UNPKG

@tensorflow/tfjs-core

Version:

Hardware-accelerated JavaScript library for machine intelligence

1,152 lines (1,151 loc) 82.1 kB
"use strict"; Object.defineProperty(exports, "__esModule", { value: true }); var tf = require("../index"); var jasmine_util_1 = require("../jasmine_util"); var test_util_1 = require("../test_util"); var util = require("../util"); var selu_util = require("./selu_util"); jasmine_util_1.describeWithFlags('relu', test_util_1.ALL_ENVS, function () { it('basic', function () { var a = tf.tensor1d([1, -2, 0, 3, -0.1]); var result = tf.relu(a); test_util_1.expectArraysClose(result, [1, 0, 0, 3, 0]); }); it('does nothing to positive values', function () { var a = tf.scalar(1); var result = tf.relu(a); test_util_1.expectNumbersClose(result.get(), 1); }); it('sets negative values to 0', function () { var a = tf.scalar(-1); var result = tf.relu(a); test_util_1.expectNumbersClose(result.get(), 0); }); it('preserves zero values', function () { var a = tf.scalar(0); var result = tf.relu(a); test_util_1.expectNumbersClose(result.get(), 0); }); it('propagates NaNs, float32', function () { var a = tf.tensor1d([1, -2, 0, 3, -0.1, NaN]); var result = tf.relu(a); expect(result.dtype).toBe('float32'); test_util_1.expectArraysClose(result, [1, 0, 0, 3, 0, NaN]); }); it('gradients: positive scalar', function () { var a = tf.scalar(3); var dy = tf.scalar(5); var grad = tf.grad(function (a) { return tf.relu(a); }); var da = grad(a, dy); expect(da.shape).toEqual(a.shape); expect(da.dtype).toEqual('float32'); test_util_1.expectArraysClose(da, [5]); }); it('gradients: negative scalar', function () { var a = tf.scalar(-3); var dy = tf.scalar(5); var grad = tf.grad(function (a) { return tf.relu(a); }); var da = grad(a, dy); expect(da.shape).toEqual(a.shape); expect(da.dtype).toEqual('float32'); test_util_1.expectArraysClose(da, [0]); }); it('gradients: array', function () { var a = tf.tensor2d([1, -1, 0, .1], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var grad = tf.grad(function (a) { return tf.relu(a); }); var da = grad(a, dy); expect(da.shape).toEqual(a.shape); expect(da.dtype).toEqual('float32'); test_util_1.expectArraysClose(da, [1, 0, 0, 4]); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.relu({}); }) .toThrowError(/Argument 'x' passed to 'relu' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('abs', test_util_1.ALL_ENVS, function () { it('basic', function () { var a = tf.tensor1d([1, -2, 0, 3, -0.1]); var result = tf.abs(a); test_util_1.expectArraysClose(result, [1, 2, 0, 3, 0.1]); }); it('propagates NaNs', function () { var a = tf.tensor1d([1, -2, 0, 3, -0.1, NaN]); var result = tf.abs(a); test_util_1.expectArraysClose(result, [1, 2, 0, 3, 0.1, NaN]); }); it('gradients: Scalar', function () { var a = tf.scalar(4); var dy = tf.scalar(8); var da = tf.grad(function (a) { return tf.abs(a); })(a, dy); expect(da.shape).toEqual(a.shape); expect(da.dtype).toEqual('float32'); test_util_1.expectArraysClose(da, [8 * 1]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([1, 2, -3, 5]); var dy = tf.tensor1d([1, 2, 3, 4]); var da = tf.grad(function (a) { return tf.abs(a); })(a, dy); expect(da.shape).toEqual(a.shape); expect(da.dtype).toEqual('float32'); test_util_1.expectArraysClose(da, [1 * 1, 2 * 1, 3 * -1, 4 * 1]); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([3, -1, -2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var da = tf.grad(function (a) { return tf.abs(a); })(a, dy); expect(da.shape).toEqual(a.shape); expect(da.dtype).toEqual('float32'); test_util_1.expectArraysClose(da, [1 * 1, 2 * -1, 3 * -1, 4 * 1]); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.abs({}); }) .toThrowError(/Argument 'x' passed to 'abs' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('step', test_util_1.ALL_ENVS, function () { it('with 1d tensor', function () { var a = tf.tensor1d([1, -2, -.01, 3, -0.1]); var result = tf.step(a); test_util_1.expectArraysClose(result, [1, 0, 0, 1, 0]); }); it('with 1d tensor and