src/mnist_loader.js

"use strict";

let _ = require("lodash");
let linearAlgebra = require("linear-algebra")();
let Matrix = linearAlgebra.Matrix;

let lib = require("./lib");

let loadTrainingData = function() {
  let trainingInput = require("../data/training_input");
  let trainingOutput = require("../data/training_output");
  return [trainingInput, trainingOutput];
};

let loadValidationData = function() {
  let validationInput = require("../data/validation_input");
  let validationOutput = require("../data/validation_output");
  return [validationInput, validationOutput];
};

let loadTestData = function() {
  let testInput = require("../data/test_input");
  let testOutput = require("../data/test_output");
  return [testInput, testOutput];
};

exports.loadTrainingDataWrapper = function() {
  let trD = loadTrainingData();
  let trainingInputs = trD[0].map(x => { return Matrix.reshape(x, 784, 1); });
  let trainingResults = trD[1].map(y => { return lib.vectorizedResult(y); });
  let trainingData = _.zip(trainingInputs, trainingResults);
  return trainingData;
};

exports.loadValidationDataWrapper = function() {
  let vaD = loadValidationData();
  let validationInputs = vaD[0].map(x => { return Matrix.reshape(x, 784, 1); });
  let validationData = _.zip(validationInputs, vaD[1]);
  return validationData;
};

exports.loadTestDataWrapper = function() {
  let teD = loadTestData();
  let testInputs = teD[0].map(x => { return Matrix.reshape(x, 784, 1); });
  let testData = _.zip(testInputs, teD[1]);
  return testData;
};