1 /**
2 * Copyright 2021 Huawei Technologies Co., Ltd
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #include "minddata/dataset/engine/datasetops/source/sampler/subset_sampler.h"
18
19 #include <algorithm>
20 #include <string>
21
22 namespace mindspore {
23 namespace dataset {
24 // Constructor.
SubsetSamplerRT(const std::vector<int64_t> & indices,int64_t num_samples,int64_t samples_per_tensor)25 SubsetSamplerRT::SubsetSamplerRT(const std::vector<int64_t> &indices, int64_t num_samples, int64_t samples_per_tensor)
26 : SamplerRT(num_samples, samples_per_tensor), indices_(indices), sample_id_(0) {}
27
28 // Initialized this Sampler.
InitSampler()29 Status SubsetSamplerRT::InitSampler() {
30 if (is_initialized) {
31 return Status::OK();
32 }
33 CHECK_FAIL_RETURN_UNEXPECTED(
34 num_rows_ > 0, "Invalid parameter, num_rows must be greater than 0, but got " + std::to_string(num_rows_) + ".\n");
35
36 // Special value of 0 for num_samples means that the user wants to sample the entire set of data.
37 // In this case, the id's are provided by the user. Cap the num_samples on the number of id's given.
38 if (num_samples_ == 0 || num_samples_ > static_cast<int64_t>(indices_.size())) {
39 num_samples_ = static_cast<int64_t>(indices_.size());
40 }
41
42 if (samples_per_tensor_ > num_samples_) {
43 samples_per_tensor_ = num_samples_;
44 }
45
46 is_initialized = true;
47 return Status::OK();
48 }
49
50 // Reset the internal variable to the initial state.
ResetSampler(const bool failover_reset)51 Status SubsetSamplerRT::ResetSampler(const bool failover_reset) {
52 // Reset the internal counters.
53 sample_id_ = 0;
54
55 if (HasChildSampler()) {
56 RETURN_IF_NOT_OK(child_[0]->ResetSampler(failover_reset));
57 }
58
59 return Status::OK();
60 }
61
62 // Get the sample ids.
GetNextSample(TensorRow * out)63 Status SubsetSamplerRT::GetNextSample(TensorRow *out) {
64 RETURN_UNEXPECTED_IF_NULL(out);
65 // All samples have been drawn
66 if (sample_id_ == num_samples_) {
67 (*out) = TensorRow(TensorRow::kFlagEOE);
68 } else {
69 if (HasChildSampler()) {
70 RETURN_IF_NOT_OK(child_[0]->GetNextSample(&child_ids_));
71 }
72
73 std::shared_ptr<Tensor> outputIds;
74
75 int64_t last_id = sample_id_ + samples_per_tensor_;
76 // Handling the return all samples at once, and when last draw is not a full batch.
77 if (last_id > num_samples_) {
78 last_id = num_samples_;
79 }
80
81 // Allocate tensor
82 RETURN_IF_NOT_OK(CreateSamplerTensor(&outputIds, last_id - sample_id_));
83
84 // Initialize tensor
85 auto id_ptr = outputIds->begin<int64_t>();
86 while (sample_id_ < last_id) {
87 if (indices_[sample_id_] >= num_rows_ || indices_[sample_id_] < 0) {
88 std::string err_msg = "Sample ID (" + std::to_string(indices_[sample_id_]) +
89 ") is out of bound, expected range [0, " + std::to_string(num_rows_ - 1) + "]";
90 RETURN_STATUS_UNEXPECTED(err_msg);
91 }
92
93 int64_t sampled_id = ((indices_[sample_id_] % num_rows_) + num_rows_) % num_rows_;
94 if (HasChildSampler()) {
95 RETURN_IF_NOT_OK(GetAssociatedChildId(&sampled_id, sampled_id));
96 }
97
98 *id_ptr = sampled_id;
99 ++id_ptr;
100 sample_id_++;
101 }
102
103 (*out) = {outputIds};
104 }
105
106 return Status::OK();
107 }
108
SamplerPrint(std::ostream & out,bool show_all) const109 void SubsetSamplerRT::SamplerPrint(std::ostream &out, bool show_all) const {
110 out << "\nSampler: SubsetSampler";
111 if (show_all) {
112 // Call the super class for displaying any common detailed info
113 SamplerRT::SamplerPrint(out, show_all);
114 // Then add our own info if any
115 }
116 }
117
to_json(nlohmann::json * out_json)118 Status SubsetSamplerRT::to_json(nlohmann::json *out_json) {
119 RETURN_UNEXPECTED_IF_NULL(out_json);
120 nlohmann::json args;
121 RETURN_IF_NOT_OK(SamplerRT::to_json(&args));
122 args["sampler_name"] = "SubsetSampler";
123 args["indices"] = indices_;
124
125 *out_json = args;
126 return Status::OK();
127 }
128
CalculateNumSamples(int64_t num_rows)129 int64_t SubsetSamplerRT::CalculateNumSamples(int64_t num_rows) {
130 int64_t child_num_rows = num_rows;
131 if (!child_.empty()) {
132 child_num_rows = child_[0]->CalculateNumSamples(num_rows);
133 // return -1 if child_num_rows is undetermined
134 if (child_num_rows == -1) {
135 return child_num_rows;
136 }
137 }
138 int64_t res = (num_samples_ > 0) ? std::min(child_num_rows, num_samples_) : child_num_rows;
139 res = std::min(res, static_cast<int64_t>(indices_.size()));
140 return res;
141 }
142 } // namespace dataset
143 } // namespace mindspore
144