Create dataset_processor.py
Browse files
my_model/dataset/dataset_processor.py
ADDED
@@ -0,0 +1,176 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
from collections import Counter
|
3 |
+
import contractions
|
4 |
+
import csv
|
5 |
+
from typing import Tuple, List, Optional
|
6 |
+
from my_model.config import dataset_config as config
|
7 |
+
|
8 |
+
class OKVQADatasetProcessor:
|
9 |
+
"""
|
10 |
+
Processes the OKVQA dataset by loading, processing, and merging question and annotation data.
|
11 |
+
|
12 |
+
Attributes:
|
13 |
+
questions_file_path (str): Path to the questions JSON file.
|
14 |
+
annotations_file_path (str): Path to the annotations JSON file.
|
15 |
+
questions (List[dict]): Extracted list of question entries from the JSON file.
|
16 |
+
annotations (List[dict]): Extracted list of annotation entries from the JSON file.
|
17 |
+
df_questions (DataFrame): DataFrame holding the questions.
|
18 |
+
df_answers (DataFrame): DataFrame holding the annotations.
|
19 |
+
merged_df (Optional[DataFrame]): DataFrame resulting from merging questions and answers, initialized as None.
|
20 |
+
"""
|
21 |
+
|
22 |
+
def __init__(self, questions_file_path: str, annotations_file_path: str) -> None:
|
23 |
+
"""
|
24 |
+
Initializes the dataset processor with file paths and loads the data into DataFrames.
|
25 |
+
|
26 |
+
Parameters:
|
27 |
+
questions_file_path (str): The file path for the questions JSON file.
|
28 |
+
annotations_file_path (str): The file path for the annotations JSON file.
|
29 |
+
"""
|
30 |
+
|
31 |
+
self.questions_file_path = questions_file_path
|
32 |
+
self.annotations_file_path = annotations_file_path
|
33 |
+
self.questions, self.annotations = self.load_data_files()
|
34 |
+
self.df_questions = pd.DataFrame(self.questions)
|
35 |
+
self.df_answers = pd.DataFrame(self.annotations)
|
36 |
+
self.merged_df = None
|
37 |
+
|
38 |
+
def load_data_files(self) -> Tuple[List[dict], List[dict]]:
|
39 |
+
"""
|
40 |
+
Loads the question and annotation data from JSON files.
|
41 |
+
|
42 |
+
Returns:
|
43 |
+
Tuple[List[dict], List[dict]]: A tuple containing lists of questions and annotations.
|
44 |
+
"""
|
45 |
+
with open(self.questions_file_path, 'r') as file:
|
46 |
+
data = json.load(file)
|
47 |
+
questions = data['questions']
|
48 |
+
|
49 |
+
with open(self.annotations_file_path, 'r') as file:
|
50 |
+
data = json.load(file)
|
51 |
+
annotations = data['annotations']
|
52 |
+
|
53 |
+
return questions, annotations
|
54 |
+
|
55 |
+
@staticmethod
|
56 |
+
def find_most_frequent(my_list: List[str]) -> Optional[str]:
|
57 |
+
"""
|
58 |
+
Determines the most frequent item in a list.
|
59 |
+
|
60 |
+
Parameters:
|
61 |
+
my_list (List[str]): The list from which to find the most frequent item.
|
62 |
+
|
63 |
+
Returns:
|
64 |
+
Optional[str]: The most frequent item or None if the list is empty.
|
65 |
+
"""
|
66 |
+
if not my_list:
|
67 |
+
return None
|
68 |
+
counter = Counter(my_list)
|
69 |
+
most_common = counter.most_common(1)
|
70 |
+
return most_common[0][0]
|
71 |
+
|
72 |
+
def merge_data(self) -> None:
|
73 |
+
"""
|
74 |
+
Merges the question and answer DataFrames on a common key.
|
75 |
+
|
76 |
+
This method sets the 'merged_df' attribute to the resulting DataFrame after merging
|
77 |
+
'df_questions' and 'df_answers' on the 'question_id' field, which is assumed to be
|
78 |
+
present in both DataFrames.
|
79 |
+
"""
|
80 |
+
|
81 |
+
self.merged_df = pd.merge(self.df_questions, self.df_answers, on=['question_id', 'image_id'])
|
82 |
+
|
83 |
+
def join_words_with_hyphen(self, sentence):
|
84 |
+
|
85 |
+
return '-'.join(sentence.split())
|
86 |
+
|
87 |
+
def process_answers(self) -> None:
|
88 |
+
"""
|
89 |
+
Processes answers from merged DataFrame by extracting and identifying the most frequent answers.
|
90 |
+
"""
|
91 |
+
if self.merged_df is not None:
|
92 |
+
self.merged_df['raw_answers'] = self.merged_df['answers'].apply(lambda x: [ans['raw_answer'] for ans in x])
|
93 |
+
self.merged_df['processed_answers'] = self.merged_df['answers'].apply(
|
94 |
+
lambda x: [ans['answer'] for ans in x])
|
95 |
+
self.merged_df['most_frequent_raw_answer'] = self.merged_df['raw_answers'].apply(self.find_most_frequent)
|
96 |
+
self.merged_df['most_frequent_processed_answer'] = self.merged_df['processed_answers'].apply(
|
97 |
+
self.find_most_frequent)
|
98 |
+
self.merged_df.drop(columns=['answers'], inplace=True)
|
99 |
+
else:
|
100 |
+
print("DataFrames have not been merged yet.")
|
101 |
+
|
102 |
+
# Apply the function to the 'most_frequent_processed_answer' column
|
103 |
+
self.merged_df['single_word_answers'] = self.merged_df['most_frequent_processed_answer'].apply(
|
104 |
+
self.join_words_with_hyphen)
|
105 |
+
|
106 |
+
def get_processed_data(self) -> Optional[pd.DataFrame]:
|
107 |
+
"""
|
108 |
+
Retrieves the processed DataFrame.
|
109 |
+
|
110 |
+
Returns:
|
111 |
+
Optional[pd.DataFrame]: The processed DataFrame or None if it is not available.
|
112 |
+
"""
|
113 |
+
|
114 |
+
if self.merged_df is not None:
|
115 |
+
return self.merged_df
|
116 |
+
else:
|
117 |
+
print("DataFrame is empty or not processed yet.")
|
118 |
+
return None
|
119 |
+
|
120 |
+
def save_to_csv(self, df: pd.DataFrame, saved_file_name: Optional[str]) -> None:
|
121 |
+
"""
|
122 |
+
Saves the DataFrame to a CSV file.
|
123 |
+
|
124 |
+
Parameters:
|
125 |
+
df (pd.DataFrame): The DataFrame to save.
|
126 |
+
saved_file_name (Optional[str]): The target file name or path.
|
127 |
+
"""
|
128 |
+
|
129 |
+
if saved_file_name is not None:
|
130 |
+
if ".csv" not in saved_file_name:
|
131 |
+
df.to_csv(os.path.join(saved_file_name, ".csv"), index=None)
|
132 |
+
else:
|
133 |
+
df.to_csv(saved_file_name, index=None)
|
134 |
+
else:
|
135 |
+
df.to_csv("data.csv", index=None)
|
136 |
+
|
137 |
+
def display_dataframe(self) -> None:
|
138 |
+
"""
|
139 |
+
Displays the processed DataFrame.
|
140 |
+
"""
|
141 |
+
if self.merged_df is not None:
|
142 |
+
print(self.merged_df)
|
143 |
+
else:
|
144 |
+
print("DataFrame is empty.")
|
145 |
+
|
146 |
+
|
147 |
+
|
148 |
+
def process_okvqa_dataset(questions_file_path: str, annotations_file_path: str, save_to_csv: bool = False,
|
149 |
+
saved_file_name: Optional[str] = None) -> Optional[pd.DataFrame]:
|
150 |
+
"""
|
151 |
+
Orchestrates the processing of the OK-VQA dataset using specified JSON file paths for questions and annotations.
|
152 |
+
|
153 |
+
Parameters:
|
154 |
+
questions_file_path (str): Path to the questions JSON file.
|
155 |
+
annotations_file_path (str): Path to the annotations JSON file.
|
156 |
+
save_to_csv (bool): Flag to determine if the processed data should be saved to CSV.
|
157 |
+
saved_file_name (Optional[str]): Filename or path to save the CSV file. If None, defaults to 'data.csv'.
|
158 |
+
|
159 |
+
Returns:
|
160 |
+
Optional[pd.DataFrame]: The processed DataFrame containing merged and processed VQA data or None if empty.
|
161 |
+
"""
|
162 |
+
# Initialize the dataset processor
|
163 |
+
processor = OKVQADatasetProcessor(questions_file_path, annotations_file_path)
|
164 |
+
|
165 |
+
# Merge question and answer data and process answers
|
166 |
+
processor.merge_data()
|
167 |
+
processor.process_answers()
|
168 |
+
|
169 |
+
# Retrieve the processed DataFrame
|
170 |
+
processed_data = processor.get_processed_data()
|
171 |
+
|
172 |
+
# Optionally save the processed DataFrame to a CSV file
|
173 |
+
if save_to_csv and processed_data is not None:
|
174 |
+
processor.save_to_csv(processed_data, saved_file_name)
|
175 |
+
|
176 |
+
return processed_data
|