removed all legacy CEHR-BERT code using tensorflow and keras

.github/workflows/tests.yml

@@ -23,6 +23,16 @@ jobs:
       uses: actions/setup-python@v3
       with:
         python-version: "3.10"
+
+    - name: Free up disk space
+      run: |
+        sudo rm -rf /usr/share/dotnet
+        sudo rm -rf /opt/ghc
+        sudo rm -rf "/usr/local/share/boost"
+        sudo rm -rf "$AGENT_TOOLSDIRECTORY"
+        sudo docker image prune --all --force
+        df -h
+
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip

constraints.txt

@@ -0,0 +1 @@
+xformers==0 # blocks installation

pyproject.toml

@@ -39,19 +39,15 @@ dependencies = [
     "numpy>=1.24.3",
     "packaging>=23.2",
     "pandas>=2.2.0",
-    "peft>=0.10.0",
+    "peft==0.10.0",
     "Pillow>=10.3.0",
     "pyarrow>=15.0.0",
     "python-dateutil==2.8.2",
     "PyYAML==6.0.1",
     "scikit-learn==1.4.0",
     "scipy==1.12.0",
-    "tensorflow>=2.15.0",
-    "keras==2.15.0",
-    "tensorflow-metal>=1.1.0; sys_platform == 'darwin'", # macOS only
-    "tensorflow-datasets>=4.5.2",
     "tqdm>=4.66.1",
-    "torch==2.4.0",
+    "torch>=2.4.0, <=2.8.0",
     "tokenizers>=0.19.0",
     "transformers>=4.41.0, <= 4.45.0",
     "accelerate>=0.31.0",

src/cehrbert/baseline_evaluation/evaluation.py

@@ -0,0 +1,44 @@
+import pandas as pd
+
+from .evaluation_parse_args import create_evaluation_args
+from .model_evaluators.frequency_model_evaluators import (
+    LogisticRegressionModelEvaluator,
+    XGBClassifierEvaluator,
+)
+
+
+def evaluate_baseline_models(args):
+    # Load the training data
+    dataset = pd.read_parquet(args.data_path)
+    test_person_ids = None
+    if args.patient_splits_folder:
+        patient_splits = pd.read_parquet(args.patient_splits_folder)
+        test_person_ids = patient_splits[patient_splits.split == "test"]
+
+    LogisticRegressionModelEvaluator(
+        dataset=dataset,
+        evaluation_folder=args.evaluation_folder,
+        num_of_folds=args.num_of_folds,
+        is_transfer_learning=args.is_transfer_learning,
+        training_percentage=args.training_percentage,
+        k_fold_test=args.k_fold_test,
+        test_person_ids=test_person_ids,
+    ).eval_model()
+
+    XGBClassifierEvaluator(
+        dataset=dataset,
+        evaluation_folder=args.evaluation_folder,
+        num_of_folds=args.num_of_folds,
+        is_transfer_learning=args.is_transfer_learning,
+        training_percentage=args.training_percentage,
+        k_fold_test=args.k_fold_test,
+        test_person_ids=test_person_ids,
+    ).eval_model()
+
+
+def main(args):
+    evaluate_baseline_models(args)
+
+
+if __name__ == "__main__":
+    main(create_evaluation_args().parse_args())

src/cehrbert/baseline_evaluation/evaluation_parse_args.py

@@ -0,0 +1,46 @@
+import argparse
+
+
+def create_evaluation_args():
+    main_parser = argparse.ArgumentParser(description="Arguments for evaluating the models")
+    main_parser.add_argument(
+        "-d",
+        "--data_path",
+        dest="data_path",
+        action="store",
+        help="The training data path",
+        required=True,
+    )
+    main_parser.add_argument(
+        "--patient_splits_folder",
+        dest="patient_splits_folder",
+        action="store",
+        help="The test person_ids data",
+        required=False,
+    )
+    main_parser.add_argument(
+        "-ef",
+        "--evaluation_folder",
+        dest="evaluation_folder",
+        action="store",
+        required=True,
+    )
+    main_parser.add_argument(
+        "-n",
+        "--num_of_folds",
+        dest="num_of_folds",
+        action="store",
+        required=False,
+        type=int,
+        default=4,
+    )
+    main_parser.add_argument("--is_transfer_learning", dest="is_transfer_learning", action="store_true")
+    main_parser.add_argument(
+        "--training_percentage",
+        dest="training_percentage",
+        required=False,
+        action="store",
+        type=float,
+        default=1.0,
+    )
+    return main_parser

