arenets

AREnets

This is an AREkit network project contributional module.

Train

run_train.py

Using the embedded tensorflow-based models. The related AREkit module provides a list of the supported models, dedicated for the sentiment relation extraction (ModelNames enum type). Model training process, based on the SentiNEREL could be launched as follows:

NOTE: considering a root project dir for this script

from arekit.contrib.networks.enum_name_types import ModelNames
from framework.arenets.train import train_nn

train_nn(output_dir="_out/serialize-nn",
         model_log_dir="_model",
         model_name=ModelNames.AttEndsCNN,
         split_filepath="data/split_fixed.txt")

The latter produces the model at _out/serialize_nn with logging information at _model dir, and data split based on the data/split_fixed.txt file.

Predict

run_predict.py

Belowe is an example which illustrates on how CNN model might be adopted as follows:

from arekit.contrib.networks.enum_name_types import ModelNames
from arekit.common.experiment.data_type import DataType
from framework.arenets.predict import predict_nn

predict_nn(model_name=ModelNames.CNN, 
           output_dir="_out/serialize-nn", data_type=DataType.Test,
           embedding_dir="_out/serialize-nn", samples_dir="_out/serialize-nn")

The pretrained state will be automaticaly searched, and the latter provided via model_io at predict_nn implementation.