transcrypto 1.7.0__py3-none-any.whl → 2.0.0__py3-none-any.whl

transcrypto/utils/stats.py

@@ -0,0 +1,360 @@
+# SPDX-FileCopyrightText: Copyright 2026 Daniel Balparda <balparda@github.com>
+# SPDX-License-Identifier: Apache-2.0
+"""Balparda's TransCrypto basic statistics library."""
+
+from __future__ import annotations
+
+import math
+from collections import abc
+
+from transcrypto.utils import base
+
+# Lanczos coefficients for g=7, n=9; provides ~15 digit accuracy for gamma
+_LANCZOS_G = 7
+_LANCZOS_COEFF: tuple[float, ...] = (
+  0.99999999999980993,
+  676.5203681218851,
+  -1259.1392167224028,
+  771.32342877765313,
+  -176.61502916214059,
+  12.507343278686905,
+  -0.13857109526572012,
+  9.9843695780195716e-6,
+  1.5056327351493116e-7,
+)
+_TINY: float = 1e-30
+_NSG: abc.Callable[[float], float] = (
+  lambda z: _TINY if abs(z) < _TINY else z  # numerical stability guard
+)
+_BETA_INCOMPLETE_MAX_ITER: int = 200
+_BETA_INCOMPLETE_TOL: float = 1e-14
+_STUDENT_SMALL: float = 1e-12
+
+
+def GammaLanczos(z: float, /) -> float:
+  """Compute the gamma function Γ(z) using the Lanczos approximation.
+
+  The Lanczos approximation provides an efficient method to compute
+  the gamma function with high accuracy (~15 digits). It uses the
+  reflection formula for z < 0.5.
+
+  Args:
+    z (float): Input value. For z ≤ 0 where z is a non-positive integer,
+      the function will return ±inf.
+
+  Returns:
+    float: Γ(z), the gamma function evaluated at z.
+
+  Notes:
+    - Uses coefficients optimized for g=7, n=9.
+    - For z < 0.5, uses the reflection formula:
+        Γ(z) = π / (sin(πz) · Γ(1-z))
+
+  """
+  if z < 0.5:  # noqa: PLR2004
+    # Reflection formula: Γ(z) = π / (sin(πz) Γ(1-z))
+    return math.pi / (math.sin(math.pi * z) * GammaLanczos(1.0 - z))
+  z -= 1.0
+  x: float = _LANCZOS_COEFF[0]
+  for i in range(1, len(_LANCZOS_COEFF)):
+    x += _LANCZOS_COEFF[i] / (z + i)
+  t: float = z + _LANCZOS_G + 0.5
+  tz: float = t ** (z + 0.5)
+  return math.sqrt(2.0 * math.pi) * tz * math.exp(-t) * x
+
+
+def BetaIncompleteCF(a: float, b: float, x: float, /) -> float:
+  """Compute continued fraction for the regularized incomplete beta function.
+
+  Uses the modified Lentz algorithm to evaluate the continued fraction
+  expansion of I_x(a, b) efficiently and stably.
+
+  Args:
+    a (float): First shape parameter (> 0).
+    b (float): Second shape parameter (> 0).
+    x (float): Point at which to evaluate (0 ≤ x ≤ 1).
+
+  Returns:
+    float: The continued fraction value.
+
+  Notes:
+    - Internal helper for `_BetaIncomplete`.
+    - Convergence is typically achieved in < 100 iterations for typical inputs.
+    - Uses a floor of 1e-30 to prevent division by zero.
