File size: 5,503 Bytes
d6bc023
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
import json
import os
import math
import copy
import argparse
import numpy as np

def write_jsonl(data, filename):
    with open(filename, 'w') as f:
        for item in data:
            f.write(json.dumps(item) + '\n')

def RelaxedAccuracy(pred, gt):
    try:
        gt = float(gt)
        pred = float(pred)
        if gt == 0.0:
            if pred == gt:
                return 1.0
            else:
                return 0.0
        else:
            if abs(pred-gt) / gt <= 0.05:
                return 1.0
            else:
                return 0.0
    except:
        if str(gt) == str(pred):
            return 1.0
        else:
            return 0.0

def evaluate_cmds(cmds):
    for cmd in cmds:
        exec(cmd)
    answer = eval('Answer')
    if (isinstance(answer, list) or isinstance(answer, np.ndarray)) and len(answer) == 1:
        answer = answer[0]
    if isinstance(answer, list) or isinstance(answer, np.ndarray):
        new_answer = answer[0]
        for i in range(1, len(answer)-1):
            new_answer = new_answer + ', ' + answer[i]
        new_answer += ' and ' + answer[-1]
        answer = new_answer
    if isinstance(answer, bool) or isinstance(answer, np.bool_):
        if answer == True:
            answer = 'Yes'
        elif answer == False:
            answer = 'No'
    return answer

def parse_model_output(cmdstr):
    lines = cmdstr.split('\n')
    new_lines = []
    for line in lines:
        if '<step>' in line or '</step>' in line:
            line = line.replace('<step>', '').replace('</step>', '')
            new_lines.append(line)
    return new_lines

def chartqa_evaluator(data, key='final_model_answer'):
    acc = 0
    for item in data:
        item['relaxed_acc'] = RelaxedAccuracy(item[key], item['gt_answer'].split('<pot_note>')[0])
        if item['relaxed_acc'] == 1.0:
            acc += 1
    accuracy = acc/len(data)
    return data, accuracy

def chartqapot_evaluator(output_data):
    correct_items = []
    wrong_items = []
    error_items = []
    output_data = copy.deepcopy(output_data)
    acc = 0
    for item in output_data:
        # cmds = parse_gpt_cmd(gpt_item['eval_cmd'])
        eval_cmds = parse_model_output(item['model_answer'])
        try:
            answer = evaluate_cmds(eval_cmds)
            item['final_model_answer'] = str(answer)
        except:
            error_items.append(item)
            item['final_model_answer'] = 'Execute <error>'
            item['relaxed_acc'] = 0.0
            continue
        item['gt_answer'] = item['gt_answer'].split('<cot_note>')[0]
        item['relaxed_acc'] = RelaxedAccuracy(str(answer), item['gt_answer'])
        
        if item['relaxed_acc'] == 1.0:
            correct_items.append(item)
        else:
            wrong_items.append(item)
    total = len(output_data)
    accuracy = len(correct_items)/total
    error_rate = len(error_items)/total
    
    return output_data, accuracy, error_rate

def rule_based_divider(question):
    calculate_words = [
        'sum', 'difference', 'times', 'summation', 'exceed', 
        'below', 'addition', 'fewer', 'subtract', ' mode ', 
        'ratio', 'division', 'average', 'mean', 'bigger', 
        'greater', ' less ', 'tallest', 'number', 'divide', 
        ' add ', 'absolute', 'dividing', 'differ', ' minus ', 
        'how many colors', 'lowest', 'what is the value', 'higher', 
        'longer', ' biggest ', 'lowest'
    ]
        
    for w in calculate_words:
        if w in question.lower():
            return 'pot'
    return 'direct'

def chartqa_rule_merger_evaluator(direct_data, pot_data):
    direct_data, _ = chartqa_evaluator(direct_data, key='model_answer')
    assert len(direct_data) == len(pot_data), 'direct and pot num inconsistent'
    acc_count = 0
    merged_data = []
    for datum1, datum2 in zip(direct_data, pot_data):
        if rule_based_divider(datum1['question']) == 'pot' and '<error>' not in datum2['final_model_answer'] and datum2['final_model_answer'] not in ['inf', '-inf', 'nan', 'np.nan', 'np.inf', '-np.inf']:
            acc_count += datum2['relaxed_acc']
            merged_data.append(datum2)
        else:
            acc_count += datum1['relaxed_acc']
            merged_data.append(datum1)
    accuracy = acc_count/len(direct_data)
    return merged_data, accuracy

def chartqa_oracle_merger_evaluator(direct_data, pot_data):
    direct_data, _ = chartqa_evaluator(direct_data, key='model_answer')   
    assert len(direct_data) == len(pot_data), 'direct and pot num inconsistent'
    acc_count = 0
    merged_data = []
    for datum1, datum2 in zip(direct_data, pot_data):
        if datum1['relaxed_acc'] != 1.0:
            acc_count += datum2['relaxed_acc']
            merged_data.append(datum2)
        else:
            acc_count += datum1['relaxed_acc']
            merged_data.append(datum1)
    accuracy = acc_count/len(direct_data)
    return merged_data, accuracy


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--direct', default='../eval_iter12000_0226/ChartQA_test_12000_pred.jsonl')
    parser.add_argument('--pot', default='../eval_iter12000_0226/ChartQA_test_pot_12000_eval.jsonl')
    parser.add_argument('--output', default='../eval_iter12000_0226/ChartQA_test_pot_12000_merged.jsonl')
    
    args = parser.parse_args()
    
    merged = oracle_merger(args.direct, args.pot)
    merged = rule_based_merger(args.direct, args.pot)
    
    write_jsonl(merged, args.output)