pyscivex 0.1.0

The complete data science toolkit — powered by Rust

pyscivex

The complete data science toolkit — powered by Rust.

One pip install pyscivex replaces numpy, pandas, scipy, scikit-learn, matplotlib, and more — with Rust performance and zero external Python dependencies.

---

Quick Start

import pyscivex as sv

# Tensors (numpy replacement)
a = sv.Tensor([[1.0, 2.0], [3.0, 4.0]])
b = sv.Tensor.ones([2, 2])
c = a @ b
print(c.shape())  # [2, 2]

# DataFrames (pandas replacement)
df = sv.DataFrame({"x": [1.0, 2.0, 3.0], "y": [4.0, 5.0, 6.0]})
print(df.shape())  # (3, 2)

# Statistics (scipy.stats replacement)
data = [2.0, 4.0, 4.0, 4.0, 5.0, 5.0, 7.0, 9.0]
print(sv.mean(data))      # 5.0
print(sv.std_dev(data))   # ~2.0

# Machine Learning (scikit-learn replacement)
model = sv.ml.LinearRegression()
model.fit(sv.Tensor([[1.0], [2.0], [3.0]]), sv.Tensor([2.0, 4.0, 6.0]))
pred = model.predict(sv.Tensor([[4.0]]))

# Visualization (matplotlib replacement)
fig = sv.Figure()
fig.line_plot([1, 2, 3, 4], [1, 4, 9, 16], label="y=x^2")
fig.title("My Plot")
fig.save_svg("plot.svg")

# GPU Acceleration
dev = sv.gpu.Device()
gt = sv.gpu.GpuTensor(dev, [1.0, 2.0, 3.0, 4.0], [2, 2])
result = sv.gpu.matmul(gt, gt)
print(result.to_list())

---

What's Included

Submodule	Replaces	Highlights
Core (sv.Tensor)	numpy	N-d tensors, broadcasting, element-wise math, linear algebra
sv.DataFrame	pandas	Column-oriented DataFrames, GroupBy, joins, pivots, SQL
sv.stats	scipy.stats	14 distributions, hypothesis tests, regression, effect sizes
sv.ml	scikit-learn	Trees, forests, SVM, KNN, NB, clustering, PCA, t-SNE, pipelines
sv.nn	PyTorch	Autograd, Linear/Conv/LSTM/GRU/Attention layers, optimizers
sv.optim	scipy.optimize	Root finding, minimization, ODE solvers, curve fitting, LP
sv.viz	matplotlib	Line, scatter, bar, histogram, heatmap, box, violin, SVG output
sv.signal	scipy.signal + librosa	FFT, STFT, wavelets, filters, MFCC, beat tracking, WAV I/O
sv.image	OpenCV + Pillow	Filters, transforms, features, contours, Hough, augmentation
sv.nlp	NLTK + spaCy + gensim	Tokenizers, TF-IDF, Word2Vec, sentiment, POS tagging, NER, LDA
sv.graph	NetworkX	Graph/DiGraph, BFS/DFS, Dijkstra, PageRank, max flow, MST
sv.sym	SymPy	Symbolic expressions, differentiation, integration, solving
sv.rl	Stable-Baselines3	CartPole/MountainCar/GridWorld, DQN, PPO, A2C, SAC, TD3
sv.gpu	CuPy	GPU tensors via wgpu/Metal/Vulkan, matmul, activations
sv.linalg	numpy.linalg	LU, QR, SVD, Eigendecomposition, Cholesky
sv.fft	numpy.fft	1D/2D FFT, real FFT, inverse FFT
sv.io	pandas I/O	CSV, JSON read/write

---

Key Features

• Zero dependencies — no numpy, no pandas, no scipy. Pure Rust underneath.
• 697 tests — comprehensive test coverage across all modules.
• Full type stubs — .pyi stubs for IDE autocomplete and mypy support.
• GPU acceleration — wgpu backend (Metal on macOS, Vulkan on Linux/Windows).
• 107 Python classes — from Tensor to DQN to Word2Vec.
• 207 functions — statistics, ML metrics, signal processing, image filters, and more.
• 16 submodules — one import covers the entire data science stack.

---

Installation

pip install pyscivex

Requires Python 3.9+. No compiler needed — prebuilt wheels available for:

• macOS (arm64, x86_64)
• Linux (x86_64, aarch64)
• Windows (x64)

---

Examples

Neural Network Training

import pyscivex as sv

# Build model
model = sv.nn.Sequential([
    sv.nn.Linear(784, 128, seed=42),
    sv.nn.ReLU(),
    sv.nn.Linear(128, 10, seed=43),
])

# Training loop
optimizer = sv.nn.Adam(model.parameters(), lr=0.001)
for epoch in range(10):
    x = sv.nn.tensor([[1.0] * 784], requires_grad=True)
    y = sv.nn.tensor([[1.0, 0, 0, 0, 0, 0, 0, 0, 0, 0]])
    pred = model.forward(x)
    loss = sv.nn.mse_loss(pred, y)
    loss.backward()
    optimizer.step()
    optimizer.zero_grad()

Statistical Analysis

import pyscivex as sv

group_a = [23.0, 25.0, 28.0, 30.0, 32.0]
group_b = [18.0, 20.0, 22.0, 24.0, 26.0]

result = sv.stats.ttest_ind(group_a, group_b)
print(f"t={result['statistic']:.3f}, p={result['p_value']:.4f}")

effect = sv.stats.cohens_d(group_a, group_b)
print(f"Cohen's d = {effect:.3f}")

Reinforcement Learning

import pyscivex as sv

agent = sv.rl.DQN(4, 2, learning_rate=0.001, seed=42)
result = agent.train_cartpole(episodes=100)
print(f"Mean reward (last 10): {result['episode_rewards'][-10:]}")

Graph Analysis

import pyscivex as sv

g = sv.graph.Graph.from_edges([(0, 1, 1.0), (1, 2, 2.0), (0, 2, 4.0)])
dists, _ = g.dijkstra(0)
print(f"Shortest distances from 0: {dists}")

mst, cost = g.kruskal()
print(f"MST cost: {cost}")

Symbolic Math

import pyscivex as sv

x = sv.sym.var("x")
expr = x ** sv.sym.constant(2.0) + sv.sym.constant(3.0) * x
deriv = sv.sym.sym_diff(expr, "x")
print(deriv.eval({"x": 2.0}))  # 7.0 (2x + 3 at x=2)

GPU Computing

import pyscivex as sv

dev = sv.gpu.Device()
print(dev.info())  # {'name': 'Apple M...', 'backend': 'Metal', ...}

a = sv.gpu.GpuTensor(dev, [1.0, 2.0, 3.0, 4.0], [2, 2])
b = sv.gpu.GpuTensor(dev, [5.0, 6.0, 7.0, 8.0], [2, 2])
c = sv.gpu.matmul(a, b)
print(c.to_list())  # [19.0, 22.0, 43.0, 50.0]

---

License

MIT