Added multi GPU support (basic version)

database_interface.py

@@ -138,7 +138,7 @@ def _check_if_database_exists(self):
         This method checks if the database defined by self.db_dir and self.db_name exists. If not,
         an exception is raise.
         """
-        db_path = self.db_dir + "\\" + self.db_name  # create full path to database
+        db_path = os.path.join(self.db_dir, self.db_name)  # create full path to database
 
         # check if database exists
         if not os.path.exists(db_path):  # database doesn't exists: raise exception
@@ -222,7 +222,7 @@ def _open_db(self):
         This methods opens the database defined by self.db_dir and self.db_name by creating a
         database connection and a cursor.
         """
-        db_path = self.db_dir + "\\" + self.db_name  # create full path to the database
+        db_path = os.path.join(self.db_dir, self.db_name)  # create full path to the database
 
         # create database connection and a cursor
         self.db_connection = sqlite3.connect(db_path)

logging_config.py

@@ -1,21 +1,25 @@
 """Configurations for logging."""
 
 import logging
+import os
 
 
-
-def init_logging():
+def init_logging(db_dir, db_name):
     """Initializes logging.
 
     Configures logging. Error messages are logged to the 'error.log' file. Info messages are logged
     to the console. The results are save in a 'result_' log file.
+
+    Args:
+        db_dir -- directory of the database, used to create file for results
+        db_name -- name of the database, used to create file name for results
     """
     # create logger for traditional-SA project
     logger = logging.getLogger('RNN-SA')
     logger.setLevel(logging.INFO)
 
     # create file handler which logs error messages
-    log_file_handler = logging.FileHandler('error.log', mode='w+')
+    log_file_handler = logging.FileHandler(os.path.join(db_dir, 'error.log'), mode='w+')
     log_file_handler.setLevel(logging.ERROR)
 
     # create console handler with a lower log level (e.g debug or info)

main.py

@@ -9,6 +9,9 @@
 from random import shuffle  # for shuffle of the task-sets
 
 import keras
+import tensorflow as tf
+import matplotlib
+matplotlib.use('agg')
 import matplotlib.pyplot as plt
 import numpy as np
 import sklearn
@@ -18,6 +21,7 @@
 import logging_config
 import ml_models
 import params
+import os
 
 # default indices of all task attributes (column indices of 'Task')
 DEFAULT_FEATURES = ['Task_ID', 'Priority', 'Deadline', 'Quota', 'CAPS', 'PKG', 'Arg', 'CORES',
@@ -39,10 +43,10 @@
 def main():
     """Main function of project 'RNN-SA'."""
     # determine database directory and name
-    db_dir, db_name = "..\\Datenbanken\\", "panda_v2.db"
+    db_dir, db_name = os.getcwd(), "panda_v3.db"
 
     # create and initialize logger
-    logger = logging_config.init_logging()
+    logger = logging_config.init_logging(db_dir, db_name)
 
     # load the data
     logger.info("Loading and pre-processing data from the database...")
@@ -58,16 +62,38 @@ def main():
     logger.info("Doing hyperparameter exploration...")
     start_time = time.time()
     #h = hyperparameter_exploration(data=data, name='LSTM', num='0')
-    out, model = ml_models.LSTM_model(data['train_X'], data['train_y'], data['val_X'],
-                                      data['val_y'], params.hparams)
+    # out, model = ml_models.LSTM_model(data['train_X'], data['train_y'], data['val_X'],
+    #                                   data['val_y'], params.hparams)
     end_time = time.time()
     logger.info("Finished hyperparameter exploration!")
     logger.info("Best result: ")
     logger.info("Time elapsed: %f s \n", end_time - start_time)
 
+    # multi GPU support
+    model = ml_models._build_LSTM_model(params.hparams, params.config)
+    try:
+        parallel_model = keras.utils.multi_gpu_model(model, cpu_relocation=True)
+        print("Training using multiple GPUs...")
+    except ValueError:
+        parallel_model = model
+        print("Training using single GPU or CPU...")
+    parallel_model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
+    out = parallel_model.fit(
+        x=data['train_X'],
+        y=data['train_y'],
+        batch_size=params['batch_size'],
+        epochs=params['num_epochs'],
+        verbose=params.config['verbose_training'],
+        callbacks=ml_models._init_callbacks(params.hparams, params.config),
+        validation_data=[data['val_X'], data['val_y']],
+        shuffle=True,
+    )
+
+
+
     # evaluate
     loss, acc = model.evaluate(data['test_X'], data['test_y'], batch_size=params.hparams['batch_size'])
-    print("Loss: %f --- Accuracy: %f", loss, acc)
+    logger.info("Loss: %f --- Accuracy: %f", loss, acc)
 
 
 def hyperparameter_exploration(data, name, num):