keras_char_cnn_kim: keras_char_cnn_kim

In systats/deeplyr: Pretrained keras models for predicting ideology from tweets.

Description

Character Level Convolutional Neural Network as described in Kim et al., 2015 (https://arxiv.org/abs/1508.06615) https://github.com/mhjabreel/CharCnn_Keras/blob/master/models/char_cnn_kim.py

Usage

keras_char_cnn_kim(input_dim, embed_dim = 128, seq_len = 50,
  filter_sizes = c(5, 4, 4), conv_layers = c(100, 60, 30),
  pool_size = 2, dens_layers = c(50, 30, 10), dropout = 0.2,
  output_dim = 1, output_fun = "sigmoid")

Arguments

input_dim	Number of unique vocabluary/tokens
embed_dim	Number of word vectors
seq_len	Length of the input sequences
conv_layers	number of filters by layer
pool_size	pooling dimension (filters)
dens_layers	Number of neurons by dense layer
dropout	a scalar between 0 and .5
output_dim	Number of neurons of the output layer
output_fun	Output activation function
filter_size	the window size (kernel_size)

Value

keras model

systats/deeplyr documentation built on Oct. 4, 2020, 7:59 p.m.