File size: 9,636 Bytes
0c40d71
729a987
0c40d71
 
 
 
 
e9796a8
 
199738e
 
70b3d72
840d980
0c40d71
 
a77552d
0c40d71
74ad48b
e23b00b
 
0c40d71
 
d73a02d
 
 
 
0c40d71
729a987
a68f5c3
 
729a987
 
 
 
 
 
 
863a898
729a987
 
a68f5c3
729a987
e23b00b
 
 
840d980
 
e23b00b
729a987
4aff4d5
 
 
729a987
 
a77552d
 
 
0c40d71
a77552d
 
e9796a8
a77552d
 
1335971
0c40d71
a77552d
 
 
 
 
 
 
 
 
 
 
 
 
0c40d71
e23b00b
74ad48b
e9796a8
 
199738e
 
 
e9796a8
 
840d980
e9796a8
e23b00b
729a987
e9796a8
 
 
 
 
 
 
705cade
e9796a8
 
 
 
 
 
705cade
e9796a8
 
 
e23b00b
a68f5c3
 
e23b00b
 
 
 
 
 
 
863a898
e23b00b
 
a68f5c3
e23b00b
 
 
 
 
e9796a8
 
 
 
 
 
e23b00b
e9796a8
 
 
840d980
 
e9796a8
 
 
199738e
 
 
e9796a8
 
70b3d72
74ad48b
70b3d72
 
 
 
 
 
 
840d980
70b3d72
 
 
 
 
 
 
840d980
 
70b3d72
 
0c40d71
74ad48b
e23b00b
 
840d980
e23b00b
199738e
0c40d71
 
199738e
0c40d71
863a898
199738e
 
729a987
 
 
 
a68f5c3
 
729a987
 
 
 
 
 
 
 
863a898
729a987
 
a68f5c3
729a987
 
 
 
199738e
 
0c40d71
 
729a987
0c40d71
729a987
 
 
7b689ca
 
729a987
0c40d71
729a987
 
7b689ca
729a987
 
 
0c40d71
 
 
74ad48b
0c40d71
 
 
 
 
 
 
e9796a8
0c40d71
 
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
import json
from operator import mod
from pathlib import Path
from PIL import Image
import io
from sentence_transformers import util
import torch
import secrets
import time
import numpy as np
from numpy.linalg import norm
import os
import time


def find(modality, query, relatedness, activation, noise, return_embeddings, auth_result, text_encoder, text_image_encoder, silent=False):
    authorized_thoughts = get_authorized_thoughts(auth_result)
    knowledge_base_path = Path('..') / 'knowledge'
    query_embeddings = encode(
        modality, query, text_encoder, text_image_encoder)

    if len(authorized_thoughts) == 0:
        return {
            'authorized_thoughts': [],
            'query_embeddings': query_embeddings
        }

    sims = []
    text_image_scaling = 1
    image_image_scaling = 0.4
    for e in authorized_thoughts:
        if modality == 'text':
            if e['modality'] == 'text':
                sims += [np.dot(e['embeddings']['text'], query_embeddings['text']) / (
                    norm(e['embeddings']['text']) * norm(query_embeddings['text']))]
            elif e['modality'] == 'image':
                sims += [np.dot(e['embeddings']['text_image'], query_embeddings['text_image']) / (
                    norm(e['embeddings']['text_image']) * norm(query_embeddings['text_image'])) * text_image_scaling]
        elif modality == 'image':
            sims += [np.dot(e['embeddings']['text_image'], query_embeddings['text_image']) / (
                norm(e['embeddings']['text_image']) * norm(query_embeddings['text_image'])) * image_image_scaling]

    if not silent and auth_result['custodian']:
        for e_idx, e in enumerate(sims):
            authorized_thoughts[e_idx]['interest'] += e
        open(knowledge_base_path / 'metadata.json',
             'w').write(json.dumps(authorized_thoughts))

    for e_idx, e in enumerate(sims):
        authorized_thoughts[e_idx]['relatedness'] = float(e)
        authorized_thoughts[e_idx]['interest'] = float(
            authorized_thoughts[e_idx]['interest'])
        authorized_thoughts[e_idx]['content'] = get_content(
            authorized_thoughts[e_idx], True)
        if not return_embeddings:
            if 'embeddings' in authorized_thoughts[e_idx]:
                authorized_thoughts[e_idx].pop('embeddings')

    authorized_thoughts = rank(
        authorized_thoughts, relatedness, activation, noise)

    response = {
        'authorized_thoughts': authorized_thoughts
    }

    if return_embeddings:
        response['query_embeddings'] = query_embeddings

    return response


def rank(authorized_thoughts, relatedness, activation, noise):
    return sorted(authorized_thoughts, reverse=True, key=lambda x:
                  (relatedness * x['relatedness'] +
                   activation * (np.log(x['interest'] / (1 - 0.9)) - 0.9 * np.log(
                       (time.time() - x['timestamp']) / (3600 * 24) + 0.1))) *
                  np.random.normal(1, noise))


def save(modality, query, auth_result, text_encoder, text_image_encoder, silent=False):
    knowledge_base_path = Path('..') / 'knowledge'

    if auth_result['custodian'] == False:
        return {
            'message': 'Only the conceptarium\'s custodian can save thoughts in it.'
        }
    else:
        if not (knowledge_base_path / 'metadata.json').exists():
            open(knowledge_base_path / 'metadata.json', 'w').write(json.dumps([]))

        query_embeddings = encode(
            modality, query, text_encoder, text_image_encoder)
        thoughts = json.load(open(knowledge_base_path / 'metadata.json'))

