Keras NALU (Neural Arithmetic Logic Units)

Keras implementation of a NALU layer (Neural Arithmetic Logic Units). See: https://arxiv.org/pdf/1808.00508.pdf.

Installation

pip install keras-nalu

Usage

from keras.layers import Input
from keras.models import Model
from keras.optimizers import RMSprop
from keras_nalu.nalu import NALU

# Your dataset
X_test = ... # Interpolation data
Y_test = ... # Interpolation data

X_validation = ... # Extrapolation data (validation)
Y_validation = ... # Extrapolation data (validation)

X_test = ... # Extrapolation data (test)
Y_test = ... # Extrapolation data (test)

# Hyper parameters
epoch_count = 1000
learning_rate = 0.05
sequence_len = 100

inputs = Input(shape=(sequence_len, ))
hidden = NALU(units=2)(inputs)
hidden = NALU(units=2)(hidden)
outputs = NALU(units=1)(hidden)

model = Model(inputs=inputs, outputs=outputs)
model.summary()
model.compile(loss='mse', optimizer=RMSprop(lr=learning_rate))

model.fit(
    batch_size=256,
    epochs=epoch_count,
    validation_data=(X_validation, Y_validation),
    x=X_train,
    y=Y_train,
)

extrapolation_loss = model.evaluate(
    batch_size=256,
    x=X_test,
    y=Y_test,
)

Options

cell

Cell to use in the NALU layer. May be 'a' (addition/subtraction), 'm' (multiplication/division/power), or None which, will apply a gating function to toggle between 'a' or 'm'.

• Default: None
• Type: ?('a' | 'm' | None)

e

Epsilon value added to inputs in order to prevent calculating the log of zero.

• Default: 1e-7
• Type: ?float