+
+  """
+  qab: float = a + b
+  qap: float = a + 1.0
+  qam: float = a - 1.0
+  c: float = 1.0
+  d: float = 1.0 / _NSG(1.0 - qab * x / qap)
+  h: float = d
+  aa: float
+  delta: float
+  m2: int
+  for m in range(1, _BETA_INCOMPLETE_MAX_ITER + 1):
+    m2 = 2 * m
+    # even step
+    aa = m * (b - m) * x / ((qam + m2) * (a + m2))
+    c, d = _NSG(1.0 + aa / c), 1.0 / _NSG(1.0 + aa * d)
+    h *= d * c
+    # odd step
+    aa = -(a + m) * (qab + m) * x / ((a + m2) * (qap + m2))
+    c, d = _NSG(1.0 + aa / c), 1.0 / _NSG(1.0 + aa * d)
+    delta = d * c
+    h *= delta
+    if abs(delta - 1.0) < _BETA_INCOMPLETE_TOL:
+      break
+  return h
+
+
+def BetaIncomplete(a: float, b: float, x: float, /) -> float:
+  """Compute the regularized incomplete beta function I_x(a, b).
+
+  The regularized incomplete beta function is defined as:
+    I_x(a, b) = B(x; a, b) / B(a, b)
+  where B(x; a, b) is the incomplete beta function and B(a, b) is the
+  complete beta function.
+
+  Args:
+    a (float): First shape parameter (> 0).
+    b (float): Second shape parameter (> 0).
+    x (float): Upper limit of integration (0 ≤ x ≤ 1).
+
+  Returns:
+    float: I_x(a, b), the regularized incomplete beta at x.
+
+  Raises:
+    base.InputError: If x is outside [0, 1].
+
+  Notes:
+    - Uses continued fraction expansion with Lentz algorithm.
+    - For numerical stability, uses the symmetry relation when
+      x > (a + 1) / (a + b + 2).
+
+  """
+  if x < 0.0 or x > 1.0:
+    raise base.InputError(f'x must be in [0, 1], got {x}')
+  if x == 0.0:
+    return 0.0
+  if x == 1.0:
+    return 1.0
+  log_beta: float = math.lgamma(a) + math.lgamma(b) - math.lgamma(a + b)
+  front: float = math.exp(math.log(x) * a + math.log(1.0 - x) * b - log_beta) / a
+  if x < (a + 1.0) / (a + b + 2.0):
+    return front * BetaIncompleteCF(a, b, x)
+  return 1.0 - front * BetaIncompleteCF(b, a, 1.0 - x) * a / b
+
+
+def StudentTCDF(t_val: float, df: float, /) -> float:
+  """Compute the cumulative distribution function (CDF) of Student's t-distribution.
+
+  The CDF gives the probability P(T ≤ t) where T follows a t-distribution
+  with `df` degrees of freedom.
+
+  Args:
+    t_val (float): The t-statistic value.
+    df (float): Degrees of freedom (> 0).
+
+  Returns:
+    float: Probability P(T ≤ t_val), in range [0, 1].
+
+  Notes:
+    - Uses the relationship between the t-distribution CDF and the
+      regularized incomplete beta function.
+    - For t ≥ 0: CDF = 0.5 + 0.5 * (1 - I_x(df/2, 0.5))
+    - For t < 0: CDF = 0.5 * I_x(df/2, 0.5)
+    - where x = df / (df + t²)
+
+  """
+  x: float = df / (df + t_val * t_val)
+  prob: float = 0.5 * BetaIncomplete(df / 2.0, 0.5, x)
+  return 0.5 + (0.5 - prob) if t_val >= 0 else prob
+
+
+def StudentTPPF(q: float, df: float, /) -> float:
+  """Compute the percent point function (inverse CDF) of Student's t-distribution.
+
+  Given a probability q, find the value t such that P(T ≤ t) = q,
+  where T follows a t-distribution with `df` degrees of freedom.
+
+  Args:
+    q (float): Probability (0 < q < 1).
+    df (float): Degrees of freedom (> 0).
+
+  Returns:
+    float: The t-value such that CDF(t) = q.
+
+  Raises:
+    base.InputError: If q is not in (0, 1).
+
+  Notes:
+    - Uses Newton-Raphson iteration with an initial guess from
+      the normal distribution approximation.
+    - Converges to ~12 decimal places in typical cases.
