probkit 0.2.2.dev1

Toolkit for modifying probabilities and shaping curves.

probkit

Toolkit for modifying probabilities and shaping curves.

Features

• Tunable sigmoid curves - Transform distributions with controllable steepness/flatness
• Probability modification - Scale probabilities using ratios with proper mathematical behavior
• Pure functional design - Deterministic functions for precise control
• Random sampling - Convenient random versions for generating samples from curves
• Zero dependencies - Only uses Python standard library
• Robust validation - Input validation and comprehensive error handling
• Well tested - Thoroughly unit tested with edge case coverage

Quick Start

Deterministic Functions

from probkit import ntsig, biased_curve, modified_probability

# Sigmoid-like curve through (0,0), (0.5,0.5), (1,1)
y = ntsig(k=0.5, x=0.3)  # k controls steepness

# Custom curve between any two points
y = biased_curve(k=0.2, a=10, b=100, x=0.7)  # From (0,10) to (1,100)

# Modify probability with a ratio
new_prob = modified_probability(0.3, 1.5)  # Scale 30% by 1.5x
new_prob = modified_probability(0.3, 3, 2)  # Scale 30% by ratio 3/2

Random Sampling

from probkit.sampling import rng

# Optionally set seed for reproducible results
rng.seed(42)

# Sample from curves with random x values
sample = rng.ntsig(k=0.5)
sample = rng.biased_curve(k=0.2, a=10, b=100)

# Generate multiple samples
samples = [rng.ntsig(0.3) for _ in range(1000)]

# Generate random values using the singleton RNG
random_val = rng.random()  # Random float in [0,1)
choices = rng.choices(['a', 'b', 'c'], k=5)  # All random.Random methods available

# Independent RNG instances for parallel work
fork1 = rng.fork()  # Clone current state
spawn1 = rng.spawn(123)  # Fresh RNG with seed 123

# Context managers that don't affect main RNG
with rng.forked() as r:
    values = [r.ntsig(0.5) for _ in range(10)]
with rng.spawned(456) as r:
    reproducible_values = [r.nthsig(0.3) for _ in range(10)]

API Reference

Curve Functions

• ntsig(k, x) - Normalized tunable sigmoid. Negative k is flat (logit-like), positive k is steep (sigmoid-like)
• nthsig(k, x) - Normalized tunable half-sigmoid. Negative k is convex, positive k is concave
• biased_curve(k, a, b, x) - Custom curve between points (0,a) and (1,b) with bias k

Probability Functions

• modified_probability(k, a, b=None) - Scale probability by ratio a (or a/b if b provided) with proper saturation

Random Sampling

• probkit.sampling.rng - Singleton RNG with all random.Random methods plus probkit helpers
• rng.ntsig(k) - Sample from ntsig with random x
• rng.nthsig(k) - Sample from nthsig with random x
• rng.biased_curve(k, a, b) - Sample from biased_curve with random x
• rng.fork() - Clone current RNG state into independent instance
• rng.spawn(seed) - Create fresh RNG instance with specified seed
• rng.forked() - Context manager yielding forked RNG (doesn't affect main state)
• rng.spawned(seed) - Context manager yielding spawned RNG (doesn't affect main state)

Utilities

• clamp(val, min_val, max_val) - Constrain value to range
• transform_range(x, old_range, new_range) - Linear transformation between ranges
• effective_ratio(a, b) - Safe division with edge case handling

Use Cases

• Game development - Procedural generation, difficulty curves, loot tables
• Simulations - Monte Carlo methods, statistical modeling
• Data science - Distribution transformation, probability weighting
• Machine learning - Custom activation functions, data preprocessing

Install

Clone this repo or copy the probkit folder into your project. No external dependencies required.

# Example: install with pip from local folder
pip install .

Testing

Run all tests with:

python -m unittest discover tests -v

Deployment (notes for Taylor)

PyPI is set up to receive releases from the main branch or when tagged with v*. This is accomplished using PyPI OIDC and GitHub Actions. Pushing to main will create a new dev release with automatic version bump. Creating a v* tag will create a production release using that version number.

git tag v0.1.0 && git push --tags

Contributing

Pull requests and suggestions welcome! Open an issue or PR on GitHub.

License

MIT License. See LICENSE file for details.