dict(seq_length=seq_length))  # Optional argument.
+# encoder_inputs = bert_pack_inputs(tokenized_inputs)
+# ###
+
+
+
+# encoder = hub.KerasLayer(
+#     "https://tfhub.dev/tensorflow/bert_en_uncased_L-12_H-768_A-12/4",
+#     trainable=True)
+# outputs = encoder(encoder_inputs)
+# pooled_output = outputs["pooled_output"]      # [batch_size, 768].
+# sequence_output = outputs["sequence_output"]  # [batch_size, seq_length, 768].
+# embedding_model = tf.keras.Model(inp, pooled_output)
+
+
+###
 text_input = tf.keras.layers.Input(shape=(), dtype=tf.string)
 preprocessor = hub.KerasLayer(
-    "https://tfhub.dev/tensorflow/bert_en_uncased_preprocess/3")
+    "https://kaggle.com/models/tensorflow/bert/TensorFlow2/en-uncased-preprocess/3")
 encoder_inputs = preprocessor(text_input)
 encoder = hub.KerasLayer(
-    "https://tfhub.dev/tensorflow/bert_en_uncased_L-12_H-768_A-12/4",
+    "https://www.kaggle.com/models/tensorflow/bert/TensorFlow2/bert-en-uncased-l-10-h-128-a-2/2",
     trainable=True)
 outputs = encoder(encoder_inputs)
-pooled_output = outputs["pooled_output"]      # [batch_size, 768].
-sequence_output = outputs["sequence_output"]  # [batch_size, seq_length, 768].
+pooled_output = outputs["pooled_output"]      # [batch_size, 128].
+sequence_output = outputs["sequence_output"]  # [batch_size, seq_length, 128].
+
 embedding_model = tf.keras.Model(text_input, pooled_output)
+###
+
 
 ## Load the Data set
 raw_text = pd.read_csv(data_file, dtype='object')
@@ -68,7 +103,7 @@
 labels = labels.values
 data = raw_text.values
 
-labels_tensor = tf.constant(labels, dtype=tf.int8)
+labels_tensor = tf.constant(labels, dtype=tf.float32)
 data_tensor = tf.constant(data, dtype=tf.string)
 
 TIME = pendulum.now(tz='America/New_York').__str__()[:16]\