chore: moving the reshape method to ops/tensor_ops.ts; adding more ex…

…ception tests

src/lib/ops/tensor_ops.ts

@@ -1,4 +1,5 @@
 import * as tf from '@tensorflow/tfjs';
+import { flattenDeep } from 'lodash';
 import { Type1DMatrix, Type2DMatrix, TypeMatrix } from '../types';
 
 /**
@@ -84,3 +85,61 @@ export function validateMatrix2D(X: Type2DMatrix<any>): number[] {
   }
   return shape;
 }
+
+/**
+ * Reshapes any size of array into a new shape. The code was copied from
+ * math.js, https://github.com/josdejong/mathjs/blob/5750a1845442946d236822505c607a522be23474/src/utils/array.js#L258
+ * in order to use specific method from Math.js instead of install an entire library.
+ *
+ * @example
+ * reshape([1, 2, 3, 4, 5, 6], [2, 3]); // [[1, 2, 3], [4, 5, 6]]
+ *
+ * @param array - Target array
+ * @param sizes - New array shape to resize into
+ * @ignore
+ */
+export function reshape(
+  array: TypeMatrix<any>,
+  sizes: number[]
+): TypeMatrix<any> {
+  // Initial validations
+  if (!Array.isArray(array)) {
+    throw new TypeError('The input array must be an array!');
+  }
+
+  if (!Array.isArray(sizes)) {
+    throw new TypeError('The sizes must be an array!');
+  }
+
+  const deepFlatArray = flattenDeep(array);
+  // If the reshaping is to single dimensional
+  if (sizes.length === 1 && deepFlatArray.length === sizes[0]) {
+    return deepFlatArray;
+  } else if (sizes.length === 1 && deepFlatArray.length !== sizes[0]) {
+    throw new TypeError(
+      `Target array shape [${
+        deepFlatArray.length
+      }] cannot be reshaped into ${sizes}`
+    );
+  }
+
+  // testing if there are enough elements for the requested shape
+  let tmpArray = deepFlatArray;
+  let tmpArray2;
+  // for each dimensions starting by the last one and ignoring the first one
+  for (let sizeIndex = sizes.length - 1; sizeIndex > 0; sizeIndex--) {
+    const size = sizes[sizeIndex];
+
+    tmpArray2 = [];
+
+    // aggregate the elements of the current tmpArray in elements of the requested size
+    const length = tmpArray.length / size;
+    for (let i = 0; i < length; i++) {
+      tmpArray2.push(tmpArray.slice(i * size, (i + 1) * size));
+    }
+    // set it as the new tmpArray for the next loop turn or for return
+    tmpArray = tmpArray2;
+  }
+
+  return tmpArray;
+}

test/ops/tensor_ops.test.ts

@@ -1,6 +1,7 @@
 import { Iris } from '../../src/lib/datasets';
 import {
   inferShape,
+  reshape,
   validateFitInputs,
   validateMatrix1D,
   validateMatrix2D
@@ -157,3 +158,41 @@ describe('ops', () => {
     });
   });
 });
+
+describe('utils.reshape', () => {
+  it('should reshape an array of shape [1] into [2, 3]', () => {
+    const result = reshape([1, 2, 3, 4, 5, 6], [2, 3]);
+    expect(result).toEqual([[1, 2, 3], [4, 5, 6]]);
+  });
+
+  it('should reshape an array of shape [2, 3] into [1]', () => {
+    // console.log(tf.tensor1d([1, 2, 3]).shape);
+    const result = reshape([[1, 2, 3], [4, 5, 6]], [6]);
+    expect(result).toEqual(result);
+  });
+
+  it('should reshape an array of shape [1] into [2, 3, 1]', () => {
+    const result = reshape([1, 2, 3, 4, 5, 6], [2, 3, 1]);
+    expect(result).toEqual([[[1], [2], [3]], [[4], [5], [6]]]);
+  });
+
+  it('should reshape an array of shape [2, 3] into [2, 3, 1]', () => {
+    const result = reshape([[1, 2, 3], [4, 5, 6]], [2, 3, 1]);
+    expect(result).toEqual([[[1], [2], [3]], [[4], [5], [6]]]);
+  });
+
+  it('should not reshape invalid inputs', () => {
+    expect(() => reshape(null, [1])).toThrow(
+      'The input array must be an array!'
+    );
+    expect(() => reshape([], [1])).toThrow(
+      'Target array shape [0] cannot be reshaped into 1'
+    );
+    expect(() => reshape([[1, 2, 3]], null)).toThrow(
+      'The sizes must be an array!'
+    );
+    expect(() => reshape([[1, 2, 3]], 1)).toThrow(
+      'The sizes must be an array!'
+    );
+  });
+});