Statistics (DDSketch, Histogram)

Namespace: dftracer::utils::utilities::common::statistics

source

class DDSketch

Public Functions

explicit DDSketch(double relative_accuracy = 0.01)

void add(double value, double weight = 1.0)

void merge(const DDSketch &other)

double quantile(double q) const

void reset()

inline std::uint64_t count() const

inline bool empty() const

inline double min() const

inline double max() const

std::size_t memory_usage() const

std::vector<std::uint8_t> serialize() const

void serialize_into(std::vector<std::uint8_t> &buf) const

Public Static Functions

static DDSketch deserialize(const std::uint8_t *data, std::size_t len)

Public Static Attributes

static constexpr int MAX_BINS

source

struct FittedDistribution

Public Members

DistributionKind kind

std::array<double, 3> params

double ks_stat

double log_likelihood

double bic

bool valid

source

struct FittedMixture

Public Members

std::vector<double> weights

std::vector<GmmComponent> components

double log_likelihood

double bic

int iterations

bool converged

bool valid

source

struct GmmComponent

Public Members

double mean

double stddev

source

class Log2Histogram

Log2-scale histogram with 65 fixed bins.

Bin 0: value == 0 Bin k (1 <= k <= 64): values in [2^(k-1), 2^k)

Covers the full uint64_t range (0 to 2^63). Suitable for both microsecond durations and byte sizes.

Public Functions

Log2Histogram() = default

void add(std::uint64_t value, std::uint64_t count = 1)

void merge(const Log2Histogram &other)

double approx_percentile(double p) const

std::string render_ascii(std::size_t max_width, const std::string &unit) const

std::string render_blocks(std::size_t max_width, const std::string &unit, const std::string &indent = "      ") const

std::string to_json() const

std::string to_json_detailed() const

inline std::uint64_t total_count() const

inline const std::array<std::uint64_t, NUM_BINS> &bins() const

Public Static Functions

static Log2Histogram from_json(const std::string &json)

static std::size_t bin_index(std::uint64_t value)

static std::uint64_t bin_lower(std::size_t bin)

static std::uint64_t bin_upper(std::size_t bin)

Public Static Attributes

static constexpr std::size_t NUM_BINS

source

struct MixtureFitOptions

Public Members

int max_iter

double tol

double variance_floor

std::uint64_t seed

source

struct ModelSelection

Public Members

BestModel model

double bic

int free_params

source

class Statistic

Public Functions

Statistic() = default

inline void attach_sketch(std::shared_ptr<const DDSketch> sketch)

inline void update(double value)

inline double quantile(double q) const

inline double min() const

inline double max() const

inline double mean() const

inline std::uint64_t count() const

source

class TimestampHistogram

Public Functions

TimestampHistogram() = default

void add(std::uint64_t timestamp_us)

void merge(const TimestampHistogram &other)

std::uint64_t count_in_range(std::uint64_t ts_start_us, std::uint64_t ts_end_us) const

double selectivity(std::uint64_t ts_start_us, std::uint64_t ts_end_us) const

std::vector<double> expansion_weights(std::uint64_t bucket_start_us, std::uint64_t bucket_end_us, std::size_t num_sub_buckets) const

std::vector<std::uint8_t> serialize() const

inline std::uint64_t total_count() const

inline bool empty() const

inline std::size_t num_bins() const

inline const std::vector<std::pair<std::uint64_t, std::uint64_t>> &bins() const

Public Static Functions

static TimestampHistogram deserialize(const std::uint8_t *data, std::size_t len)

static inline std::uint64_t bin_index(std::uint64_t timestamp_us)

static inline std::uint64_t bin_start_us(std::uint64_t bin_idx)

static inline std::uint64_t bin_end_us(std::uint64_t bin_idx)

Public Static Attributes

static constexpr std::uint64_t BIN_WIDTH_US

Free Functions

source

std::optional<FittedDistribution> dftracer::utils::utilities::common::statistics::best_fit_by_ks(const std::vector<FittedDistribution> &fits)

source

double dftracer::utils::utilities::common::statistics::cdf(const FittedDistribution &fit, double x)

source

double dftracer::utils::utilities::common::statistics::cdf(const FittedMixture &mix, double x)

source

double dftracer::utils::utilities::common::statistics::cdf(const BestModel &m, double x)

source

std::string_view dftracer::utils::utilities::common::statistics::distribution_name(DistributionKind k)

source

std::vector<FittedDistribution> dftracer::utils::utilities::common::statistics::fit_all_single_distributions(const std::vector<double> &data)

source

FittedMixture dftracer::utils::utilities::common::statistics::fit_gaussian_mixture(const std::vector<double> &data, int K, const MixtureFitOptions &options = {})

source

FittedDistribution dftracer::utils::utilities::common::statistics::fit_single_distribution(DistributionKind kind, const std::vector<double> &data)

source

int dftracer::utils::utilities::common::statistics::free_parameter_count(DistributionKind kind)

source

int dftracer::utils::utilities::common::statistics::free_parameter_count(const FittedMixture &mix)

source

Sampler dftracer::utils::utilities::common::statistics::make_sampler(const FittedDistribution &fit, std::optional<double> min_bound = std::nullopt, std::optional<double> max_bound = std::nullopt)

source

Sampler dftracer::utils::utilities::common::statistics::make_sampler(const FittedMixture &mix, std::optional<double> min_bound = std::nullopt, std::optional<double> max_bound = std::nullopt)

source

Sampler dftracer::utils::utilities::common::statistics::make_sampler(const BestModel &m, std::optional<double> min_bound = std::nullopt, std::optional<double> max_bound = std::nullopt)

source

double dftracer::utils::utilities::common::statistics::pdf(const FittedDistribution &fit, double x)

source

double dftracer::utils::utilities::common::statistics::pdf(const FittedMixture &mix, double x)

source

double dftracer::utils::utilities::common::statistics::pdf(const BestModel &m, double x)

source

double dftracer::utils::utilities::common::statistics::quantile(const FittedDistribution &fit, double p)

source

std::optional<ModelSelection> dftracer::utils::utilities::common::statistics::select_best_model(const std::vector<FittedDistribution> &single_fits, const std::vector<FittedMixture> &mixtures)