7.1. Neural Network Graph
- fhez.nn.nn.NN
alias of
fhez.nn.nn.NeuralNetwork
- class fhez.nn.nn.NeuralNetwork(graph=None)
Multi-Directed Neural Network Graph Handler.
This class handles traversing computational graphs toward some end-state, while computing forward(s), backward(s), and update(s) of the respective components described within.
- backward(gradient, current_node, end_node)
Traverse backwards until all nodes processed.
- backwards(gradients, current_node, end_node)
Calculate backward pass for multiple examples simultaneously.
- forward(x, current_node, end_node)
Traverse and activate nodes until all nodes processed.
- forwards(xs, current_node, end_node)
Calculate forward pass for multiple examples simultaneously.
- property g
Get computational graph.
- probe_shape(lst: list)
Get the shape of a list, assuming each sublist is the same length.
This function is recursive, sending the sublists down and terminating once a type error is thrown by the final point being a non-list
- update(current_node, end_node)
Update weights of all nodes using oldest single example gradient.
- updates(current_node, end_node)
Update the weights of all nodes by taking the average gradient.