brazilian_court_decisions / convert_to_hf_dataset.py
joelniklaus's picture
removed different configuration because of missing native huggingface support
31ca8e7
raw
history blame
6.7 kB
import os
import numpy as np
import pandas as pd
"""
Dataset url: https://github.com/lagefreitas/predicting-brazilian-court-decisions/blob/main/dataset.zip
Paper url: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9044329/
There are no splits available ==> Make random split ourselves
"""
pd.set_option('display.max_colwidth', None)
pd.set_option('display.max_columns', None)
def perform_original_preprocessing():
# Original Preprocessing from: https://github.com/lagefreitas/predicting-brazilian-court-decisions/blob/main/predicting-brazilian-court-decisions.py#L81
# Loading the labeled decisions
data = pd.read_csv("dataset.csv", sep='<=>', header=0)
print('data.shape=' + str(data.shape) + ' full data set')
# Removing NA values
data = data.dropna(subset=[data.columns[9]]) # decision_description
data = data.dropna(subset=[data.columns[11]]) # decision_label
print('data.shape=' + str(data.shape) + ' dropna')
# Removing duplicated samples
data = data.drop_duplicates(subset=[data.columns[1]]) # process_number
print('data.shape=' + str(data.shape) + ' removed duplicated samples by process_number')
data = data.drop_duplicates(subset=[data.columns[9]]) # decision_description
print('data.shape=' + str(data.shape) + ' removed duplicated samples by decision_description')
# Removing not relevant decision labels and decision not properly labeled
data = data.query('decision_label != "conflito-competencia"')
print('data.shape=' + str(data.shape) + ' removed decisions labeled as conflito-competencia')
data = data.query('decision_label != "prejudicada"')
print('data.shape=' + str(data.shape) + ' removed decisions labeled as prejudicada')
data = data.query('decision_label != "not-cognized"')
print('data.shape=' + str(data.shape) + ' removed decisions labeled as not-cognized')
data_no = data.query('decision_label == "no"')
print('data_no.shape=' + str(data_no.shape))
data_yes = data.query('decision_label == "yes"')
print('data_yes.shape=' + str(data_yes.shape))
data_partial = data.query('decision_label == "partial"')
print('data_partial.shape=' + str(data_partial.shape))
# Merging decisions whose labels are yes, no, and partial to build the final data set
data_merged = data_no.merge(data_yes, how='outer')
data = data_merged.merge(data_partial, how='outer')
print('data.shape=' + str(data.shape) + ' merged decisions whose labels are yes, no, and partial')
# Removing decision_description and decision_labels whose values are -1 and -2
indexNames = data[(data['decision_description'] == str(-1)) | (data['decision_description'] == str(-2)) | (
data['decision_label'] == str(-1)) | (data['decision_label'] == str(-2))].index
data.drop(indexNames, inplace=True)
print('data.shape=' + str(data.shape) + ' removed -1 and -2 decision descriptions and labels')
data.to_csv("dataset_processed_original.csv", index=False)
def perform_additional_processing():
df = pd.read_csv("dataset_processed_original.csv")
# remove strange " characters sometimes occurring in the beginning and at the end of a line
df.ementa_filepath = df.ementa_filepath.str.replace('^"', '')
df.decision_unanimity = df.decision_unanimity.str.replace('"$', '')
# removing process_type and judgment_date, since they are the same everywhere (-)
# decisions only contains 'None', nan and '-2'
# ementa_filepath refers to the name of file in the filesystem that we created when we scraped the data from the Court. It is temporary data and can be removed
# decision_description = ementa_text - decision_text - decision_unanimity_text
df = df.drop(['process_type', 'judgment_date', 'decisions', 'ementa_filepath'], axis=1)
# some rows are somehow not read correctly. With this, we can filter them
df = df[df.decision_text.str.len() > 1]
# rename "-2" to more descriptive name ==> -2 means, that they were not able to determine it
df.decision_unanimity = df.decision_unanimity.replace('-2', 'not_determined')
# rename cols for more clarity
df = df.rename(columns={"decision_unanimity": "unanimity_label"})
df = df.rename(columns={"decision_unanimity_text": "unanimity_text"})
df = df.rename(columns={"decision_text": "judgment_text"})
df = df.rename(columns={"decision_label": "judgment_label"})
df.to_csv("dataset_processed_additional.csv", index=False)
return df
perform_original_preprocessing()
df = perform_additional_processing()
# perform random split 80% train (3234), 10% validation (404), 10% test (405)
train, validation, test = np.split(df.sample(frac=1, random_state=42), [int(.8 * len(df)), int(.9 * len(df))])
def save_splits_to_jsonl(config_name):
# save to jsonl files for huggingface
if config_name: os.makedirs(config_name, exist_ok=True)
train.to_json(os.path.join(config_name, "train.jsonl"), lines=True, orient="records", force_ascii=False)
validation.to_json(os.path.join(config_name, "validation.jsonl"), lines=True, orient="records", force_ascii=False)
test.to_json(os.path.join(config_name, "test.jsonl"), lines=True, orient="records", force_ascii=False)
def print_split_table_single_label(train, validation, test, label_name):
train_counts = train[label_name].value_counts().to_frame().rename(columns={label_name: "train"})
validation_counts = validation[label_name].value_counts().to_frame().rename(columns={label_name: "validation"})
test_counts = test[label_name].value_counts().to_frame().rename(columns={label_name: "test"})
table = train_counts.join(validation_counts)
table = table.join(test_counts)
table[label_name] = table.index
total_row = {label_name: "total",
"train": len(train.index),
"validation": len(validation.index),
"test": len(test.index)}
table = table.append(total_row, ignore_index=True)
table = table[[label_name, "train", "validation", "test"]] # reorder columns
print(table.to_markdown(index=False))
save_splits_to_jsonl("")
print_split_table_single_label(train, validation, test, "judgment_label")
print_split_table_single_label(train, validation, test, "unanimity_label")
# create second config by filtering out rows with unanimity label == not_determined, while keeping the same splits
# train = train[train.unanimity_label != "not_determined"]
# validation = validation[validation.unanimity_label != "not_determined"]
# test = test[test.unanimity_label != "not_determined"]
# it is a very small dataset and very imbalanced (only very few not-unanimity labels)
# save_splits_to_jsonl("unanimity")