documentation for pytorch.py

urnai/models/memory_representations/neural_network/pytorch.py

@@ -6,6 +6,43 @@
 import numpy as np
 
 class PyTorchDeepNeuralNetwork(ABNeuralNetwork):
+    """
+    Implementation of a Generic Deep Neural Network using PyTorch
+
+    This class inherits from ABNeuralNetwork, so it already has all abstract methods
+    necessary for learning, predicting outputs, and building a model. All that this
+    class does is implement those abstract methods, and implement the methods necessary
+    for adding Neural Network layers, such as add_input_layer(), add_output_layer(),
+    add_fully_connected_layer().
+
+    Differently from KerasDeepNeuralNetwork, this class is not able to dynamically build
+    Neural Network architectures that use convolutional layers, due to the complexity 
+    of PyTorch's initialization of convolutional layers and a general difficulty in 
+    fitting that complexity to URNAI's achitecture. To use PyTorch with convolutional 
+    layers, one has to inherit from this class and manually create your model, we recommend
+    you do that in self.make_model().
+
+    This class also implements the methodes necessary for saving and loading the model
+    from local memory.
+
+    Parameters:
+        action_output_size: int
+            size of our output
+        state_input_shape: tuple
+            shape of our input
+        build_model: Python dict
+            A dict representing the NN's layers. Can be generated by the 
+            ModelBuilder.get_model_layout() method from an instantiated ModelBuilder object.
+        gamma: Float
+            Gamma parameter for the Deep Q Learning algorithm
+        alpha: Float
+            This is the Learning Rate of the model
+        seed: Integer (default None)
+            Value to assing to random number generators in Python and our ML libraries to try 
+            and create reproducible experiments
+        batch_size: Integer
+            Size of our learning batch to be passed to the Machine Learning library
+    """
 
     def __init__(self, action_output_size, state_input_shape, build_model, gamma, alpha, seed = None, batch_size=32):