Quickstart#

Train and predict#

import numpy as np
import mojolearn

rng = np.random.default_rng(0)
X = rng.random((100_000, 20), dtype=np.float32)
y = (X[:, 0] + X[:, 1] > 1.0).astype(np.float32)

model = mojolearn.GradientBoosting(loss="Logloss", n_estimators=200, max_depth=6)
model.fit(X, y)
print(model.predict_proba(X[:5]))
print(model.numeric_mode_used(), mojolearn.vendor())

vendor() names the backend that ran, and numeric_mode_used() the numeric mode. In identical mode, mojolearn returns the same bits on Apple, NVIDIA and AMD GPUs. Run the same script with the same release on any of them and the model and its predictions are the same bits.

The estimator interface#

Estimators follow scikit-learn’s conventions: construct with hyperparameters, then fit, predict, predict_proba, transform or score as the estimator supports. Inputs are float32 arrays; NumPy arrays work directly, and the library itself does not require NumPy.

from mojolearn import KMeans, PCA, RandomForestClassifier

labels = KMeans(n_clusters=8).fit(X).labels_
Z = PCA(n_components=5).fit_transform(X)
clf = RandomForestClassifier(n_estimators=100).fit(X, y)

Choose a numeric mode#

identical is the default. Select another mode for the process or per call:

previous = mojolearn.set_numeric_mode("fast")            # returns the previous mode
mojolearn.set_numeric_mode("identical")                  # back to the default

You can also set MOJOLEARN_NUMERIC_MODE in the environment before import. fast is available on Apple GPUs only.

Estimators that take a numeric_mode= argument, such as RandomForestClassifier, override the process mode per call. Asking a family for a mode it does not ship raises a named error rather than running something weaker. See Numeric modes.

Next#