+
+  """
+  if not 0.0 < q < 1.0:
+    raise base.InputError(f'q must be in (0, 1), got {q}')
+  # Special case: q=0.5 is exactly 0 by symmetry
+  if q == 0.5:  # noqa: PLR2004
+    return 0.0
+  # Initial guess using inverse normal approximation (Abramowitz & Stegun 26.2.23)
+  if q < 0.5:  # noqa: PLR2004
+    sign: float = -1.0
+    p: float = q
+  else:
+    sign = 1.0
+    p = 1.0 - q
+  # Protect against log(0) when p is very close to 0
+  p = max(p, 1e-300)
+  t_approx: float = math.sqrt(-2.0 * math.log(p))
+  c0 = 2.515517
+  c1 = 0.802853
+  c2 = 0.010328
+  d1 = 1.432788
+  d2 = 0.189269
+  d3 = 0.001308
+  x0: float = sign * (
+    t_approx
+    - (c0 + c1 * t_approx + c2 * t_approx**2)
+    / (1 + d1 * t_approx + d2 * t_approx**2 + d3 * t_approx**3)
+  )
+  # Newton-Raphson refinement
+  for _ in range(50):
+    cdf_val: float = StudentTCDF(x0, df)
+    # PDF of Student's t-distribution (computed in log-space to avoid overflow for large df)
+    log_pdf: float = (
+      math.lgamma((df + 1) / 2)
+      - 0.5 * math.log(df * math.pi)
+      - math.lgamma(df / 2)
+      - ((df + 1) / 2) * math.log(1 + x0**2 / df)
+    )
+    pdf_val: float = _NSG(math.exp(log_pdf))
+    x1: float = x0 - (cdf_val - q) / pdf_val
+    if abs(x1 - x0) < _STUDENT_SMALL:
+      return x1
+    x0 = x1
+  return x0  # pragma: no cover - Newton-Raphson always converges for t-distribution
+
+
+def SampleVariance(data: list[int | float], mean: float, /) -> float:
+  """Compute sample variance with Bessel's correction (n-1 denominator).
+
+  Args:
+    data (list[int | float]): Sequence of numeric measurements, with len(data) >= 2.
+    mean (float): Pre-computed mean of the data.
+
+  Returns:
+    float: Sample variance s² = Σ(xᵢ - x̄)² / (n - 1).
+
+  Raises:
+    base.InputError: If len(data) < 2.
+
+  """
+  if (data_sz := len(data)) < 2:  # noqa: PLR2004
+    raise base.InputError(f'sample variance requires at least 2 data points, got {data_sz}')
+  return sum((x - mean) ** 2 for x in data) / float(data_sz - 1)
+
+
+def StandardErrorOfMean(data: list[int | float], /) -> tuple[float, float]:
+  """Compute the mean and standard error of the mean (SEM).
+
+  The SEM is the standard deviation of the sampling distribution of the
+  sample mean, computed as s / √n where s is the sample standard deviation.
+
+  Args:
+    data (list[int | float]): Sequence of numeric measurements (n >= 2).
+
+  Returns:
+    tuple[float, float]: (mean, SEM) where:
+      - mean: arithmetic mean of the data
+      - SEM: standard error of the mean (σ / √n)
+
+  Notes:
+    - Assumes len(data) >= 2; returns (mean, inf) for single element handled by caller.
+    - Uses sample standard deviation (Bessel's correction).
+
+  """  # noqa: RUF002
+  n: int = len(data)
+  mean: float = sum(data) / n
+  variance: float = SampleVariance(data, mean)
+  return (mean, math.sqrt(variance / n))
+
+
+def StudentTInterval(
+  confidence: float, df: int, loc: float, scale: float, /
+) -> tuple[float, float]:
+  """Compute a symmetric confidence interval using Student's t-distribution.
+
+  Args:
+    confidence (float): Confidence level (e.g., 0.95 for 95% CI).
+    df (int): Degrees of freedom (n - 1 for a sample of size n).
+    loc (float): Center of the interval (typically the sample mean).
+    scale (float): Scale parameter (typically the SEM).
+
+  Returns:
+    tuple[float, float]: (lower_bound, upper_bound) of the confidence interval.
