A simple example for how to build your own model using AllenNLP as a dependency. An explanation of all of the code in this repository is given in the part 1 and part 2 of the AllenNLP tutorial.
There are two main pieces of code you need to write in order to make a new model: a
DatasetReader
and a Model
. In this repository, we constructed a DatasetReader
for reading
academic papers formatted as a JSON lines file (you can see an example of the data in
tests/fixtures/s2_papers.jsonl
). We then constructed a model
to classify the papers given some label (which we specified as the paper's venue in the
DatasetReader
). Finally, we added a script to use AllenNLP's training commands from a
third-party repository, and an experiment configuration for running a real model on real data.
To train this model, after setting up your development environment by running
pip install -r requirements.txt
, you run:
allennlp train experiments/venue_classifier.json -s /tmp/your_output_dir_here --include-package my_library
This example was written by the AllenNLP team. You can see a similar example repository written by others here.