Added a generic model, where you only have to provide a loss and a mo…

…del function to the constructor.

models/generic_model.py

@@ -0,0 +1,27 @@
+from models.model import Model
+
+
+class GenericModel(Model):
+    def __init__(self, hyper_params_filepath, create_model_fn, create_loss_fn):
+        """
+        Create a generic model by passing in a create model function and a create loss function.
+        :param hyper_params_filepath: The hyperparameters filepath like for every other model.
+            This is then loaded into a hyperparams object.
+        :param create_model_fn: A function that gets in the input_tensor and if it should reuse weights
+            (and hyperparams object) and outputs the output tensors.
+            (outputs = create_model_fn(input_tensor, reuse_weights, hyperparams)
+        :param create_loss_fn: A function that gets in the output tensors of the create model the labels and the
+            validation labels (and hyperparams object). It must return the training operation, the loss operation and
+            the validation loss operation.
+            (train_op, loss_op, validation_loss_op = create_loss_fn(outputs, labels, validation_labels, hyperparams))
+        """
+        super(GenericModel, self).__init__(hyper_params_filepath)
+        self.__create_model = create_model_fn
+        self.__create_loss = create_loss_fn
+
+    def _create_model(self, input_tensor, reuse_weights):
+        self.outputs = self.__create_model(input_tensor, reuse_weights, self.hyper_params)
+
+    def _create_loss(self, labels, validation_labels=None):
+        train_op, loss_op, validation_loss_op = self.__create_loss(self.outputs, labels, validation_labels, self.hyper_params)
+        return train_op, loss_op, validation_loss_op