Decision Tree

A decision tree is a supervised learning model that splits data into branches based on feature values, forming a tree-like structure of decisions. Each internal node tests a feature (e.g., 'is VRAM > 8 GB?'), each branch represents the outcome, and each leaf gives a prediction (e.g., 'model fits'). Operators encounter decision trees in classical ML pipelines for tabular data, often as baselines or interpretable alternatives to neural networks. They are fast to train and run on CPU, making them practical for small datasets on local hardware.

An operator training a classifier to predict whether a GPU can run a given model might build a decision tree with features like VRAM, model size, and quantization level. The tree would split on VRAM first: if VRAM < 8 GB, predict 'cannot run'; else, check model size. This yields a simple, interpretable rule set that runs in microseconds on CPU.

Using scikit-learn on a local machine, an operator runs from sklearn.tree import DecisionTreeClassifier; clf.fit(X_train, y_train) to train on a CSV of hardware specs. They then export the tree with export_graphviz to visualize splits. The model is small (kilobytes) and loads instantly, useful for quick inference on new hardware queries without GPU acceleration.