Fix invalid wokers numbers for classification (#213)

When multiprocessing is disabled, the flag for running the prediction in the main process was not updated.
Signed-off-by: Luca Foppiano <luca@foppiano.org>

Author: lfoppiano

delft/textClassification/models.py

@@ -250,8 +250,9 @@ def predict(self, predict_generator, use_main_thread_only=False):
         multiprocessing = True
 
         if use_main_thread_only:
-            # worker at 0 means the training will be executed in the main thread
-            nb_workers = 0
+            # tf_keras 2.17 requires workers >= 1; workers=1 with
+            # use_multiprocessing=False runs in the main thread.
+            nb_workers = 1
             multiprocessing = False
 
         y = self.model.predict(predict_generator, use_multiprocessing=multiprocessing, workers=nb_workers)