Method test

src/neural-network.js:740–816

* * @param data * @returns { * { * error: number, * misclasses: Array * } * }

(data)

   * }
   */
  test(data) {
    data = this._formatData(data);

    // for binary classification problems with one output node
    let isBinary = data[0].output.length === 1;
    let falsePos = 0;
    let falseNeg = 0;
    let truePos = 0;
    let trueNeg = 0;

    // for classification problems
    let misclasses = [];

    // run each pattern through the trained network and collect
    // error and misclassification statistics
    let sum = 0;
    for (let i = 0; i < data.length; i++) {
      let output = this.runInput(data[i].input);
      let target = data[i].output;

      let actual, expected;
      if (isBinary) {
        actual = output[0] > this.binaryThresh ? 1 : 0;
        expected = target[0];
      }
      else {
        actual = output.indexOf(max(output));
        expected = target.indexOf(max(target));
      }

      if (actual !== expected) {
        let misclass = data[i];
        Object.assign(misclass, {
          actual: actual,
          expected: expected
        });
        misclasses.push(misclass);
      }

      if (isBinary) {
        if (actual === 0 && expected === 0) {
          trueNeg++;
        } else if (actual === 1 && expected === 1) {
          truePos++;
        } else if (actual === 0 && expected === 1) {
          falseNeg++;
        } else if (actual === 1 && expected === 0) {
          falsePos++;
        }
      }

      let errors = output.map((value, i) => {
        return target[i] - value;
      });
      sum += mse(errors);
    }
    let error = sum / data.length;