MultiTRON/src/sample.py at main · otto-de/MultiTRON · GitHub

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
import itertools
from random import sample

import numpy as np


def _uniform_negatives(num_items, shape):
    return np.random.randint(1, num_items + 1, shape)


def _uniform_negatives_session_rejected(num_items, shape, in_session_items):
    negatives = []
    for _ in range(np.prod(shape)):
        negative = np.random.randint(1, num_items + 1)
        while negative in in_session_items:
            negative = np.random.randint(1, num_items + 1)
        negatives.append(negative)
    return np.array(negatives).reshape(shape)


def _infer_shape(session_len, num_uniform_negatives, sampling_style):
    if sampling_style == "eventwise":
        return [session_len, num_uniform_negatives]
    elif sampling_style == "sessionwise":
        return [num_uniform_negatives]
    else:
        return []


def sample_uniform(num_items, shape, in_session_items, reject_session_items):
    if reject_session_items:
        return _uniform_negatives_session_rejected(num_items, shape, in_session_items)
    else:
        return _uniform_negatives(num_items, shape)


def sample_uniform_negatives_with_shape(clicks, num_items, session_len, num_uniform_negatives, sampling_style,
                                        reject_session_items):
    in_session_items = set(clicks)
    shape = _infer_shape(session_len, num_uniform_negatives, sampling_style)
    if shape:
        negatives = sample_uniform(num_items, shape, in_session_items, reject_session_items)
    else:
        negatives = np.array([])
    return negatives


def sample_in_batch_negatives(batch_positives, num_in_batch_negatives, batch_session_len, reject_session_items):
    in_batch_negatives = []
    positive_indices = itertools.accumulate(batch_session_len)
    positive_indices = [0] + [p for p in positive_indices]
    if reject_session_items:
        for i in range(len(positive_indices[:-1])):
            candidate_positives = batch_positives[:positive_indices[i]] + batch_positives[positive_indices[i + 1]:]
            in_batch_negatives.append(sample(candidate_positives, num_in_batch_negatives))
    else:
        for i in range(len(batch_session_len)):
            in_batch_negatives.append(sample(batch_positives, num_in_batch_negatives))
    return in_batch_negatives