mindspore2022/mindspore/ccsrc/dataset/kernels/data/data_utils.h

/**
 * Copyright 2019 Huawei Technologies Co., Ltd
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 * http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
#ifndef DATASET_KERNELS_DATA_DATA_UTILS_H_
#define DATASET_KERNELS_DATA_DATA_UTILS_H_

#include <memory>
#include <string>
#include <vector>
#include "dataset/core/constants.h"
#include "dataset/core/cv_tensor.h"
#include "dataset/core/data_type.h"
#include "dataset/core/tensor.h"

namespace mindspore {
namespace dataset {
// Returns Onehot encoding of the input tensor.
//          Example: if input=2 and numClasses=3, the output is [0 0 1].
// @param input: Tensor has type DE_UINT64, the non-one hot values are stored
//               along the first dimensions or rows..
//               If the rank of input is not 1 or the type is not DE_UINT64,
//               then it will fail.
// @param output: Tensor. The shape of the output tensor is <input_shape, numClasses>
//                and the type is same as input.
// @param num_classes: Number of classes to.
Status OneHotEncoding(std::shared_ptr<Tensor> input, std::shared_ptr<Tensor> *output, dsize_t num_classes);

Status OneHotEncodingUnsigned(const std::shared_ptr<Tensor> &input, std::shared_ptr<Tensor> *output,
                              dsize_t num_classes, int64_t index);

Status OneHotEncodingSigned(const std::shared_ptr<Tensor> &input, std::shared_ptr<Tensor> *output, dsize_t num_classes,
                            int64_t index);

// Returns a type changed input tensor.
//          Example: if input tensor is float64, the output will the specified dataType. See DataTypes.cpp
// @param input  Tensor
// @param output Tensor. The shape of the output tensor is same as input with the type changed.
// @param data_type: type of data to cast data to
// @note: this operation will do a memcpy and if the value is truncated then precision will be lost

template <typename T>
void CastFrom(const std::shared_ptr<Tensor> &input, std::shared_ptr<Tensor> *output);

template <typename FROM, typename TO>
void Cast(const std::shared_ptr<Tensor> &input, std::shared_ptr<Tensor> *output);

Status ToFloat16(const std::shared_ptr<Tensor> &input, std::shared_ptr<Tensor> *output);

Status TypeCast(const std::shared_ptr<Tensor> &input, std::shared_ptr<Tensor> *output, const DataType &data_type);

// Pad input tensor according pad_shape, need to have same rank.
// Based on the type of the input tensor, PadEndNumeric/String will be called.
// @param std::shared_ptr<Tensor> src - tensor to pad from
// @param std::shared_ptr<Tensor> *dst - return tensor padded
// @param std::vector<dsize_t> pad_shape - shape to pad to
// @param std::shared_ptr<Tensor> pad_val - value to pad with in Tensor format,
// @return - The error code return
Status PadEnd(const std::shared_ptr<Tensor> &src, std::shared_ptr<Tensor> *dst, const std::vector<dsize_t> &pad_shape,
              const std::shared_ptr<Tensor> &pad_val);

// Pad input numeric tensor according pad_shape, need to have same rank.
// @param std::shared_ptr<Tensor> src - tensor to pad from
// @param std::shared_ptr<Tensor> *dst - return tensor padded
// @param std::vector<dsize_t> pad_shape - shape to pad to
// @param float pad_val - value to pad with
// @return - The error code return
Status PadEndNumeric(const std::shared_ptr<Tensor> &src, std::shared_ptr<Tensor> *dst,
                     const std::vector<dsize_t> &pad_shape, float pad_val);

// recursive helper function for padding numric tensors. This function could be very expensive if called on a
// multi-dimensional tensor it is only meant to be called by PadEndNumeric.
// @tparam T - type of tensor and fill value
// @param std::shared_ptr<Tensor> src - Tensor to pad from
// @param std::shared_ptr<Tensor>* dst - Tensor to pad to, return value
// @param std::vector<dsize_t> cur_ind - recursion helper
// @param T pad_val - value to pad tensor with
// @param size_t cur_dim - recursion helper
// @return Status - The error code return
Status PadEndNumericHelper(const std::shared_ptr<Tensor> &src, std::shared_ptr<Tensor> dst,
                           std::vector<dsize_t> cur_ind, size_t cur_dim = 0);

// Pad input string tensor according pad_shape, need to have same rank.
// @param std::shared_ptr<Tensor> src - tensor to pad from
// @param std::shared_ptr<Tensor> *dst - return tensor padded
// @param std::vector<dsize_t> pad_shape - shape to pad to
// @param std::string pad_val - value to pad with
// @return - The error code return
Status PadEndString(const std::shared_ptr<Tensor> &src, std::shared_ptr<Tensor> *dst,
                    const std::vector<dsize_t> &pad_shape, const std::string &pad_val);

// recursive helper function for padding string tensors. This function could be very expensive if called on a
// multi-dimensional tensor it is only meant to be called by PadEndNumeric.
// @tparam T - type of tensor and fill value
// @param std::shared_ptr<Tensor> src - Tensor to pad from
// @param std::shared_ptr<Tensor>* dst - Tensor to pad to, return value
// @param std::vector<dsize_t> cur_ind - recursion helper
// @param std::string pad_val - value to pad tensor with
// @param size_t cur_dim - recursion helper
// @return Status - The error code return
Status PadEndStringHelper(const std::shared_ptr<Tensor> &src, std::vector<std::string> *dst,
                          const TensorShape &dst_shape, std::vector<dsize_t> cur_ind, size_t cur_dim,
                          const std::string &pad_value);
}  // namespace dataset
}  // namespace mindspore

#endif  // DATASET_KERNELS_DATA_DATA_UTILS_H_