File size: 8,971 Bytes
714a85e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import re
from pathlib import Path

import numpy as np

from maze_generator import *
from prompt_builder import *

def encode_maze(maze, encoding_type):
    if encoding_type == "matrix":
        return encode_standard_matrix_maze(maze)
    elif encoding_type == "coord_list":
        return encode_coordinate_list_maze(maze)
    elif encoding_type == "vision":
        return encode_standard_matrix_maze(maze)
    else:
        raise NotImplementedError

# RECOGNIZE 
def contains_word_case_insensitive(s, word_list):
    word_list = [re.escape(word) for word in word_list]
    pattern = r'\b(?:' + '|'.join(word_list) + r')\b'
    return re.search(pattern, s, re.IGNORECASE) is not None

def is_correct_recognize(response, shape):
    if shape == 'square':
        word_list = ['square', 'box', 'cube', 'symmetrical rectangle', 'right-angled quadrilateral', 'symmetric quadrilateral', "four-sided polygon"]
    elif shape == 'triangle':
        word_list = ['triangle', 'pyramid', 'trilateral', 'three-sided polygon', 'isosceles']
    elif shape == 'spiral':
        word_list = ['spiral', 'helix', 'coil', 'whorl', 'swirl', 'vortex']
    elif shape == 'cross':
        word_list = ['cross', 'times', 'multiplication', 'X', 'X-shape', 'crossed lines', 'crossing']
    elif shape == 'C':
        word_list = ['C', 'crescent', 'half-circle', 'curved line', 'semi-circle', 'semi circle', 'open curve']
    elif shape == 'Z':
        word_list = ['Z', 'zigzag', 'lightning bolt']

    return contains_word_case_insensitive(response, word_list)

# GENERATE

def is_correct_generate(response, encoding_type, size, shape, orig_maze):
    all_mazes = globals()[f"all_{shape}_path_mazes"](size)
    all_mazes_and_paths = []
    for maze in all_mazes:
        all_mazes_and_paths.extend(init_all_start_end(maze))
    all_maze_encodings = set()
    for maze in all_mazes_and_paths:
        all_maze_encodings.add(encode_maze(maze, encoding_type))
    
    all_maze_encodings.remove(encode_maze(orig_maze, encoding_type))

    for maze_encoding in all_maze_encodings:
        if maze_encoding in response:
            return True
    return False

def is_correct_solution(response, maze):
    # Find all coordinates that appear in the response.
    pattern = r'\(\s*(-?\d+)\s*,\s*(-?\d+)\s*\)'
    coordinates = re.findall(pattern, response)
    coordinates = [(int(x), int(y)) for x, y in coordinates]

    # As responses will usually talk about the solution before presenting
    # the final form, we will start from the last provided end point 
    # of the maze and work backwards until we arrive at the solution.
    start = tuple(np.argwhere(maze == POS)[0])
    end = tuple(np.argwhere(maze == END)[0])

    try:
        i = find_last_index(coordinates, end)
    except ValueError:
        return False

    while i > 0:
        curr_pos = coordinates[i]
        if curr_pos == start:
            return True
        next_pos = coordinates[i - 1]
        if next_pos in get_valid_moves(maze, curr_pos):
            i = i - 1
        else:
            break
    return False

def find_last_index(lst, element):
    return len(lst) - 1 - lst[::-1].index(element)

def get_valid_moves(maze, pos):
    x, y = pos
    rows = len(maze)
    cols = len(maze[0])
    
    # All 8 possible directions (up, down, left, right and diagonals)
    directions = [
        (-1, -1), (-1, 0), (-1, 1),
        (0, -1),         (0, 1),
        (1, -1), (1, 0), (1, 1) 
    ]

    valid_moves = []
    for dx, dy in directions:
        new_x, new_y = x + dx, y + dy
        
        # Check if the new coordinates are within bounds
        if 0 <= new_x < rows and 0 <= new_y < cols:
            if maze[new_x, new_y] != WALL:
                valid_moves.append((new_x, new_y))
    
    return valid_moves

def parse_coordinates(text):
    walls = re.findall(r"Walls: ([\d, ()]+)", text)
    empty = re.findall(r"Empty: ([\d, ()]+)", text)
    start = re.findall(r"Start: \((\d+, \d+)\)", text)
    end = re.findall(r"End: \((\d+, \d+)\)", text)

    walls = [tuple(map(int, coord.split(', '))) for coord in re.findall(r"\((\d+, \d+)\)", walls[0])]
    empty = [tuple(map(int, coord.split(', '))) for coord in re.findall(r"\((\d+, \d+)\)", empty[0])]
    start = tuple(map(int, start[0].split(', ')))
    end = tuple(map(int, end[0].split(', ')))

