tesseract  4.00.00dev
tesseract::LSTMTester Class Reference

#include <lstmtester.h>

Public Member Functions

 LSTMTester (inT64 max_memory)
 
bool LoadAllEvalData (const STRING &filenames_file)
 
bool LoadAllEvalData (const GenericVector< STRING > &filenames)
 
STRING RunEvalAsync (int iteration, const double *training_errors, const GenericVector< char > &model_data, int training_stage)
 
STRING RunEvalSync (int iteration, const double *training_errors, const GenericVector< char > &model_data, int training_stage)
 

Detailed Description

Definition at line 29 of file lstmtester.h.

Constructor & Destructor Documentation

◆ LSTMTester()

tesseract::LSTMTester::LSTMTester ( inT64  max_memory)

Definition at line 24 of file lstmtester.cpp.

25  : test_data_(max_memory), total_pages_(0), async_running_(false) {}

Member Function Documentation

◆ LoadAllEvalData() [1/2]

bool tesseract::LSTMTester::LoadAllEvalData ( const STRING filenames_file)

Definition at line 30 of file lstmtester.cpp.

30  {
31  GenericVector<STRING> filenames;
32  if (!LoadFileLinesToStrings(filenames_file, &filenames)) {
33  tprintf("Failed to load list of eval filenames from %s\n",
34  filenames_file.string());
35  return false;
36  }
37  return LoadAllEvalData(filenames);
38 }
#define tprintf(...)
Definition: tprintf.h:31
const char * string() const
Definition: strngs.cpp:198
bool LoadFileLinesToStrings(const STRING &filename, GenericVector< STRING > *lines)
bool LoadAllEvalData(const STRING &filenames_file)
Definition: lstmtester.cpp:30

◆ LoadAllEvalData() [2/2]

bool tesseract::LSTMTester::LoadAllEvalData ( const GenericVector< STRING > &  filenames)

Definition at line 43 of file lstmtester.cpp.

43  {
44  test_data_.Clear();
45  bool result = test_data_.LoadDocuments(filenames, CS_SEQUENTIAL, nullptr);
46  total_pages_ = test_data_.TotalPages();
47  return result;
48 }
bool LoadDocuments(const GenericVector< STRING > &filenames, CachingStrategy cache_strategy, FileReader reader)
Definition: imagedata.cpp:572

◆ RunEvalAsync()

STRING tesseract::LSTMTester::RunEvalAsync ( int  iteration,
const double *  training_errors,
const GenericVector< char > &  model_data,
int  training_stage 
)

Definition at line 52 of file lstmtester.cpp.

54  {
55  STRING result;
56  if (total_pages_ == 0) {
57  result.add_str_int("No test data at iteration", iteration);
58  return result;
59  }
60  if (!LockIfNotRunning()) {
61  result.add_str_int("Previous test incomplete, skipping test at iteration",
62  iteration);
63  return result;
64  }
65  // Save the args.
66  STRING prev_result = test_result_;
67  test_result_ = "";
68  if (training_errors != nullptr) {
69  test_iteration_ = iteration;
70  test_training_errors_ = training_errors;
71  test_model_data_ = model_data;
72  test_training_stage_ = training_stage;
73  SVSync::StartThread(&LSTMTester::ThreadFunc, this);
74  } else {
75  UnlockRunning();
76  }
77  return prev_result;
78 }
void add_str_int(const char *str, int number)
Definition: strngs.cpp:381
Definition: strngs.h:45
static void StartThread(void *(*func)(void *), void *arg)
Create new thread.
Definition: svutil.cpp:87

◆ RunEvalSync()

STRING tesseract::LSTMTester::RunEvalSync ( int  iteration,
const double *  training_errors,
const GenericVector< char > &  model_data,
int  training_stage 
)

Definition at line 82 of file lstmtester.cpp.

84  {
85  LSTMTrainer trainer;
86  if (!trainer.ReadTrainingDump(model_data, &trainer)) {
87  return "Deserialize failed";
88  }
89  int eval_iteration = 0;
90  double char_error = 0.0;
91  double word_error = 0.0;
92  int error_count = 0;
93  while (error_count < total_pages_) {
94  const ImageData* trainingdata = test_data_.GetPageBySerial(eval_iteration);
95  trainer.SetIteration(++eval_iteration);
96  NetworkIO fwd_outputs, targets;
97  if (trainer.PrepareForBackward(trainingdata, &fwd_outputs, &targets) !=
98  UNENCODABLE) {
99  char_error += trainer.NewSingleError(tesseract::ET_CHAR_ERROR);
100  word_error += trainer.NewSingleError(tesseract::ET_WORD_RECERR);
101  ++error_count;
102  }
103  }
104  char_error *= 100.0 / total_pages_;
105  word_error *= 100.0 / total_pages_;
106  STRING result;
107  result.add_str_int("At iteration ", iteration);
108  result.add_str_int(", stage ", training_stage);
109  result.add_str_double(", Eval Char error rate=", char_error);
110  result.add_str_double(", Word error rate=", word_error);
111  return result;
112 }
const ImageData * GetPageBySerial(int serial)
Definition: imagedata.h:335
void add_str_int(const char *str, int number)
Definition: strngs.cpp:381
Definition: strngs.h:45
void add_str_double(const char *str, double number)
Definition: strngs.cpp:391

The documentation for this class was generated from the following files: