File size: 15,777 Bytes
c996cf4
 
 
 
 
 
 
 
 
 
cddc6a6
c996cf4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cddc6a6
c996cf4
cddc6a6
c996cf4
 
cddc6a6
c996cf4
 
 
cddc6a6
c996cf4
 
 
 
 
cddc6a6
c996cf4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cddc6a6
c996cf4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cddc6a6
c996cf4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cddc6a6
c996cf4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cddc6a6
 
 
c996cf4
cddc6a6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c996cf4
 
cddc6a6
 
 
 
 
 
 
 
 
 
 
 
c996cf4
 
 
cddc6a6
 
c996cf4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cddc6a6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
import streamlit as st
import json
from collections import Counter
import contractions
import csv
import altair as alt
from typing import Tuple, List, Optional
from my_model.dataset.dataset_processor import process_okvqa_dataset
from my_model.config import dataset_config as config


class OKVQADatasetAnalyzer:
    """
    Provides tools for analyzing and visualizing distributions of question types within given question datasets.
    It supports operations such as data loading, categorization of questions based on keywords, visualization of q
    uestion distribution, and exporting data to CSV files.

    Attributes:
        train_file_path (str): Path to the training dataset file.
        test_file_path (str): Path to the testing dataset file.
        data_choice (str): Choice of dataset(s) to analyze; options include 'train', 'test', or 'train_test'.
        questions (List[str]): List of questions aggregated based on the dataset choice.
        question_types (Counter): Counter object tracking the frequency of each question type.
        Qs (Dict[str, List[str]]): Dictionary mapping question types to lists of corresponding questions.
    """

    def __init__(self, train_file_path: str, test_file_path: str, data_choice: str):
        """
        Initializes the OKVQADatasetAnalyzer with paths to dataset files and a choice of which datasets to analyze.

        Parameters:
            train_file_path (str): Path to the training dataset JSON file. This file should contain a list of questions.
            test_file_path (str): Path to the testing dataset JSON file. This file should also contain a list of
                                  questions.
            data_choice (str): Specifies which dataset(s) to load and analyze. Valid options are 'train', 'test', or
                               'train_test'indicating whether to load training data, testing data, or both.

        The constructor initializes the paths, selects the dataset based on the choice, and loads the initial data by
        calling the `load_data` method.
        It also prepares structures for categorizing questions and storing the results.
        """
        
        self.train_file_path = train_file_path
        self.test_file_path = test_file_path
        self.data_choice = data_choice
        self.questions = []
        self.question_types = Counter()
        self.Qs = {keyword: [] for keyword in config.QUESTION_KEYWORDS + ['others']}
        self.load_data()

    def load_data(self) -> None:
        """
        Loads the dataset(s) from the specified JSON file(s) based on the user's choice of 'train', 'test', or
        'train_test'.
        This method updates the internal list of questions depending on the chosen dataset.
        """

        if self.data_choice in ['train', 'train_test']:
            with open(self.train_file_path, 'r') as file:
                train_data = json.load(file)
                self.questions += [q['question'] for q in train_data['questions']]

        if self.data_choice in ['test', 'train_test']:
            with open(self.test_file_path, 'r') as file:
                test_data = json.load(file)
                self.questions += [q['question'] for q in test_data['questions']]

    def categorize_questions(self) -> None:
        """
        Categorizes each question in the loaded data into predefined categories based on keywords.
        This method updates the internal dictionary `self.Qs` and the Counter `self.question_types` with categorized
        questions.
        """

        question_keywords = self.QUESTION_KEYWORDS

        for question in self.questions:
            question = contractions.fix(question)
            words = question.lower().split()
            question_keyword = None
            if words[:2] == ['name', 'the']:
                question_keyword = 'name the'
            else:
                for word in words:
                    if word in question_keywords:
                        question_keyword = word
                        break
            if question_keyword:
                self.question_types[question_keyword] += 1
                self.Qs[question_keyword].append(question)
            else:
                self.question_types["others"] += 1
                self.Qs["others"].append(question)

    def plot_question_distribution(self) -> None:
        """
        Plots an interactive bar chart of question types using Altair and Streamlit, displaying the count and percentage
         of each type.
        The chart sorts question types by count in descending order and includes detailed tooltips for interaction.
        This method is intended for visualization in a Streamlit application.
        """

        # Prepare data
        total_questions = sum(self.question_types.values())
        items = [(key, value, (value / total_questions) * 100) for key, value in self.question_types.items()]
        df = pd.DataFrame(items, columns=['Question Keyword', 'Count', 'Percentage'])