+
+  Notes:
+    - The interval is symmetric around `loc`:
+        [loc - t_crit * scale, loc + t_crit * scale]
+    - where t_crit is the critical t-value for the given confidence and df.
+
+  """
+  alpha: float = 1.0 - confidence
+  t_crit: float = StudentTPPF(1.0 - alpha / 2.0, df)
+  margin: float = t_crit * scale
+  return (loc - margin, loc + margin)
+
+
+def MeasurementStats(
+  data: list[int | float], /, *, confidence: float = 0.95
+) -> tuple[int, float, float, float, tuple[float, float], float]:
+  """Compute descriptive statistics for repeated measurements.
+
+  Given N ≥ 1 measurements, this function computes the sample mean, the
+  standard error of the mean (SEM), and the symmetric error estimate for
+  the chosen confidence interval using Student's t distribution.
+
+  Notes:
+    • If only one measurement is given, SEM and error are reported as +∞ and
+      the confidence interval is (-∞, +∞).
+    • This function assumes the underlying distribution is approximately
+      normal, or n is large enough for the Central Limit Theorem to apply.
+
+  Args:
+    data (list[int | float]): Sequence of numeric measurements.
+    confidence (float, optional): Confidence level for the interval, 0.5 <= confidence < 1;
+        defaults to 0.95 (95% confidence interval).
+
+  Returns:
+    tuple:
+      - n (int): number of measurements.
+      - mean (float): arithmetic mean of the data
+      - sem (float): standard error of the mean, sigma / √n
+      - error (float): half-width of the confidence interval (mean ± error)
+      - ci (tuple[float, float]): lower and upper confidence interval bounds
+      - confidence (float): the confidence level used
+
+  Raises:
+    base.InputError: if the input list is empty.
+
+  """
+  # test inputs
+  n: int = len(data)
+  if not n:
+    raise base.InputError('no data')
+  if not 0.5 <= confidence < 1.0:  # noqa: PLR2004
+    raise base.InputError(f'invalid confidence: {confidence=}')
+  # solve trivial case
+  if n == 1:
+    return (n, float(data[0]), math.inf, math.inf, (-math.inf, math.inf), confidence)
+  # compute statistics using local implementation (no scipy/numpy dependency)
+  mean: float
+  sem: float
+  mean, sem = StandardErrorOfMean(data)
+  ci: tuple[float, float] = StudentTInterval(confidence, n - 1, mean, sem)
+  t_crit: float = StudentTPPF((1.0 + confidence) / 2.0, n - 1)
+  error: float = t_crit * sem  # half-width of the CI
+  return (n, mean, sem, error, ci, confidence)

transcrypto/utils/timer.py

@@ -0,0 +1,175 @@
+# SPDX-FileCopyrightText: Copyright 2026 Daniel Balparda <balparda@github.com>
+# SPDX-License-Identifier: Apache-2.0
+"""Balparda's TransCrypto timer library."""
+
+from __future__ import annotations
+
+import datetime
+import functools
+import logging
+import time
+from collections import abc
+from types import TracebackType
+from typing import Self
+
+from transcrypto.utils import base, human
+
+# Time utils
+
+MIN_TM = int(datetime.datetime(2000, 1, 1, 0, 0, 0, tzinfo=datetime.UTC).timestamp())
+TIME_FORMAT = '%Y/%b/%d-%H:%M:%S-UTC'
+TimeStr: abc.Callable[[int | float | None], str] = lambda tm: (
+  time.strftime(TIME_FORMAT, time.gmtime(tm)) if tm else '-'
+)
+Now: abc.Callable[[], int] = lambda: int(time.time())
+StrNow: abc.Callable[[], str] = lambda: TimeStr(Now())
+
+
+class Timer:
+  """An execution timing class that can be used as both a context manager and a decorator.
