34 строки
1.1 KiB
Python
34 строки
1.1 KiB
Python
# Copyright (c) Microsoft Corporation.
|
|
# Licensed under the MIT license.
|
|
|
|
import random
|
|
import numpy as np
|
|
from torch.utils.data.sampler import Sampler
|
|
|
|
class TaskSampler(Sampler):
|
|
"""
|
|
Sampler class for a fixed number of tasks per user/object.
|
|
"""
|
|
def __init__(self, num_tasks_per_item, num_items, shuffle):
|
|
"""
|
|
Creates instances of TaskSampler.
|
|
:param num_tasks_per_item: (int) Number of tasks to sample per user/object.
|
|
:param num_items: (int) Total number of users/objects.
|
|
:param shuffle: (bool) If True, shuffle tasks, otherwise do not shuffle.
|
|
:return: Nothing.
|
|
"""
|
|
self.num_tasks_per_item = num_tasks_per_item
|
|
self.num_items = num_items
|
|
self.shuffle = shuffle
|
|
|
|
def __iter__(self):
|
|
task_ids = []
|
|
for item in range(self.num_items):
|
|
task_ids.extend([item]*self.num_tasks_per_item)
|
|
if self.shuffle:
|
|
random.shuffle(task_ids)
|
|
return iter(task_ids)
|
|
|
|
def __len__(self):
|
|
return self.num_items*self.num_tasks_per_item
|