        # Sort data and handle 'others' category specifically if present
        df = df[df['Question Keyword'] != 'others'].sort_values('Count', ascending=False)
        if 'others' in self.question_types:
            others_df = pd.DataFrame([('others', self.question_types['others'],
                                       (self.question_types['others'] / total_questions) * 100)],
                                     columns=['Question Keyword', 'Count', 'Percentage'])
            df = pd.concat([df, others_df], ignore_index=True)

        # Explicitly set the order of the x-axis based on the sorted DataFrame
        order = df['Question Keyword'].tolist()

        # Create the bar chart
        bars = alt.Chart(df).mark_bar().encode(
            x=alt.X('Question Keyword:N', sort=order, title='Question Keyword', axis=alt.Axis(labelAngle=-45)),
            y=alt.Y('Count:Q', title='Question Count'),
            color=alt.Color('Question Keyword:N', scale=alt.Scale(scheme='category20'), legend=None),
            tooltip=[alt.Tooltip('Question Keyword:N', title='Type'),
                     alt.Tooltip('Count:Q', title='Count'),
                     alt.Tooltip('Percentage:Q', title='Percentage', format='.1f')]
        )

        # Create text labels for the bars with count and percentage
        text = bars.mark_text(
            align='center',
            baseline='bottom',
            dy=-5  # Nudges text up so it appears above the bar
        ).encode(
            text=alt.Text('PercentageText:N')
        ).transform_calculate(
            PercentageText="datum.Count + ' (' + format(datum.Percentage, '.1f') + '%)'"
        )

        # Combine the bar and text layers
        chart = (bars + text).properties(
            width=800,
            height=600,
        ).configure_axis(
            labelFontSize=12,
            titleFontSize=16,
            labelFontWeight='bold',
            titleFontWeight='bold',
            grid=False
        ).configure_text(
            fontWeight='bold'
        ).configure_title(
        fontSize=20,
        font='bold',
        anchor='middle'
        )

        # Display the chart in Streamlit
        st.altair_chart(chart, use_container_width=True)

    def plot_bar_chart(self, df: pd.DataFrame, category_col: str, value_col: str, chart_title: str) -> None:
        """
        Plots an interactive bar chart using Altair and Streamlit.

        Args:
            df (pd.DataFrame): DataFrame containing the data for the bar chart.
            category_col (str): Name of the column containing the categories.
            value_col (str): Name of the column containing the values.
            chart_title (str): Title of the chart.

        Returns:
            None
        """
        # Calculate percentage for each category
        df['Percentage'] = (df[value_col] / df[value_col].sum()) * 100
        df['PercentageText'] = df['Percentage'].round(1).astype(str) + '%'

        # Create the bar chart
        bars = alt.Chart(df).mark_bar().encode(
            x=alt.X(field=category_col, title='Category', sort='-y', axis=alt.Axis(labelAngle=-45)),
            y=alt.Y(field=value_col, type='quantitative', title='Percentage'),
            color=alt.Color(field=category_col, type='nominal', legend=None),
            tooltip=[
                alt.Tooltip(field=category_col, type='nominal', title='Category'),
                alt.Tooltip(field=value_col, type='quantitative', title='Percentage'),
                alt.Tooltip(field='Percentage', type='quantitative', title='Percentage', format='.1f')
            ]
        ).properties(
            width=800,
            height=600
        )

        # Add text labels to the bars
        text = bars.mark_text(
            align='center',
            baseline='bottom',
            dy=-10  # Nudges text up so it appears above the bar
        ).encode(
            text=alt.Text('PercentageText:N')
        )

        # Combine the bar chart and text labels
        chart = (bars + text).configure_title(
            fontSize=20
        ).configure_axis(
            labelFontSize=12,
            titleFontSize=16,
            labelFontWeight='bold',
            titleFontWeight='bold',
            grid=False
        ).configure_text(
            fontWeight='bold')

        # Display the chart in Streamlit
        st.altair_chart(chart, use_container_width=True)



    def export_to_csv(self, qs_filename: str, question_types_filename: str) -> None:
        """
        Exports the categorized questions and their counts to two separate CSV files.

        Parameters:
            qs_filename (str): The filename or path for exporting the `self.Qs` dictionary data.
            question_types_filename (str): The filename or path for exporting the `self.question_types` Counter data.

