nb · CLIO Core

What it does

A single pass accumulates per-class counts, sums and sums of squares. Variances are smoothed by var_smoothing * max variance, as scikit-learn does. The result is bit-reproducible across cache sizes and block counts.

Rows

n_features feature columns followed by the class label in 0..num_classes-1. clio_ml.COLUMNS["nb"] is 1.

Python

python

import numpy as np, clio_ml as cm
rng = np.random.default_rng(1)
X = rng.normal(size=(4096, 7)).astype(np.float32)
y = (X[:, 0] + X[:, 1] > 0).astype(np.float32)
data = np.column_stack([X, y]).astype(np.float32)

r = cm.fit_nb(data, num_classes=2)
m = r.model
print(m.prior, np.asarray(m.mean).reshape(m.C, m.nf))
print(cm.predict("nb", r, data[0, :7]))        # predicted class

fit_nb(X, **options).
Options (NaiveBayesOptions): num_classes (2), var_smoothing (1e-9), threads (256).
Result NaiveBayesResult: model and stats. model (NaiveBayesModel) has nf, C, prior (C), mean (C*nf) and var (C*nf, smoothed).
predict("nb", r, x) returns the class index.

C++

cpp

struct NaiveBayesOptions {
  u32 num_classes = 2;
  float var_smoothing = 1e-9f;
  u32 threads = 256;
  ClusterExchange cluster;
};
struct NaiveBayesModel {
  u32 nf = 0, C = 0;
  std::vector<float> prior;  // C
  std::vector<float> mean;   // C*nf
  std::vector<float> var;    // C*nf (smoothed)
  std::vector<float> Flatten() const;
};
struct NaiveBayesResult { NaiveBayesModel model; VecStats stats; };
NaiveBayesResult FitNaiveBayes(Dataset &ds, const NaiveBayesOptions &opt);
NaiveBayesResult FitNaiveBayesPlain(const float *data, u64 n_rows, u32 nf,
                                    u64 elems_per_page, u32 nblocks,
                                    const NaiveBayesOptions &opt);
NaiveBayesResult FitNaiveBayesReference(/* same arguments */);
u32 NaiveBayesPredict(const NaiveBayesModel &m, const float *x);

The ChiMod is module ml_nb.