    return walls, empty, start, end

def parse_coordinate_maze(text, size):
    walls, empty, start, end = parse_coordinates(text)
    maze = np.full(size, -1)
    for x, y in walls:
        maze[x, y] = WALL
    for x, y in empty:
        maze[x, y] = PATH
    maze[start[0], start[1]] = POS
    maze[end[0], end[1]] = END
    return maze

def grade_prompts(version):
    parent_dir = Path(f"results/{version}/")
    for model in parent_dir.iterdir():
        grade_info = ""
        total = 0
        total_correct = 0
        total_size = {size_t.name: 0 for size_t in (parent_dir / model.name).iterdir() if not size_t.is_file()}
        total_size_correct = {size_t.name: 0 for size_t in (parent_dir / model.name).iterdir() if not size_t.is_file()}
        total_encoding = {"standard": 0, "unusual": 0, "coordinate": 0}
        total_encoding_correct = {"standard": 0, "unusual": 0, "coordinate": 0}
        total_shape = {shape: 0 for shape in SHAPES}
        total_shape_correct = {shape: 0 for shape in SHAPES}
        for size_t in (parent_dir / model.name).iterdir():
            if size_t.is_file():
                continue
            size = (int(size_t.name[0]), int(size_t.name[2]))
            for encoding in (parent_dir / model.name / size_t.name).iterdir():
                for shape in (parent_dir / model.name / size_t.name / encoding.name).iterdir():
                    for response_file in (parent_dir / model.name / size_t.name / encoding.name / shape.name).iterdir():
                        response_text = response_file.read_text()
                        # Get coordinate version of maze given in prompt so we can decode.
                        coord_prompt_text = Path(f"prompts/v1/diag/{size_t.name}/coordinate/{shape.name}/{response_file.name}").read_text()
                        maze = parse_coordinate_maze(coord_prompt_text, size)
                        is_correct = is_correct_solution(response_text, maze)

                        grade_info += f"{size_t.name}-{encoding.name}: {response_file.name}: {is_correct}\n"

                        total += 1
                        total_correct += is_correct
                        total_size[size_t.name] += 1
                        total_size_correct[size_t.name] += is_correct
                        total_encoding[encoding.name] += 1
                        total_encoding_correct[encoding.name] += is_correct
                        total_shape[shape.name] += 1
                        total_shape_correct[shape.name] += is_correct
        
        grade_info += "\n\n### SUMMARY ###\n"
        grade_info += f"Total: {total_correct}/{total}\n\n"
        grade_info += "Following show total correct for one variable across all others:\n"
        grade_info += "# SIZES #\n"
        for size, correct in total_size_correct.items():
            grade_info += f"{size}: {correct}/{total_size[size]}\n"
        grade_info += "# ENCODING #\n"
        for encoding, correct in total_encoding_correct.items():
            grade_info += f"{encoding}: {correct}/{total_encoding[encoding]}\n"
        grade_info += "# SHAPE #\n"
        for shape, correct in total_shape_correct.items():
            grade_info += f"{shape}: {correct}/{total_shape[shape]}\n"

        (parent_dir / model.name / "grade_info.txt").write_text(grade_info)

        print(f"{model.name} - total: {total_correct}/{total}")
        import csv

        # Define the path to the CSV file
        csv_path = parent_dir / model.name / "grade_info.csv"

        # Prepare the data for the CSV
        csv_data = [
            [model.name, "Total Correct", "Total"],
            ["Summary", total_correct, total],
        ]

        # Add size data to csv_data
        csv_data.append(["SIZES", "Correct", "Total"])
        for size, correct in total_size_correct.items():
            csv_data.append([size, correct, total_size[size]])

        # Add encoding data to csv_data
        csv_data.append(["ENCODING", "Correct", "Total"])
        for encoding, correct in total_encoding_correct.items():
            csv_data.append([encoding, correct, total_encoding[encoding]])

        # Add shape data to csv_data
        csv_data.append(["SHAPE", "Correct", "Total"])
        for shape, correct in total_shape_correct.items():
            csv_data.append([shape, correct, total_shape[shape]])

        # Write the csv_data to the CSV file
        with open(csv_path, mode='w', newline='') as file:
            writer = csv.writer(file)
            writer.writerows(csv_data)
        print(f"CSV file saved at {csv_path}")

if __name__ == "__main__":
    grade_prompts("only_one_per_folder")