Hiercon

Requirements

We will take Ubuntu for example.

• python 3.6

$ sudo apt-get install 3.6

• other python packages

$ pip install -r requirements.txt

Pre-requisites

Please use install Autophrase by

$ git clone https://github.com/shangjingbo1226/AutoPhrase.git

And follows the instruction from there. Additionally, configure inside Hiercon by setting the AUTOPHRASE_PATH to be the autophrase installation path.

Training and testing

Our model works in a weakily supervised setting, where given a single text file with each row representing one specific document, along with training labels for a few rows, it predicts the label for all the documents.

Input Format

• The input text files as specified by {your prefix name here}_merged_tokenized, and the prefix name is specified in ./run.sh at
• The training documents is specified by a {your prefix name here}_merged_tokenized_training_inds_HANsFile.bin, by pickle.dump() your python list containing the row index of the training documents.
• the training labels is specified by {your prefix name here}_merged_tokenized_superspan_HANs_labels.txt, and each row contains a label for the corresponding row in the input text files. (only the rows in the training indexes are used.)

Output Format

The output prediction is specified by {your prefix name here}_merged_tokenized_prediction_result.txt. Each row contains a label for the corresponding row in the input text files.

Hyper-Parameters

The list of parameters, their default values, and a short description is attached below

parser.add_argument("--batch_size", type=int, default=16)
parser.add_argument("--num_epoches", type=int, default=5)
parser.add_argument("--log_interval", type=int, default=5)
parser.add_argument("--lr", type=float, default=0.0001)
parser.add_argument("--momentum", type=float, default=0.9)
parser.add_argument("--word_feature_size", type=int, default=4)
parser.add_argument("--sent_feature_size", type=int, default=3)
parser.add_argument("--num_bins", type=int, default=10)
parser.add_argument("--es_min_delta", type=float, default=0.0,
                    help="Early stopping's parameter: minimum change loss to qualify as an improvement")
parser.add_argument("--es_patience", type=int, default=5,
                    help="Early stopping's parameter: number of epochs with no improvement after which training will be stopped. Set to 0 to disable this technique.")
parser.add_argument("--test_interval", type=int, default=1,
                    help="Number of epoches between testing phases")
parser.add_argument("--log_path", type=str, default="tensorboard/han_voc")

Sample data

A set of sample data containing all the intermediate results in order to run the prediction is available at Google drive