Preprocessing#

Preprocessing turns raw columns into the inputs a model expects: features on a common scale, categories as numbers, missing values filled, and only the useful columns kept. Fit a transformer on the training rows, then apply the same fitted transformer to new rows.

Use

Estimators

Scaling

StandardScaler, MinMaxScaler, MaxAbsScaler, RobustScaler, Normalizer

Reshaping distributions

PowerTransformer, QuantileTransformer

Categorical features

OneHotEncoder, OrdinalEncoder, TargetEncoder

Labels

LabelEncoder, LabelBinarizer, MultiLabelBinarizer

Missing values

SimpleImputer, IterativeImputer, KNNImputer

Binning and new features

KBinsDiscretizer, Binarizer, PolynomialFeatures, SplineTransformer

Feature selection

SelectKBest, VarianceThreshold, RFE

Feature scores

f_classif(), f_regression(), chi2(), mutual_info_classif(), r_regression()

Scaling#

import numpy as np
import mojolearn

rng = np.random.default_rng(0)
X = rng.random((10_000, 8), dtype=np.float32)
X_train, X_test = X[:8_000], X[8_000:]

scaler = mojolearn.StandardScaler().fit(X_train)
Z_train = scaler.transform(X_train)
Z_test = scaler.transform(X_test)

StandardScaler and MinMaxScaler also take data in batches with partial_fit, and save to a file with save and load.

Categories and missing values#

OneHotEncoder encodes numeric category codes as one-hot columns and returns a dense float32 array. SimpleImputer fills missing values with a per-column statistic.

codes = np.array([[0.0, 1.0], [1.0, 1.0], [2.0, 0.0]], dtype=np.float32)
onehot = mojolearn.OneHotEncoder(drop="first").fit_transform(codes)

gaps = np.array([[1.0, np.nan], [3.0, 4.0], [np.nan, 8.0]], dtype=np.float32)
filled = mojolearn.SimpleImputer(strategy="median").fit_transform(gaps)

Feature selection#

X = rng.random((1_000, 20), dtype=np.float32)
y = (X[:, 3] > 0.5).astype(np.int64)
selector = mojolearn.SelectKBest(mojolearn.f_classif, k=5).fit(X, y)
X_small = selector.transform(X)
print(selector.get_support())

Devices and numeric modes#

In identical mode, mojolearn returns the same bits on Apple, NVIDIA and AMD GPUs. identical is the default. On Apple GPUs you can choose fast instead, which is tuned for speed and does not promise identical bits: call mojolearn.set_numeric_mode() with "fast". See Numeric modes.

See also#