Implementation of kim2014convolutional
Project description
kim2014convolutional
This package provides a simple implementation of the models proposed in the paper:
Kim, Y. (2014). Convolutional neural networks for sentence classification. arXiv preprint arXiv:1408.5882.
Installation
This package depends on the Keras library. This means you will need to install a backend library in order to use this module. Take a look to Keras installation to get more information.
After having installed the backend of yout choice, you just need to install this package using pip:
pip install kim2014convolutional
Usage
This package only provides a single model. To get detailed information on the parameters the model accepts, take a look to the documentation included with the module class.
Here is a complete example of instantiation of the CNN-multichannel
model proposed in the original paper using two channel of randomly
initialized word embeddings:
import numpy as np
import numpy.random as rng
vocabulary_size = 10000
embedding_size = 300
value = np.sqrt(6/embedding_size)
weights_shape = (vocabulary_size+1, embedding_size)
weights = rng.uniform(low=-value, high=value, size=weights_shape)
channels = [
{
'weights': [weights],
'trainable': False,
'input_dim': vocabulary_size + 1,
'output_dim': embedding_size,
'name': 'random-embedding-1'
},
{
'weights': [weights],
'trainable': True,
'input_dim': vocabulary_size + 1,
'output_dim': embedding_size,
'name': 'random-embedding-2'
}
]
windows = [
{
'filters': 100,
'kernel_size': 3,
'activation': 'relu',
'name': '3-grams'
},
{
'filters': 100,
'kernel_size': 4,
'activation': 'relu',
'name': '4-grams'
},
{
'filters': 100,
'kernel_size': 5,
'activation': 'relu',
'name': '5-grams'
}
]
from kim2014convolutional import Model
model = Model(channels=channels,
windows=windows,
sentence_length=37,
num_classes=6,
dropout_rate=0.5,
maxnorm_value=3,
classifier_activation='softmax',
include_top=True,
name='CNN-multichannel')
model.summary()
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.
Source Distribution
Built Distribution
Hashes for kim2014convolutional-0.1.0.tar.gz
Algorithm | Hash digest | |
---|---|---|
SHA256 | 34a56f9f6c36b36553b240eab44dec907a6ea8e7aed6c02c7266c077e37f86e1 |
|
MD5 | fd6fadf740b0f43861c47c77b79f1a57 |
|
BLAKE2b-256 | 7fd24dd8325c6cdd193e012af99dc9377aaaeaca593bc16bda272c48df0995a6 |
Hashes for kim2014convolutional-0.1.0-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 107a52f24c50f9e2f31b9b2065ac93e8d956971f96869e6371ae33afb51b138f |
|
MD5 | 9d104b8286d93cecaa2148e29b527bf6 |
|
BLAKE2b-256 | 52197e4787f48905f06128b95ac1e92a0e4712c6b3b6cfd105642bdded3c4f52 |