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 #include "minddata/dataset/engine/datasetops/source/sampler/subset_sampler.h"
17
18 namespace mindspore {
19 namespace dataset {
20 // Constructor.
SubsetSamplerRT(const std::vector<int64_t> & indices,int64_t num_samples,int64_t samples_per_tensor)21 SubsetSamplerRT::SubsetSamplerRT(const std::vector<int64_t> &indices, int64_t num_samples, int64_t samples_per_tensor)
22 : SamplerRT(num_samples, samples_per_tensor), indices_(indices), sample_id_(0) {}
23
24 // Initialized this Sampler.
InitSampler()25 Status SubsetSamplerRT::InitSampler() {
26 if (is_initialized) {
27 return Status::OK();
28 }
29 CHECK_FAIL_RETURN_UNEXPECTED(
30 num_rows_ > 0, "Invalid parameter, num_rows must be greater than 0, but got " + std::to_string(num_rows_) + ".\n");
31
32 // Special value of 0 for num_samples means that the user wants to sample the entire set of data.
33 // In this case, the id's are provided by the user. Cap the num_samples on the number of id's given.
34 if (num_samples_ == 0 || num_samples_ > static_cast<int64_t>(indices_.size())) {
35 num_samples_ = static_cast<int64_t>(indices_.size());
36 }
37
38 if (samples_per_tensor_ > num_samples_) {
39 samples_per_tensor_ = num_samples_;
40 }
41
42 is_initialized = true;
43 return Status::OK();
44 }
45
46 // Reset the internal variable to the initial state.
ResetSampler()47 Status SubsetSamplerRT::ResetSampler() {
48 // Reset the internal counters.
49 sample_id_ = 0;
50
51 if (HasChildSampler()) {
52 RETURN_IF_NOT_OK(child_[0]->ResetSampler());
53 }
54
55 return Status::OK();
56 }
57
58 // Get the sample ids.
GetNextSample(TensorRow * out)59 Status SubsetSamplerRT::GetNextSample(TensorRow *out) {
60 // All samples have been drawn
61 if (sample_id_ == num_samples_) {
62 (*out) = TensorRow(TensorRow::kFlagEOE);
63 } else {
64 if (HasChildSampler()) {
65 RETURN_IF_NOT_OK(child_[0]->GetNextSample(&child_ids_));
66 }
67
68 std::shared_ptr<Tensor> outputIds;
69
70 int64_t last_id = sample_id_ + samples_per_tensor_;
71 // Handling the return all samples at once, and when last draw is not a full batch.
72 if (last_id > num_samples_) {
73 last_id = num_samples_;
74 }
75
76 // Allocate tensor
77 RETURN_IF_NOT_OK(CreateSamplerTensor(&outputIds, last_id - sample_id_));
78
79 // Initialize tensor
80 auto id_ptr = outputIds->begin<int64_t>();
81 while (sample_id_ < last_id) {
82 if (indices_[sample_id_] >= num_rows_ || indices_[sample_id_] < 0) {
83 std::string err_msg = "Sample ID (" + std::to_string(indices_[sample_id_]) +
84 ") is out of bound, expected range [0, " + std::to_string(num_rows_ - 1) + "]";
85 RETURN_STATUS_UNEXPECTED(err_msg);
86 }
87
88 int64_t sampled_id = ((indices_[sample_id_] % num_rows_) + num_rows_) % num_rows_;
89 if (HasChildSampler()) {
90 RETURN_IF_NOT_OK(GetAssociatedChildId(&sampled_id, sampled_id));
91 }
92
93 *id_ptr = sampled_id;
94 ++id_ptr;
95 sample_id_++;
96 }
97
98 (*out) = {outputIds};
99 }
100
101 return Status::OK();
102 }
103
SamplerPrint(std::ostream & out,bool show_all) const104 void SubsetSamplerRT::SamplerPrint(std::ostream &out, bool show_all) const {
105 out << "\nSampler: SubsetSampler";
106 if (show_all) {
107 // Call the super class for displaying any common detailed info
108 SamplerRT::SamplerPrint(out, show_all);
109 // Then add our own info if any
110 }
111 }
112
to_json(nlohmann::json * out_json)113 Status SubsetSamplerRT::to_json(nlohmann::json *out_json) {
114 nlohmann::json args;
115 RETURN_IF_NOT_OK(SamplerRT::to_json(&args));
116 args["sampler_name"] = "SubsetSampler";
117 args["indices"] = indices_;
118
119 *out_json = args;
120 return Status::OK();
121 }
122
CalculateNumSamples(int64_t num_rows)123 int64_t SubsetSamplerRT::CalculateNumSamples(int64_t num_rows) {
124 int64_t child_num_rows = num_rows;
125 if (!child_.empty()) {
126 child_num_rows = child_[0]->CalculateNumSamples(num_rows);
127 // return -1 if child_num_rows is undetermined
128 if (child_num_rows == -1) return child_num_rows;
129 }
130 int64_t res = (num_samples_ > 0) ? std::min(child_num_rows, num_samples_) : child_num_rows;
131 res = std::min(res, static_cast<int64_t>(indices_.size()));
132 return res;
133 }
134 } // namespace dataset
135 } // namespace mindspore
136