make_origin_dataset.py

from collections import Counter
from typing import List

import datasets
import matplotlib.pyplot as plt
import pandas as pd
from constants import (event_centered_2_descriptions,
                       physical_entity_2_descriptions, relations_map,
                       social_intercation_2_descriptions)


def show_bar(relation: List):
    c = dict(Counter(relation))

    keys = list(c.keys())
    values = list(c.values())
    plt.bar(keys, values)
    plt.xticks(rotation=25, fontsize=8)
    plt.yticks(fontsize=8)

    plt.xlabel('relations')
    plt.ylabel('numbers')
    plt.title('relations analysis')

    for i in range(len(keys)):
        plt.text(i, values[i] + 10, str(values[i]), ha='center', fontsize=10)

    plt.show()

def read_file(data_path: str):
    df = pd.read_csv(data_path, sep='\t', header=None)

    df.columns = ['event', 'relation', 'tail']
    print(df.head())

    event = df['event'].tolist()
    relation = df['relation'].tolist()
    tail = df['tail'].tolist()

    return event, relation, tail

def make_base_dataset(event: List[str], relation: List[str], tail: List[str]):
    new_event, new_relation, new_tail = [], [], []
    knowledge_type = []
    relation_description = []

    prev_event, prev_relation, prev_tail = None, None, None

    for i in range(len(event)):
        if i > 0 and event[i] == prev_event and relation[i] == prev_relation:
            new_tail[-1].extend( [tail[i]] )
        else:
            new_event.append(event[i])
            new_relation.append(relation[i])
            # insert knowledge type
            relation_list = []
            for r in list(relations_map.values()):
                relation_list.extend(r)
            if relation[i] not in relation_list:
                raise ValueError(f'dont find match knowledge type named {relation[i]}, please check it!')
            
            for k, v in relations_map.items():
                if relation[i] in v:
                    knowledge_type.append(k)
                    if k == 'social_intercation':
                        relation_description.append(social_intercation_2_descriptions[relation[i]])
                    elif k == 'physical_entity':
                        relation_description.append(physical_entity_2_descriptions[relation[i]])
                    elif k == 'event_centered':
                        relation_description.append(event_centered_2_descriptions[relation[i]])
                    else:
                        raise KeyError(f"dont find match relation type named {relation[i]} in dict, please check it!")
                    
            new_tail.append( [tail[i]] )
        prev_event, prev_relation, prev_tail = event[i], relation[i], tail[i]
    
    df = pd.DataFrame({
        'knowledge_type': knowledge_type,
        'event': new_event,
        'relation': new_relation,
        'relation_description': relation_description,
        'tail': new_tail,
    })

    print(df.head())
    return df

def get_dataset(data_path: str):
    event, relation, tail = read_file(data_path=data_path)
    df = make_base_dataset(event=event, relation=relation, tail=tail)
    dataset = datasets.Dataset.from_pandas(df, split='train')

    print(dataset)
    return dataset

def upload_dataset(dataset, repo_id :str, access_token: str, private: bool):
    dataset.push_to_hub(
        repo_id = repo_id,
        private = private,
        token   = access_token,
    )

if __name__ == '__main__':
    train_dataset = get_dataset('./dataset/train.tsv')
    valid_dataset = get_dataset('./dataset/dev.tsv')
    test_dataset = get_dataset('./dataset/test.tsv')
    
    dataset = datasets.DatasetDict({
        'train': train_dataset,
        'validation': valid_dataset,
        'test': test_dataset
    })

    print(dataset)

    upload_dataset(dataset, repo_id='', private=False, access_token='')