scientific-computing-system-2.0

CDS v2 is a scientific computing platform built on the scientific Python stack — NumPy, SciPy, pandas and matplotlib. The algorithms proven in the pure-Python scientific-computing-system (v1.x) form its foundation; v2 rebuilds them for speed and adds new domain modules on top.

Installation

pip install scientific-computing-system-2.0

From source:

git clone https://github.com/Furox-Art/scientific-computing-system-2.0.git
cd scientific-computing-system-2.0
pip install -e .[dev]

Quick start

import numpy as np
import cds2

# Linear algebra
A = [[3.0, 1.0], [1.0, 2.0]]
b = [9.0, 8.0]
x = cds2.linalg.solve(A, b)

# Statistics
r = cds2.stats.independent_t_test([1, 2, 3, 4, 5], [3, 4, 5, 6, 7])

# Optimization
res = cds2.optimize.minimize(lambda v: (v[0] - 2) ** 2 + (v[1] + 1) ** 2, x0=[0.0, 0.0])
print(res.x)  # ~ [2.0, -1.0]

# Signals
freqs, psd = cds2.signals.power_spectrum(np.sin(np.linspace(0, 100, 1024)), fs=256.0)

# Graphs with PageRank
adj = cds2.graph.from_edges(4, [(0, 1), (0, 2), (1, 3), (2, 3)], directed=True)
scores = cds2.graph.pagerank(adj)

# Information theory
h = cds2.infotheory.entropy([0.25, 0.25, 0.25, 0.25])
mi = cds2.infotheory.mutual_information([[0.5, 0.0], [0.0, 0.5]])

# Chaos / nonlinear dynamics
series = cds2.chaos.logistic_map(3.99, length=400, seed=1)
lyap = cds2.chaos.largest_lyapunov_exponent(series)

# Bayesian conjugate updates
post = cds2.bayes.beta_binomial_update(successes=7, failures=3)
print(post.mean)  # 0.7

# Metaheuristics
res = cds2.metaheuristics.pso_minimize(lambda v: (v[0] - 3) ** 2, [(-10, 10)], seed=1)

# Geometry
area = cds2.geometry.hull_area([(0, 0), (1, 0), (0, 1)])

# Reinforcement learning
q_values, returns = cds2.rl.q_learn(cds2.rl.GridWorld(4, 4), episodes=300, seed=1)

Modules

Facade vs Real Capability

Some cds2.* modules are thin convenience wrappers that only coerce args and unify return types; others contain CDS-native science with no SciPy equivalent. Knowing which is which tells you when cds2 saves time and when to import upstream directly.

¹ Deprecated since 4.3.0, removed in 5.0.0 — DeprecationWarning. ² cds2.scientific is 100% native (CODATA CONSTANTS, physics formulas, convert_units). ³ cds2.sparse already has real value: jacobi_preconditioner/ilu_preconditioner → LinearOperator, residual_norm diagnostics.

# Convenience re-export — use SciPy
from scipy import special as sps
from scipy import stats

sps.gamma([0.5, 1, 2])
stats.norm.pdf(0.0, loc=0, scale=1)

# CDS-native — use cds2
from cds2 import sde

ens = sde.sde_milstein(
    lambda y, t: 0.05 * y,
    lambda y, t: 0.20 * y,
    y0=[100.0],
    t_span=(0, 1),
    dt=1e-3,
    n_paths=8192,
    seed=0,
)

CLI

cds2 info
cds2 stats 1,2,3,4,5
cds2 integrate sin --a 0 --b 3.14159
cds2 linsolve --a "3,1;1,2" --b "9,8"
cds2 entropy "0.25,0.25,0.25,0.25"
cds2 units 5 --from-unit km --to-unit mile
cds2 solve --coeffs "1,-5,6"
cds2 plot 1,3,2,5,4 --file out.png

Relationship to CDS v1.x

The original zero-dependency pure-Python line lives at Furox-Art/scientific-computing-system and remains available. v2 is an independent project that trades that constraint for the speed and breadth of the scientific Python ecosystem.

Runnable case studies live in examples/ - see the docs page for details.

Benchmarks

cds2 races the scientific stack head-to-head — and ships its own compiled C kernels where they help. Current scoreboard (full methodology in docs/benchmarks.md):

Wrapper APIs hold parity with raw NumPy/SciPy; the KMeans Lloyd loop and PageRank power iteration are from-scratch C extensions (cds2._fast_kmeans, cds2._fast_pagerank) that beat the specialist libraries. A pure-Python fallback wheel keeps compiler-less installs working.

python benchmarks/run_benchmarks.py            # full run
python benchmarks/run_benchmarks.py --quick    # smoke run

Development

pip install -e .[dev]
pytest            # run the test suite
ruff check .      # lint

License

MIT — see LICENSE.