1 /* Copyright 2016 The TensorFlow Authors. All Rights Reserved. 2 3 Licensed under the Apache License, Version 2.0 (the "License"); 4 you may not use this file except in compliance with the License. 5 You may obtain a copy of the License at 6 7 http://www.apache.org/licenses/LICENSE-2.0 8 9 Unless required by applicable law or agreed to in writing, software 10 distributed under the License is distributed on an "AS IS" BASIS, 11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 See the License for the specific language governing permissions and 13 limitations under the License. 14 ==============================================================================*/ 15 16 // See docs in ../ops/string_ops.cc. 17 18 #include <string> 19 20 #include "tensorflow/core/framework/kernel_def_builder.h" 21 #include "tensorflow/core/framework/op_kernel.h" 22 #include "tensorflow/core/framework/tensor.h" 23 #include "tensorflow/core/lib/core/errors.h" 24 #include "tensorflow/core/lib/core/status.h" 25 #include "tensorflow/core/lib/strings/str_util.h" 26 27 namespace tensorflow { 28 29 namespace { 30 31 std::vector<string> Split(const string& str, const string& delimiter, 32 const bool skipEmpty) { 33 if (!delimiter.empty()) { 34 if (skipEmpty) { 35 return str_util::Split(str, delimiter, str_util::SkipEmpty()); 36 } 37 return str_util::Split(str, delimiter); 38 } 39 std::vector<string> char_vector(str.size()); 40 for (size_t i = 0; i < str.size(); ++i) { 41 char_vector[i] = str[i]; 42 } 43 return char_vector; 44 } 45 46 } // namespace 47 48 class StringSplitOp : public OpKernel { 49 public: 50 explicit StringSplitOp(OpKernelConstruction* context) 51 : OpKernel(context), skip_empty_(true) { 52 bool skip_empty; 53 // By default skip_empty_ is true. We only get the value from attr if it is 54 // available, so that it is backward compatible. 55 if (context->GetAttr("skip_empty", &skip_empty).ok()) { 56 skip_empty_ = skip_empty; 57 } 58 } 59 60 void Compute(OpKernelContext* ctx) override { 61 const Tensor* input_tensor; 62 OP_REQUIRES_OK(ctx, ctx->input("input", &input_tensor)); 63 OP_REQUIRES(ctx, TensorShapeUtils::IsVector(input_tensor->shape()), 64 errors::InvalidArgument("input must be a vector, got shape: ", 65 input_tensor->shape().DebugString())); 66 67 const auto input_vec = input_tensor->vec<string>(); 68 const int64 batch_size = input_vec.dimension(0); 69 70 const Tensor* delimiter_tensor; 71 OP_REQUIRES_OK(ctx, ctx->input("delimiter", &delimiter_tensor)); 72 OP_REQUIRES( 73 ctx, TensorShapeUtils::IsScalar(delimiter_tensor->shape()), 74 errors::InvalidArgument("delimiter must scalar, got shape: ", 75 delimiter_tensor->shape().DebugString())); 76 const auto delimiter_vec = delimiter_tensor->flat<string>(); 77 const string& delimiter = delimiter_vec(0); 78 // Empty delimiter means split the input character by character. 79 std::vector<string> tokens; 80 // Guess that we'll be unpacking a handful of tokens per example. 81 static constexpr int kReserveSize = 4; 82 tokens.reserve(batch_size * kReserveSize); 83 84 int64 output_size = 0; 85 int64 max_num_entries = 0; 86 std::vector<int64> num_indices(batch_size); 87 for (int64 i = 0; i < batch_size; ++i) { 88 std::vector<string> parts = Split(input_vec(i), delimiter, skip_empty_); 89 int64 n_entries = parts.size(); 90 num_indices[i] = n_entries; 91 output_size += n_entries; 92 max_num_entries = std::max(max_num_entries, n_entries); 93 tokens.insert(tokens.end(), parts.begin(), parts.end()); 94 } 95 96 Tensor* sp_indices_t; 97 OP_REQUIRES_OK(ctx, ctx->allocate_output(0, TensorShape({output_size, 2}), 98 &sp_indices_t)); 99 Tensor* sp_tokens_t; 100 OP_REQUIRES_OK( 101 ctx, ctx->allocate_output(1, TensorShape({output_size}), &sp_tokens_t)); 102 Tensor* sp_shape_t; 103 OP_REQUIRES_OK(ctx, ctx->allocate_output(2, TensorShape({2}), &sp_shape_t)); 104 105 auto sp_indices = sp_indices_t->matrix<int64>(); 106 auto sp_tokens = sp_tokens_t->vec<string>(); 107 auto sp_shape = sp_shape_t->vec<int64>(); 108 sp_shape(0) = batch_size; 109 sp_shape(1) = max_num_entries; 110 size_t c = 0; 111 for (size_t i = 0; i < batch_size; ++i) { 112 for (size_t j = 0; j < num_indices[i]; ++j) { 113 sp_indices(c, 0) = i; 114 sp_indices(c, 1) = j; 115 sp_tokens(c) = tokens[c]; 116 ++c; 117 } 118 } 119 } 120 121 private: 122 bool skip_empty_; 123 }; 124 125 REGISTER_KERNEL_BUILDER(Name("StringSplit").Device(DEVICE_CPU), StringSplitOp); 126 127 } // namespace tensorflow 128