src/cehrbert/evaluations/model_evaluators/frequency_model_evaluators.py → src/cehrbert/baseline_evaluation/model_evaluators/frequency_model_evaluators.py

@@ -10,7 +10,7 @@
 from tensorflow.keras.preprocessing.text import Tokenizer
 from xgboost import XGBClassifier
 
-from cehrbert.evaluations.model_evaluators.model_evaluators import AbstractModelEvaluator
+from .model_evaluators import AbstractModelEvaluator
 from cehrbert.utils.model_utils import compute_binary_metrics
 
 

src/cehrbert/evaluations/model_evaluators/model_evaluators.py → src/cehrbert/baseline_evaluation/model_evaluators/model_evaluators.py

@@ -1,43 +1,62 @@
 import copy
 import os
 import pathlib
-from abc import abstractmethod
+import logging
+from abc import ABC, abstractmethod
+from cehrbert.utils.model_utils import create_folder_if_not_exist
 
-import tensorflow as tf
 
-from cehrbert.trainers.model_trainer import AbstractModel
+class AbstractModel(ABC):
+    def __init__(self, *args, **kwargs):
+        super().__init__()
+        self._model = self._create_model(*args, **kwargs)
 
+    @abstractmethod
+    def _create_model(self, *args, **kwargs):
+        pass
+
+    @abstractmethod
+    def train_model(self, *args, **kwargs):
+        pass
+
+    @abstractmethod
+    def get_model_folder(self):
+        pass
+
+    def get_model_metrics_folder(self):
+        return create_folder_if_not_exist(self.get_model_folder(), "metrics")
+
+    def get_model_test_metrics_folder(self):
+        return create_folder_if_not_exist(self.get_model_folder(), "test_metrics")
+
+    def get_model_test_prediction_folder(self):
+        return create_folder_if_not_exist(self.get_model_folder(), "test_prediction")
 
-def get_metrics():
-    """
-    Standard metrics used for compiling the models.
+    def get_model_history_folder(self):
+        return create_folder_if_not_exist(self.get_model_folder(), "history")
 
-    :return:
-    """
+    @classmethod
+    def get_logger(cls):
+        return logging.getLogger(cls.__name__)
 
-    return [
-        "binary_accuracy",
-        tf.keras.metrics.Recall(name="recall"),
-        tf.keras.metrics.Precision(name="precision"),
-        tf.keras.metrics.AUC(curve="PR", name="pr_auc"),
-        tf.keras.metrics.AUC(name="auc"),
-    ]
+    def __str__(self):
+        return str(self.__class__.__name__)
 
 
 class AbstractModelEvaluator(AbstractModel):
     def __init__(
-        self,
-        dataset,
-        evaluation_folder,
-        num_of_folds,
-        is_transfer_learning: bool = False,
-        training_percentage: float = 1.0,
-        learning_rate: float = 1e-4,
-        is_chronological_test: bool = False,
-        k_fold_test: bool = False,
-        test_person_ids=None,
-        *args,
-        **kwargs,
+            self,
+            dataset,
+            evaluation_folder,
+            num_of_folds,
+            is_transfer_learning: bool = False,
+            training_percentage: float = 1.0,
+            learning_rate: float = 1e-4,
+            is_chronological_test: bool = False,
+            k_fold_test: bool = False,
+            test_person_ids=None,
+            *args,
+            **kwargs,
     ):
         self._dataset = copy.copy(dataset)
         self._evaluation_folder = evaluation_folder