File size: 9,646 Bytes
d88615f
 
 
 
ea2621e
d88615f
 
cd8c3a9
 
d88615f
 
 
 
 
 
 
 
c4c5fc5
d88615f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cd8c3a9
 
 
 
d88615f
 
 
 
 
 
 
 
 
 
cd8c3a9
 
 
 
d88615f
 
 
 
 
 
 
cd8c3a9
d88615f
 
cd8c3a9
d88615f
 
cd8c3a9
 
 
 
d88615f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cd8c3a9
 
 
 
 
d88615f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cd8c3a9
 
d88615f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cd8c3a9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d88615f
cd8c3a9
ea2621e
 
d88615f
 
 
c4c5fc5
ecbf45e
d88615f
 
 
 
 
 
 
ecbf45e
d88615f
 
 
 
 
 
 
cd8c3a9
d88615f
 
 
 
 
 
 
 
cd8c3a9
 
 
 
e6d1b9e
cd8c3a9
e6d1b9e
 
 
 
cd8c3a9
d88615f
 
 
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
import logging
import os
from datetime import datetime
from decimal import Decimal
from typing import List

import boto3
from boto3.dynamodb.conditions import Attr, Key
from datasets import Dataset

logging.basicConfig(level=os.getenv("LOG_LEVEL", "INFO"))

# Create a DynamoDB client
dynamodb = boto3.resource('dynamodb', region_name='us-east-1')


def _create_arena_table():
    dynamodb.create_table(
        TableName='oaaic_chatbot_arena',
        KeySchema=[
            {
                'AttributeName': 'arena_battle_id',
                'KeyType': 'HASH'
            },
        ],
        AttributeDefinitions=[
            {
                'AttributeName': 'arena_battle_id',
                'AttributeType': 'S'
            },
            {
                'AttributeName': 'timestamp',
                'AttributeType': 'S'
            },
        ],
        ProvisionedThroughput={
            'ReadCapacityUnits': 5,
            'WriteCapacityUnits': 5
        },
        GlobalSecondaryIndexes=[
            {
                'IndexName': 'TimestampIndex',
                'KeySchema': [
                    {
                        'AttributeName': 'arena_battle_id',
                        'KeyType': 'HASH'
                    },
                    {
                        'AttributeName': 'timestamp',
                        'KeyType': 'RANGE'
                    },
                ],
                'Projection': {
                    'ProjectionType': 'ALL',
                },
                'ProvisionedThroughput': {
                    'ReadCapacityUnits': 5,
                    'WriteCapacityUnits': 5,
                }
            },
        ]
    )

def _create_elo_scores_table():
    dynamodb.create_table(
        TableName='elo_scores',
        KeySchema=[
            {
                'AttributeName': 'chatbot_name',
                'KeyType': 'HASH'  # Partition key
            },
        ],
        AttributeDefinitions=[
            {
                'AttributeName': 'chatbot_name',
                'AttributeType': 'S'
            },
        ],
        ProvisionedThroughput={
            'ReadCapacityUnits': 5,
            'WriteCapacityUnits': 5
        }
    )


def _create_elo_logs_table():
    dynamodb.create_table(
        TableName='elo_logs',
        KeySchema=[
            {
                'AttributeName': 'arena_battle_id',
                'KeyType': 'HASH'  # Partition key
            },
            {
                'AttributeName': 'battle_timestamp',
                'KeyType': 'RANGE'  # Sort key
            },
        ],
        AttributeDefinitions=[
            {
                'AttributeName': 'arena_battle_id',
                'AttributeType': 'S'
            },
            {
                'AttributeName': 'battle_timestamp',
                'AttributeType': 'S'
            },
            {
                'AttributeName': 'all',
                'AttributeType': 'S'
            }
        ],
        ProvisionedThroughput={
            'ReadCapacityUnits': 10,
            'WriteCapacityUnits': 10
        },
        GlobalSecondaryIndexes=[
            {
                'IndexName': 'AllTimestampIndex',
                'KeySchema': [
                    {
                        'AttributeName': 'all',
                        'KeyType': 'HASH'  # Partition key for the GSI
                    },
                    {
                        'AttributeName': 'battle_timestamp',
                        'KeyType': 'RANGE'  # Sort key for the GSI
                    }
                ],
                'Projection': {
                    'ProjectionType': 'ALL'
                },
                'ProvisionedThroughput': {
                    'ReadCapacityUnits': 10,
                    'WriteCapacityUnits': 10
                }
            },
        ]
    )


def get_unprocessed_battles(last_processed_timestamp):
    # Use boto3 to create a DynamoDB resource and reference the table
    table = dynamodb.Table('oaaic_chatbot_arena')

    # Use a query to retrieve unprocessed battles in temporal order
    response = table.scan(
        FilterExpression=Attr('timestamp').gt(last_processed_timestamp),
        # ScanIndexForward=True
    )

    return response['Items']


def calculate_elo(rating1, rating2, result, K=32):
    # Convert ratings to float
    rating1 = float(rating1)
    rating2 = float(rating2)

