WaveGrad is a tiny Python library to create and train deep feedforward neural netwroks. Documentation
It's possible to create different networks with different topology. You just need to import the Sequential
module and add Layers
into it:
from wavegrad.network import Sequential
from wavegrad.layers import LayerDense
from wavegrad.activations import *
from wavegrad.losses import MSE
from wavegrad.optimizers import *
# network
net = Sequential()
net.add(LayerDense(17, 5, sigmoid))
# You can always add more layers
net.add(LayerDense(5, 1, tanh))
# You can alway add optimizer and loss function
optim = GD(net.layers, lr=0.01, momentum=0.9)
net.use(MSE)
# train
net.fit(Xtrain, ytrain, epochs=500, optimizer=optim)
See activations
for the list of activation functions.
See optimizers
for the list of optimizers.
See losses
for the list of losees functions.
- implement the validation split from yourself without use the library sklearn
- Refactor the optimizator