Preprocessing: Encoders

1/**2 * Example 17: Preprocessing: Encoders3 *4 * Turn categorical and label data into numbers, and back again. Each encoder5 * is fitted on the training values, then transforms and inverse-transforms.6 * Feature encoders (OneHotEncoder, OrdinalEncoder) expect a 2D input with one7 * column per feature, so a 1D list of values is reshaped to [n, 1] first.8 */9 10import { tensor } from "deepbox/ndarray";11import {12  LabelBinarizer,13  LabelEncoder,14  MultiLabelBinarizer,15  OneHotEncoder,16  OrdinalEncoder,17} from "deepbox/preprocess";18 19console.log("=== Preprocessing: Encoders ===\n");20 21// ---------------------------------------------------------------------------22// Part 1: LabelEncoder: map string labels to integers23// ---------------------------------------------------------------------------24console.log("--- Part 1: LabelEncoder ---");25 26const le = new LabelEncoder();27le.fit(tensor(["cat", "dog", "bird", "cat", "bird"]));28 29// Classes are sorted alphabetically, so bird = 0, cat = 1, dog = 2.30console.log("Classes:", le.classes?.toString());31 32const encoded = le.transform(tensor(["bird", "cat", "dog"]));33console.log("Encoded:", encoded.toString());34 35const decoded = le.inverseTransform(encoded);36console.log("Decoded:", decoded.toString());37 38// ---------------------------------------------------------------------------39// Part 2: OneHotEncoder: one-hot vectors for categorical features40// ---------------------------------------------------------------------------41console.log("\n--- Part 2: OneHotEncoder ---");42 43// By default an unseen category throws. { handleUnknown: "ignore" } encodes it as all zeros.44const ohe = new OneHotEncoder();45ohe.fit(tensor(["red", "green", "blue", "red", "blue"]).reshape([5, 1]));46 47const oneHot = ohe.transform(tensor(["red", "blue", "green"]).reshape([3, 1]));48console.log("One-hot encoded shape:", oneHot.shape);49console.log("One-hot encoded:\n", oneHot.toString());50console.log("Categories per column:", JSON.stringify(ohe.categories));51 52// ---------------------------------------------------------------------------53// Part 3: OrdinalEncoder: ordinal integer encoding54// ---------------------------------------------------------------------------55console.log("\n--- Part 3: OrdinalEncoder ---");56 57// With the default categories: "auto", values are numbered in alphabetical order,58// so high = 0, low = 1, medium = 2. That is not the order low < medium < high.59const oe = new OrdinalEncoder();60oe.fit(tensor(["low", "medium", "high", "medium", "low"]).reshape([5, 1]));61 62const ordinal = oe.transform(tensor(["low", "high", "medium"]).reshape([3, 1]));63console.log("Ordinal encoded (alphabetical):", ordinal.toString());64 65// Pass the categories in the order you want to get low = 0, medium = 1, high = 2.66const ordered = new OrdinalEncoder({ categories: [["low", "medium", "high"]] });67ordered.fit(tensor(["low", "medium", "high"]).reshape([3, 1]));68console.log(69  "Ordinal encoded (explicit order):",70  ordered.transform(tensor(["low", "high", "medium"]).reshape([3, 1])).toString()71);72 73const ordinalDecoded = oe.inverseTransform(ordinal);74console.log("Decoded:", ordinalDecoded.toString());75 76// ---------------------------------------------------------------------------77// Part 4: LabelBinarizer: binary indicator for multi-class labels78// ---------------------------------------------------------------------------79console.log("\n--- Part 4: LabelBinarizer ---");80 81const lb = new LabelBinarizer();82lb.fit(tensor(["cat", "dog", "bird"]));83 84const binarized = lb.transform(tensor(["cat", "bird", "dog"]));85console.log("Binarized shape:", binarized.shape);86console.log("Binarized:\n", binarized.toString());87 88const binarizedDecoded = lb.inverseTransform(binarized);89console.log("Decoded:", binarizedDecoded.toString());90 91// ---------------------------------------------------------------------------92// Part 5: MultiLabelBinarizer: multi-label binary encoding93// ---------------------------------------------------------------------------94console.log("\n--- Part 5: MultiLabelBinarizer ---");95 96const mlb = new MultiLabelBinarizer();97mlb.fit([["cat", "dog"], ["bird"], ["cat", "bird", "dog"]]);98 99const multiEncoded = mlb.transform([["cat", "bird"], ["dog"]]);100console.log("Multi-label encoded shape:", multiEncoded.shape);101console.log("Multi-label encoded:\n", multiEncoded.toString());102 103console.log("\n=== Preprocessing: Encoders Complete ===");