Add COVER dictionary builder

This commit is contained in:
Nick Terrell 2016-12-31 21:07:44 -08:00
parent 8d7432914f
commit 96b39f65fa
2 changed files with 1093 additions and 0 deletions

1041
lib/dictBuilder/cover.c Normal file

File diff suppressed because it is too large Load Diff

View File

@ -86,6 +86,58 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_advanced(void* dictBuffer, size_t dict
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
ZDICT_params_t parameters);
/*! COVER_params_t :
For all values 0 means default.
kMin and d are the only required parameters.
*/
typedef struct {
unsigned d; /* dmer size : constraint: <= kMin : Should probably be in the range [6, 16]. */
unsigned kMin; /* Minimum segment size : constraint: > 0 */
unsigned kStep; /* Try kStep segment lengths uniformly distributed in the range [kMin, kMax] : 0 (default) only if kMax == 0 */
unsigned kMax; /* Maximum segment size : 0 = kMin (default) : constraint : 0 or >= kMin */
unsigned smoothing; /* Higher smoothing => larger segments are selected. Only useful if kMax > kMin. */
unsigned notificationLevel; /* Write to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
unsigned dictID; /* 0 means auto mode (32-bits random value); other : force dictID value */
int compressionLevel; /* 0 means default; target a specific zstd compression level */
} COVER_params_t;
/*! COVER_trainFromBuffer() :
Train a dictionary from an array of samples using the COVER algorithm.
Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
The resulting dictionary will be saved into `dictBuffer`.
@return : size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
or an error code, which can be tested with ZDICT_isError().
Tips : In general, a reasonable dictionary has a size of ~ 100 KB.
It's obviously possible to target smaller or larger ones, just by specifying different `dictBufferCapacity`.
In general, it's recommended to provide a few thousands samples, but this can vary a lot.
It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
*/
ZDICTLIB_API size_t COVER_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
COVER_params_t parameters);
/*! COVER_optimizeTrainFromBuffer() :
The same requirements as above hold for all the parameters except `parameters`.
This function tries many parameter combinations and picks the best parameters.
`*parameters` is filled with the best parameters found, and the dictionary
constructed with those parameters is stored in `dictBuffer`.
All of the {d, kMin, kStep, kMax} are optional, and smoothing is ignored.
If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8, 10, 12, 14, 16}.
If kStep is non-zero then it is used, otherwise we pick 8.
If kMin and kMax are non-zero, then they limit the search space for kMin and kMax,
otherwise we check kMin and kMax values in the range [32, 1024].
@return : size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
or an error code, which can be tested with ZDICT_isError().
On success `*parameters` contains the parameters selected.
*/
ZDICTLIB_API size_t COVER_optimizeTrainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
const void* samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
COVER_params_t *parameters);
/*! ZDICT_finalizeDictionary() :