        This method writes the contents of `self.Qs` and `self.question_types` to the specified files in CSV format.
        Each CSV file includes headers for better understanding and use of the exported data.
        """

        # Export self.Qs dictionary
        with open(qs_filename, mode='w', newline='', encoding='utf-8') as file:
            writer = csv.writer(file)
            writer.writerow(['Question Type', 'Questions'])
            for q_type, questions in self.Qs.items():
                for question in questions:
                    writer.writerow([q_type, question])

        # Export self.question_types Counter
        with open(question_types_filename, mode='w', newline='', encoding='utf-8') as file:
            writer = csv.writer(file)
            writer.writerow(['Question Type', 'Count'])
            for q_type, count in self.question_types.items():
                writer.writerow([q_type, count])



def run_dataset_analyzer():
    datasets_comparison_table = pd.read_excel("dataset_analyses.xlsx", sheet_name="VQA Datasets Comparison")
    okvqa_dataset_characteristics = pd.read_excel("dataset_analyses.xlsx", sheet_name="OK-VQA Dataset Characteristics")
    val_data = process_okvqa_dataset('OpenEnded_mscoco_val2014_questions.json', 'mscoco_val2014_annotations.json',
                                     save_to_csv=False)
    train_data = process_okvqa_dataset('OpenEnded_mscoco_train2014_questions.json', 'mscoco_train2014_annotations.json',
                                       save_to_csv=False)



    dataset_analyzer = OKVQADatasetAnalyzer('OpenEnded_mscoco_train2014_questions.json',
                                            'OpenEnded_mscoco_val2014_questions.json', 'train_test')

    with st.container():
        st.markdown("## Overview of KB-VQA Datasets")
        col1, col2 = st.columns([2, 1])
        with col1:
            st.write(" ")
            with st.expander("1 - Knowledge-Based VQA (KB-VQA)"):
                st.markdown(""" [Knowledge-Based VQA (KB-VQA)](https://arxiv.org/abs/1511.02570): One of the earliest 
                                datasets in this domain, KB-VQA comprises 700 images and 2,402 questions, with each 
                                question associated with both an image and a knowledge base (KB). The KB encapsulates 
                                facts about the world, including object names, properties, and relationships, aiming to
                                 foster models capable of answering questions through reasoning over both the image 
                                 and the KB.\n""")
            with st.expander("2 - Factual VQA (FVQA)"):
                st.markdown(""" [Factual VQA (FVQA)](https://arxiv.org/abs/1606.05433): This dataset includes 2,190 
                                images and 5,826 questions, accompanied by a knowledge base containing 193,449 facts. 
                                The FVQA's questions are predominantly factual and less open-ended compared to those 
                                in KB-VQA, offering a different challenge in knowledge-based reasoning.\n""")
            with st.expander("3 - Outside-Knowledge VQA (OK-VQA)"):
                st.markdown(""" [Outside-Knowledge VQA (OK-VQA)](https://arxiv.org/abs/1906.00067): OK-VQA poses a more 
                                demanding challenge than KB-VQA, featuring an open-ended knowledge base that can be 
                                updated during model training. This dataset contains 14,055 questions and 14,031 images.
                                Questions are carefully curated to ensure they require reasoning beyond the image 
                                content alone.\n""")
            with st.expander("4 - Augmented OK-VQA (A-OKVQA)"):
                st.markdown(""" [Augmented OK-VQA (A-OKVQA)](https://arxiv.org/abs/2206.01718): Augmented successor of 
                                OK-VQA dataset, focused on common-sense knowledge and reasoning rather than purely 
                                factual knowledge, A-OKVQA offers approximately 24,903 questions across 23,692 images. 
                                Questions in this dataset demand commonsense reasoning about the scenes depicted in the
                                images, moving beyond straightforward knowledge base queries. It also provides 
                                rationales for answers, aiming to be a significant testbed for the development of AI 
                                models that integrate visual and natural language reasoning.\n""")
        with col2:
            st.markdown("#### KB-VQA Datasets Comparison")
            st.write(datasets_comparison_table, use_column_width=True)
    st.write("-----------------------")
    with st.container():
        st.write("\n" * 10)
        st.markdown("## OK-VQA Dataset")
        st.write("This model was fine-tuned and evaluated using OK-VQA dataset.\n")
        col1, col2, col3 = st.columns([2, 5, 5])

        with col1:
            st.markdown("#### OK-VQA Dataset Characteristics")
            st.write(okvqa_dataset_characteristics)
        with col2:
            df = pd.read_excel("dataset_analyses.xlsx", sheet_name="Question Category Dist")
            st.markdown("#### Questions Distribution over Knowledge Category")
            dataset_analyzer.plot_bar_chart(df, "Knowledge Category", "Percentage", "Questions Distribution over "
                                                                                      "Knowledge Category")

        with col3:

            #with st.expander("Distribution of Question Keywords"):
            dataset_analyzer.categorize_questions()
            st.markdown("#### Distribution of Question Keywords")
            dataset_analyzer.plot_question_distribution()

    with st.container():
        with st.expander("Show Dataset Samples"):
            st.write(train_data[:10])