feat(lib): Column value mapping capability (#64)

* feat(lib): Column value mapping capability

* test(lib): Tests for applyMappings

* chore(config): Ignore shrinkwrap.yaml

* fix(lib): Fix mappings implementation

Co-authored-by: Glitch (isair-tensorflow-load-csv) <none>

Author: isair

.gitignore

@@ -24,3 +24,4 @@ pnpm-debug.log*
 dist
 docs
 coverage
+shrinkwrap.yaml

README.md

@@ -55,6 +55,10 @@ const {
 } = loadCsv('./data.csv', {
   featureColumns: ['lat', 'lng', 'height'],
   labelColumns: ['temperature'],
+  mappings: {
+    height: (ft) => ft * 0.3048, // feet to meters
+    temperature: (f) => (f - 32) / 1.8, // fahrenheit to celsius
+  }, // Map values based on which column they are in before they are loaded into tensors.
   shuffle: true, // Pass true to shuffle with a fixed seed, or a string to use it as a seed for the shuffling.
   splitTest: true, // Splits your data in half. You can also provide a certain row count for the test data.
   prependOnes: true, // Prepends a column of 1s to your features and testFeatures tensors, useful for linear regression.

src/applyMappings.ts

@@ -0,0 +1,22 @@
+import { CsvTable, CsvReadOptions } from './loadCsv.models';
+
+const applyMappings = (
+  table: CsvTable,
+  mappings: NonNullable<CsvReadOptions['mappings']>
+) => {
+  if (table.length < 2) {
+    return table;
+  }
+  const mappingsByIndex = table[0].map((columnName) => mappings[columnName]);
+  return table.map((row, index) =>
+    index === 0
+      ? row
+      : row.map((value, columnIndex) =>
+          mappingsByIndex[columnIndex]
+            ? mappingsByIndex[columnIndex](value)
+            : value
+        )
+  );
+};
+
+export default applyMappings;

src/loadCsv.models.ts

@@ -7,6 +7,13 @@ export interface CsvReadOptions {
    * Names of the columns to be included in the labels and testLabels tensors.
    */
   labelColumns: string[];
+  /**
+   * Used for transforming values of entire columns. Key is column label, value is transformer function. Each value belonging to
+   * that column will be put through the transformer function and be overwritten with the return value of it.
+   */
+  mappings?: {
+    [columnName: string]: (value: string | number) => string | number;
+  };
   /**
    * If true, shuffles all rows with a fixed seed, meaning that shuffling the same data will always result in the same shuffled data.
    *

src/loadCsv.ts

@@ -6,13 +6,15 @@ import { shuffle } from 'shuffle-seed';
 import { CsvReadOptions, CsvTable } from './loadCsv.models';
 import filterColumns from './filterColumns';
 import splitTestData from './splitTestData';
+import applyMappings from './applyMappings';
 
 const defaultShuffleSeed = 'mncv9340ur';
 
 const loadCsv = (filename: string, options: CsvReadOptions) => {
   const {
     featureColumns,
     labelColumns,
+    mappings = {},
     shuffle: shouldShuffle = false,
     splitTest = false,
     prependOnes = false,
@@ -43,6 +45,10 @@ const loadCsv = (filename: string, options: CsvReadOptions) => {
   tables.labels.shift();
   tables.features.shift();
 
+  for (const key of Object.keys(tables)) {
+    tables[key] = applyMappings(tables[key], mappings);
+  }
+
   if (shouldShuffle) {
     const seed =
       typeof shouldShuffle === 'string' ? shouldShuffle : defaultShuffleSeed;

tests/applyMappings.test.ts

@@ -0,0 +1,38 @@
+/* eslint-disable @typescript-eslint/ban-ts-comment */
+import { CsvReadOptions } from '../src/loadCsv.models';
+import applyMappings from '../src/applyMappings';
+
+const data = [
+  ['lat', 'lng', 'height', 'temperature'],
+  [0.234, 1.47, 849.7, 64.4],
+  [-293.2, 103.34, 715.2, 73.4],
+];
+
+const mappings: NonNullable<CsvReadOptions['mappings']> = {
+  height: (ft) => Number(ft) * 0.3048, // feet to meters
+  temperature: (f) => (Number(f) - 32) / 1.8, // fahrenheit to celsius
+};
+
+test('Applying mappings works correctly', () => {
+  const mappedData = applyMappings(data, mappings);
+  // @ts-ignore
+  expect(mappedData).toBeDeepCloseTo(
+    [
+      ['lat', 'lng', 'height', 'temperature'],
+      [0.234, 1.47, 258.98856, 18],
+      [-293.2, 103.34, 217.99296, 23],
+    ],
+    3
+  );
+});
+
+test('Applying mappings does not break with a table with just headers', () => {
+  const tableOnlyHeaders = [['lat', 'lng', 'height', 'temperature']];
+  const mappedData = applyMappings(tableOnlyHeaders, mappings);
+  expect(mappedData).toMatchObject(tableOnlyHeaders);
+});
+
+test('Applying mappings does not break with an empty table', () => {
+  const mappedData = applyMappings([], mappings);
+  expect(mappedData).toMatchObject([]);
+});