Machine Learning and AI

Prerequisites: Importing data and Dataset

Machine learning in the Wolfram Language is highly automated: one function chooses a method, preprocesses the data, and trains a model.

Clustering

FindClusters[{1, 2, 10, 11, 12}]

{{1, 2}, {10, 11, 12}}

FindClusters groups similar values (numbers, vectors, strings, images) without labels. ClusterClassify returns a reusable classifier.

Classification and prediction

(* illustrative: trained models vary between runs and versions *)
training = {
  {1.0, 1.1} -> "small", {1.2, 0.9} -> "small", {0.8, 1.0} -> "small",
  {5.0, 5.2} -> "large", {5.3, 4.9} -> "large", {4.8, 5.1} -> "large"};
c = Classify[training];
c[{1.1, 1.0}]
c[{5.1, 5.0}, "Probabilities"]

Classify learns labels from examples written as input -> label. Here c[{1.1, 1.0}] returns small, and "Probabilities" gives the model's confidence for each label.
Predict does the same for numeric outputs (regression).
Method -> "RandomForest", "LogisticRegression", "GradientBoostedTrees", and others choose the algorithm explicitly.
ClassifierMeasurements[c, testSet] reports accuracy, a confusion matrix, and more — always evaluate on data the model didn't see.

Inputs can be numbers, text, images, or associations of mixed features.

Neural networks

(* illustrative: downloads a pretrained model *)
net = NetModel["LeNet Trained on MNIST Data"];
net[image]

The Wolfram Neural Net Repository has pretrained networks for images, text, and audio, loaded with NetModel.
NetChain and NetGraph build networks from layers; NetTrain trains them on CPU or GPU.
ImageIdentify, TextSentences, SpeechRecognize, and similar functions use built-in models directly.

Large language models

(* illustrative: needs an LLM service configured *)
summarize = LLMFunction["Summarize in one sentence: ``"];
summarize["The Wolfram Language is a symbolic language ..."]

LLMFunction, LLMSynthesize, and chat notebooks connect to large language models (through a Wolfram subscription or your own API key), and LLMTool lets a model call your Wolfram Language functions. Version 15 builds an AI assistant into notebooks.

Good practice

Split data into training and test sets (RandomSample then TakeDrop), and seed randomness with SeedRandom for reproducible experiments.
Treat automatic model choice as a strong baseline, then compare methods with ClassifierMeasurements or PredictorMeasurements.
Save trained models with Export["model.wmlf", c] and load them in deployed APIs.