Neural Sequence Learning Using Tensorflow
The Neural Monkey package provides a higher level abstraction for sequential models, most prominently in Natural Language Processing (NLP), built on TensorFlow. It can be used for fast prototyping of sequential models in NLP which can be used e.g. for neural machine translation or sentence classification.
The higher-level API brings together a collection of standard building blocks (RNN encoder and decoder, multi-layer percetpron) and a simple way of adding new building blocks implemented directly in TensorFlow.
neuralmonkey-train <EXPERIMENT_INI>
bin
: Directory with executables and toolsexamples
: Example configuration files for ready-made experimentslib
: Third party softwareneuralmonkey
: Python package filestests
: Test files
The neuralmonkey
package is organized into subpackages as follows:
encoders
: The encoders (get sequences of data and encode them)decoders
: The decoders (get outputs of encoders and decode them)nn
: The NN (as in Neural Network components)trainers
: The trainers (train the whole thing)runners
: The runners (run the trained thing)readers
: The readers (read files)processors
: The pre- and postprocessors (pre- and postprocess data)evaluators
: The evaluators (used for measuring performance)config
: The configuration (loading, saving, parsing)logbook
: The Logbook (server for checking up on experiments)tests
: The unit tests
caffe_image_features.py
extracts features from images using pre-trained networkinstall_caffe.sh
installs caffe including prerequisitesprecompute_image_features.py
deprecatedreformat_downloaded_image_features.py
deprecatedtokenize_data.py
tokenizes datatokenize_persistable.sh
manages the tokenize_data script
- Install Tensorflow by following their installation docs here
- Clone this repository and follow the installation procedure outlined in
prepare_env.sh
. If the training crashes on an unknown dependency, just install it with pip.
- tflearn -- a more general and less abstract deep learning toolkit built over TensorFlow
- nlpnet -- deep learning tools for tagging and parsing
- NNBlocks -- a library build over Theano containing NLP specific models
The software is distributed under the BSD License.