---

name: ml-algorithms description: "Use this skill for MLAlgorithms (mla) educational machine-learning implementations: classical estimators, clustering/reduction, metrics, NeuralNet building blocks, and DQN examples." disable-model-invocation: true metadata: disco-role: operating license: MIT

MLAlgorithms Repo Skill

Use this skill when a task asks about the rushter/MLAlgorithms package, the mla Python distribution, or its minimal NumPy/SciPy/autograd implementations of common machine-learning algorithms. The package is educational rather than production-optimized: favor small examples, explicit NumPy arrays, deterministic seeds, and direct metric checks.

First checks

• Install the package and its scientific Python dependencies in an isolated environment. The public distribution name is mla.

• Confirm the package imports:

  python - <<'PY'
  import mla
  from mla.linear_models import LinearRegression
  from mla.kmeans import KMeans
  from mla.neuralnet import NeuralNet
  print("mla import ok")
  PY
  

• Run scripts/run_import_smoke.py --json from this skill directory to inspect package imports, dependency versions, important signatures, and compatibility warnings without training, plotting, downloading, or reading original examples.

• Read references/repo-provenance.md before deciding whether this skill is current for a checkout or should be refreshed.

Route by user goal

• Classical supervised estimators: use sub-skills/classical-estimators/SKILL.md for linear/logistic regression, KNN, Naive Bayes, SVM kernels, random forests, gradient boosting, and experimental factorization machines.
• Unsupervised and reduction workflows: use sub-skills/unsupervised-and-reduction/SKILL.md for KMeans, GaussianMixture, PCA, t-SNE, RBM, demo dataset loaders, distances, and no-display clustering/reduction checks.
• Neural network building blocks: use sub-skills/neural-network-building-blocks/SKILL.md for NeuralNet, layers, activations, initializers, constraints, regularizers, optimizers, CNN/RNN/LSTM recipes, and DQN wiring.

Common decisions

• Package name: install/query distribution mla; import package mla.
• No CLIs: the project exposes Python APIs and example modules, not console entry points. Use bundled skill scripts for safe checks.
• Backend: selected workflows are CPU-only. A visible GPU is not required for this package.
• Compatibility: the current dataset text loader uses deprecated np.bool; prefer NumPy <1.24 or patch the loader before using load_nietzsche() with modern NumPy. The DQN loop expects legacy Gym reset/step signatures, so do not assume Gymnasium compatibility.
• Shapes: most estimators expect 2D feature arrays. Neural layers use explicit 2D dense, 3D sequence, or 4D image tensors.
• Plots and long examples: repo examples include plotting, MNIST ConvNet, RNN text generation, and CartPole DQN training. Treat those as reference workflows unless the user explicitly authorizes longer runs or display side effects.

Root references

• references/api-reference.md: package-wide imports, metrics, datasets, dependency facts, and public API index.
• references/workflows.md: how to select and combine the sub-skills for common tasks.
• references/troubleshooting.md: cross-cutting install/import, dependency, data-loader, plotting, and runtime problems.
• references/repo-provenance.md: source snapshot and refresh triggers.
• references/repo-routing-metadata.json: structured metadata for the managed repo-skills router.

Bundled helpers

• scripts/run_import_smoke.py: safe package/dependency/signature check for the active Python environment.
• sub-skills/classical-estimators/scripts/run_classical_smoke.py: small supervised estimator checks.
• sub-skills/unsupervised-and-reduction/scripts/run_unsupervised_smoke.py: small clustering/reduction/RBM checks.
• sub-skills/neural-network-building-blocks/scripts/run_neural_smoke.py: small dense/RBM/DQN-wiring checks.

Minimal examples

Classical estimator:

from mla.linear_models import LogisticRegression

model = LogisticRegression(lr=0.01, max_iters=300)
model.fit(X_train, y_train)
proba = model.predict(X_test)
labels = (proba >= 0.5).astype(int)

Unsupervised estimator:

from mla.kmeans import KMeans

model = KMeans(K=3, init="++", max_iters=50)
model.fit(X)
labels = model.predict()

Neural model:

from mla.neuralnet import NeuralNet
from mla.neuralnet.layers import Dense, Activation
from mla.neuralnet.optimizers import Adam

model = NeuralNet([Dense(16), Activation("relu"), Dense(1)], Adam(), loss="mse", max_epochs=5)