+
+  Examples:
+    # As a context manager
+    with Timer('Block timing'):
+      time.sleep(1.2)
+
+    # As a decorator
+    @Timer('Function timing')
+    def slow_function():
+      time.sleep(0.8)
+
+    # As a regular object
+    tm = Timer('Inline timing')
+    tm.Start()
+    time.sleep(0.1)
+    tm.Stop()
+    print(tm)
+
+  Attributes:
+    label (str, optional): Timer label
+    emit_log (bool, optional): If True (default) will logging.info() the timer, else will not
+    emit_print (bool, optional): If True will print() the timer, else (default) will not
+
+  """
+
+  def __init__(
+    self,
+    label: str = '',
+    /,
+    *,
+    emit_log: bool = True,
+    emit_print: abc.Callable[[str], None] | None = None,
+  ) -> None:
+    """Initialize the Timer.
+
+    Args:
+      label (str, optional): A description or name for the timed block or function
+      emit_log (bool, optional): Emit a log message when finished; default is True
+      emit_print (Callable[[str], None] | None, optional): Emit a print() message when
+          finished using the provided callable; default is None
+
+    """
+    self.emit_log: bool = emit_log
+    self.emit_print: abc.Callable[[str], None] | None = emit_print
+    self.label: str = label.strip()
+    self.start: float | None = None
+    self.end: float | None = None
+
+  @property
+  def elapsed(self) -> float:
+    """Elapsed time. Will be zero until a measurement is available with start/end.
+
+    Raises:
+        base.Error: negative elapsed time
+
+    Returns:
+        float: elapsed time, in seconds
+
+    """
+    if self.start is None or self.end is None:
+      return 0.0
+    delta: float = self.end - self.start
+    if delta <= 0.0:
+      raise base.Error(f'negative/zero delta: {delta}')
+    return delta
+
+  def __str__(self) -> str:
+    """Get current timer value.
+
+    Returns:
+        str: human-readable representation of current time value
+
+    """
+    if self.start is None:
+      return f'{self.label}: <UNSTARTED>' if self.label else '<UNSTARTED>'
+    if self.end is None:
+      return (
+        f'{self.label}: ' if self.label else ''
+      ) + f'<PARTIAL> {human.HumanizedSeconds(time.perf_counter() - self.start)}'
+    return (f'{self.label}: ' if self.label else '') + f'{human.HumanizedSeconds(self.elapsed)}'
+
+  def Start(self) -> None:
+    """Start the timer.
+
+    Raises:
+        base.Error: if you try to re-start the timer
+
+    """
+    if self.start is not None:
+      raise base.Error('Re-starting timer is forbidden')
+    self.start = time.perf_counter()
+
+  def __enter__(self) -> Self:
+    """Start the timer when entering the context.
+
+    Returns:
+        Timer: context object (self)
+
+    """
+    self.Start()
+    return self
+
+  def Stop(self) -> None:
+    """Stop the timer and emit logging.info with timer message.
+
+    Raises:
+        base.Error: trying to re-start timer or stop unstarted timer
+
+    """
+    if self.start is None:
+      raise base.Error('Stopping an unstarted timer')
+    if self.end is not None:
+      raise base.Error('Re-stopping timer is forbidden')
+    self.end = time.perf_counter()
+    message: str = str(self)
+    if self.emit_log:
+      logging.info(message)
+    if self.emit_print is not None:
+      self.emit_print(message)
+
+  def __exit__(
+    self,
+    unused_exc_type: type[BaseException] | None,
+    unused_exc_val: BaseException | None,
+    exc_tb: TracebackType | None,
+  ) -> None:
+    """Stop the timer when exiting the context."""
+    self.Stop()
+
+  def __call__[**F, R](self, func: abc.Callable[F, R]) -> abc.Callable[F, R]:
+    """Allow the Timer to be used as a decorator.
+
+    Args:
+      func: The function to time.
+
+    Returns:
+      The wrapped function with timing behavior.
+
+    """
+
+    @functools.wraps(func)
+    def _Wrapper(*args: F.args, **kwargs: F.kwargs) -> R:
+      with self.__class__(self.label, emit_log=self.emit_log, emit_print=self.emit_print):
+        return func(*args, **kwargs)
+
+    return _Wrapper