Easy neural networks
Project description
Nait
Neural network module
Made to be easy and fast
Network - class
Syntax
Network()
train - method
Syntax
train(x, y, layer_size=4, layers=1, activation_function="linear", generate_network=True, learning_rate=0.01, optimizer="randomize", sample_size=10, epochs=100, exit_increase=0.1, backup=None)
Arguments
-
x
list
--- training inputs -
y
list
--- training outputs -
layer_size
integer
--- number of neurons in hidden layers -
layers
integer
--- if the time when logged should be displayed in the message -
activation_function
string
--- function applied to the output of each layer (linear / relu / step / sigmoid / leaky_relu) -
generate_network
boolean
--- if the program should generate a new network or try to train the existing one -
learning_rate
float
--- how fast the network learns (faster learning may result in lower precision) -
optimizer
string
--- which optimizer to use (randomize / backpropagate) - 'backpropagate' optimizer is highly unstable and needs normalized inputs -
sample_size
int
--- how many network samples in each batch - only for 'randomize' optimizer -
epochs
integer
--- number of epochs the network should train for -
exit_increase
float
--- exits if the loss gets _ over the lowest loss - only for 'backpropagate' optimizer -
backup
string
--- file to backup network while training (if left blank or set to None the network will not backup)
predict - method
Syntax
predict(input)
Arguments
- input
list
--- input
save - method
Syntax
save(file)
Arguments
- file
string
--- file to save network to
load - method
Syntax
load(file)
Arguments
- file
string
--- file to load network from
evaluate - method
Syntax
evaluate(x, y)
Arguments
-
x
list
--- testing inputs -
y
list
--- testing outputs
Nait v2.0.1 - Change Log
-
Complete re-write
-
Added backpropagate optimizer
-
Changed training display
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.