        if modality == 'text':
            duplicates = [e for e in thoughts if e['modality'] ==
                          'text' and open(knowledge_base_path / e['filename']).read() == query]

            if len(duplicates) == 0:
                filename = secrets.token_urlsafe(16) + '.md'
                open(knowledge_base_path / filename, 'w').write(query)
        elif modality == 'image':
            duplicates = [e for e in thoughts if e['modality'] ==
                          'image' and open(knowledge_base_path / e['filename'], 'rb').read() == query]

            if len(duplicates) == 0:
                filename = secrets.token_urlsafe(16) + '.jpg'
                query = Image.open(io.BytesIO(query)).convert('RGB')
                query.save(knowledge_base_path / filename, quality=50)

        sims = []
        text_image_scaling = 1
        image_image_scaling = 0.4
        for e in thoughts:
            if modality == 'text':
                if e['modality'] == 'text':
                    sims += [np.dot(e['embeddings']['text'], query_embeddings['text']) / (
                        norm(e['embeddings']['text']) * norm(query_embeddings['text']))]
                elif e['modality'] == 'image':
                    sims += [np.dot(e['embeddings']['text_image'], query_embeddings['text_image']) / (
                        norm(e['embeddings']['text_image']) * norm(query_embeddings['text_image'])) * text_image_scaling]
            elif modality == 'image':
                sims += [np.dot(e['embeddings']['text_image'], query_embeddings['text_image']) / (
                    norm(e['embeddings']['text_image']) * norm(query_embeddings['text_image'])) * image_image_scaling]

        if not silent:
            for e_idx, e in enumerate(sims):
                thoughts[e_idx]['interest'] += e

        if len(duplicates) == 0:
            new_thought = {
                'filename': filename,
                'modality': modality,
                'timestamp': time.time(),
                'interest': 1,
                'embeddings': query_embeddings
            }

            thoughts += [new_thought]
            open(knowledge_base_path / 'metadata.json',
                 'w').write(json.dumps(thoughts))

            return new_thought
        else:
            return {
                'message': 'Duplicate thought found.'
            }


def remove(auth_result, filename):
    knowledge_base_path = Path('..') / 'knowledge'

    if auth_result['custodian'] == False:
        return {
            'message': 'Only the conceptarium\'s custodian can remove thoughts from it.'
        }
    else:
        if not (knowledge_base_path / 'metadata.json').exists():
            open(knowledge_base_path / 'metadata.json', 'w').write(json.dumps([]))

        thoughts = json.load(open(knowledge_base_path / 'metadata.json'))
        target = [e for e in thoughts if e['filename'] == filename]

        if len(target) > 0:
            os.remove(knowledge_base_path / filename)
            thoughts.remove(target[0])
            open(knowledge_base_path / 'metadata.json',
                 'w').write(json.dumps(thoughts))


def get_authorized_thoughts(auth_result):
    metadata_path = Path('..') / 'knowledge' / 'metadata.json'

    if not (metadata_path).exists():
        open(metadata_path, 'w').write(json.dumps([]))

    thoughts = json.load(open(metadata_path))

    if auth_result['custodian'] == True:
        return thoughts
    else:
        similarity_threshold = 0.3
        authorized_microverse = auth_result['authorized_microverse']

        if authorized_microverse == []:
            return []

        query_embeddings = authorized_microverse[0]['embeddings']
        text_image_scaling = 1
        image_image_scaling = 0.4
        sims = []
        for e in thoughts:
            if authorized_microverse[0]['modality'] == 'text':
                if e['modality'] == 'text':
                    sims += [np.dot(e['embeddings']['text'], query_embeddings['text']) / (
                        norm(e['embeddings']['text']) * norm(query_embeddings['text']))]
                elif e['modality'] == 'image':
                    sims += [np.dot(e['embeddings']['text_image'], query_embeddings['text_image']) / (
                        norm(e['embeddings']['text_image']) * norm(query_embeddings['text_image'])) * text_image_scaling]
            elif authorized_microverse[0]['modality'] == 'image':
                sims += [np.dot(e['embeddings']['text_image'], query_embeddings['text_image']) / (
                    norm(e['embeddings']['text_image']) * norm(query_embeddings['text_image'])) * image_image_scaling]

        scored_thoughts = zip(thoughts, sims)
        authorized_thoughts = [e[0]
                               for e in scored_thoughts if e[1] > similarity_threshold]

        return authorized_thoughts


def encode(modality, content, text_encoder, text_image_encoder):
    if modality == 'text':
        return {
            'text_model': 'sentence-transformers/multi-qa-mpnet-base-cos-v1',
            'text_image_model': 'clip-ViT-B-32',
            'text': [round(e, 5) for e in text_encoder.encode(content).tolist()],
            'text_image': [round(e, 5) for e in text_image_encoder.encode(content).tolist()]
        }
    elif modality == 'image':
        return {
            'text_image_model': 'clip-ViT-B-32',
            'text_image': [round(e, 5) for e in text_image_encoder.encode(Image.open(io.BytesIO(content))).tolist()]
        }
    else:
        raise Exception('Can\'t encode content of modality "' + modality + '"')


def get_content(thought, json_friendly=False):
    knowledge_base_path = Path('..') / 'knowledge'

    if thought['modality'] == 'text':
        content = open(knowledge_base_path / thought['filename']).read()
    elif thought['modality'] == 'image':
        content = open(knowledge_base_path / thought['filename'], 'rb').read()

        if json_friendly:
            content = thought['filename']

    return content