    # Calculate the expected outcomes
    expected_outcome1 = 1.0 / (1.0 + 10.0 ** ((rating2 - rating1) / 400.0))
    expected_outcome2 = 1.0 - expected_outcome1

    # Calculate the new Elo ratings
    new_rating1 = rating1 + K * (result - expected_outcome1)
    new_rating2 = rating2 + K * ((1.0 - result) - expected_outcome2)

    return Decimal(new_rating1).quantize(Decimal('0.00')), Decimal(new_rating2).quantize(Decimal('0.00'))


def get_last_processed_timestamp():
    table = dynamodb.Table('elo_logs')

    # Scan the table sorted by timestamp in descending order
    response = table.query(
        IndexName='AllTimestampIndex',
        KeyConditionExpression=Key('all').eq('ALL'),
        ScanIndexForward=False,
        Limit=1
    )

    # If there are no items in the table, return a default timestamp
    if not response['Items']:
        return '1970-01-01T00:00:00'

    # Otherwise, return the timestamp of the latest item
    return response['Items'][0]['battle_timestamp']


def log_elo_update(arena_battle_id, battle_timestamp, new_rating1, new_rating2):
    # Reference the elo_logs table
    table = dynamodb.Table('elo_logs')

    # Update the table
    table.put_item(
        Item={
            'arena_battle_id': arena_battle_id,
            'battle_timestamp': battle_timestamp,  # Use the timestamp of the battle
            'log_timestamp': datetime.now().isoformat(),  # Also store the timestamp of the log for completeness
            'new_rating1': new_rating1,
            'new_rating2': new_rating2,
            'all': 'ALL',
        }
    )


def get_elo_score(chatbot_name, elo_scores):
    if chatbot_name in elo_scores:
        return elo_scores[chatbot_name]

    table = dynamodb.Table('elo_scores')
    response = table.get_item(Key={'chatbot_name': chatbot_name})

    # If there is no item in the table, return a default score
    if 'Item' not in response:
        return 1500

    return response['Item']['elo_score']


def update_elo_score(chatbot_name, new_elo_score):
    table = dynamodb.Table('elo_scores')

    # This will create a new item if it doesn't exist
    table.put_item(
        Item={
            'chatbot_name': chatbot_name,
            'elo_score': Decimal(str(new_elo_score)),
        }
    )


def get_elo_scores():
    table = dynamodb.Table('elo_scores')

    response = table.scan()
    data = response['Items']

    return data


def _backfill_logs():
    table = dynamodb.Table('elo_logs')

    # Initialize the scan operation
    response = table.scan()

    for item in response['Items']:
        table.update_item(
            Key={
                'arena_battle_id': item['arena_battle_id'],
                'battle_timestamp': item['battle_timestamp']
            },
            UpdateExpression="SET #all = :value",
            ExpressionAttributeNames={
                '#all': 'all'
            },
            ExpressionAttributeValues={
                ':value': 'ALL'
            }
        )

def main():
    last_processed_timestamp = get_last_processed_timestamp()
    battles: List[dict] = get_unprocessed_battles(last_processed_timestamp)
    battles = sorted(battles, key=lambda x: x['timestamp'])
    elo_scores = {}

    for battle in battles:
        print(repr(battle))
        if battle['label'] in {-1, 0, 1, 2}:
            outcome = battle['label']
            for chatbot_name in [battle['choice1_name'], battle['choice2_name']]:
                if chatbot_name not in elo_scores:
                    elo_scores[chatbot_name] = get_elo_score(chatbot_name, elo_scores)
            # 1: This means that the first player (or team) won the match.
            # 0.5: This means that the match ended in a draw.
            # 0: This means that the first player (or team) lost the match.
            if outcome == 0 or outcome == -1:
                elo_result = 0.5
            elif outcome == 1:
                elo_result = 1
            else:
                elo_result = 0

            new_rating1, new_rating2 = calculate_elo(elo_scores[battle['choice1_name']], elo_scores[battle['choice2_name']], elo_result)
            logging.info(f"{battle['choice1_name']}: {elo_scores[battle['choice1_name']]} -> {new_rating1} | {battle['choice2_name']}: {elo_scores[battle['choice2_name']]} -> {new_rating2}")
            elo_scores[battle['choice1_name']] = new_rating1
            elo_scores[battle['choice2_name']] = new_rating2
            log_elo_update(battle['arena_battle_id'], battle['timestamp'], new_rating1, new_rating2)
            update_elo_score(battle['choice1_name'], new_rating1)
            update_elo_score(battle['choice2_name'], new_rating2)
            elo_scores[battle['choice1_name']] = new_rating1
            elo_scores[battle['choice2_name']] = new_rating2

    elo_scores = get_elo_scores()
    for i, j in enumerate(elo_scores):
        j["elo_score"] = float(j["elo_score"])
        elo_scores[i] = j
    print(elo_scores)

    if battles:
        # Convert the data into a format suitable for Hugging Face Dataset
        elo_dataset = Dataset.from_list(elo_scores)
        elo_dataset.push_to_hub("openaccess-ai-collective/chatbot-arena-elo-scores", private=False)


if __name__ == "__main__":
    main()