alpha', function () { var a = tf.tensor1d([1, -2, -.01, 3, NaN]); var result = tf.step(a, 0.1); test_util_1.expectArraysClose(result, [1, 0.1, 0.1, 1, NaN]); }); it('with 2d tensor', function () { var a = tf.tensor2d([1, -5, -3, 4], [2, 2]); var result = tf.step(a); expect(result.shape).toEqual([2, 2]); test_util_1.expectArraysClose(result, [1, 0, 0, 1]); }); it('propagates NaNs', function () { var a = tf.tensor1d([1, -2, -.01, 3, NaN]); var result = tf.step(a); test_util_1.expectArraysClose(result, [1, 0, 0, 1, NaN]); }); it('gradients: Scalar', function () { var a = tf.scalar(-4); var dy = tf.scalar(8); var gradients = tf.grad(function (a) { return tf.step(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [0]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([1, 2, -3, 5]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.step(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [0, 0, 0, 0]); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([3, -1, -2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.step(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [0, 0, 0, 0]); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.step({}); }) .toThrowError(/Argument 'x' passed to 'step' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('neg', test_util_1.ALL_ENVS, function () { it('basic', function () { var a = tf.tensor1d([1, -3, 2, 7, -4]); var result = tf.neg(a); test_util_1.expectArraysClose(result, [-1, 3, -2, -7, 4]); }); it('propagate NaNs', function () { var a = tf.tensor1d([1, -3, 2, 7, NaN]); var result = tf.neg(a); var expected = [-1, 3, -2, -7, NaN]; test_util_1.expectArraysClose(result, expected); }); it('gradients: Scalar', function () { var a = tf.scalar(4); var dy = tf.scalar(8); var da = tf.grad(function (a) { return tf.neg(a); })(a, dy); expect(da.shape).toEqual(a.shape); expect(da.dtype).toEqual('float32'); test_util_1.expectArraysClose(da, [8 * -1]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([1, 2, -3, 5]); var dy = tf.tensor1d([1, 2, 3, 4]); var da = tf.grad(function (a) { return tf.neg(a); })(a, dy); expect(da.shape).toEqual(a.shape); expect(da.dtype).toEqual('float32'); test_util_1.expectArraysClose(da, [1 * -1, 2 * -1, 3 * -1, 4 * -1]); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([3, -1, -2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var da = tf.grad(function (a) { return tf.neg(a); })(a, dy); expect(da.shape).toEqual(a.shape); expect(da.dtype).toEqual('float32'); test_util_1.expectArraysClose(da, [1 * -1, 2 * -1, 3 * -1, 4 * -1]); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.neg({}); }) .toThrowError(/Argument 'x' passed to 'neg' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('sigmoid', test_util_1.ALL_ENVS, function () { it('basic', function () { var values = [1, -3, 2, 7, -4]; var a = tf.tensor1d(values); var result = tf.sigmoid(a); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = 1 / (1 + Math.exp(-values[i])); } test_util_1.expectArraysClose(result, expected); }); it('propagates NaNs', function () { var a = tf.tensor1d([3, NaN]); var res = tf.sigmoid(a); test_util_1.expectArraysClose(res, [1 / (1 + Math.exp(-3)), NaN]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([1, 2, -3, 5]); var dy = tf.tensor1d([1, 2, 3, 4]); var da = tf.grad(function (a) { return tf.sigmoid(a); })(a, dy); var expected = []; for (var i = 0; i < a.size; i++) { var y = 1 / (1 + Math.exp(-a.get(i))); expected[i] = dy.get(i) * y * (1 - y); } test_util_1.expectArraysClose(da, expected); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.sigmoid({}); }) .toThrowError(/Argument 'x' passed to 'sigmoid' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('logSigmoid', test_util_1.ALL_ENVS, function () { it('basic', function () { var values = [1, -3, 2, 7, -4]; var a = tf.tensor1d(values); var result = tf.logSigmoid(a); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = Math.log(1 / (1 + Math.exp(-values[i]))); } test_util_1.expectArraysClose(result, expected); }); it('scalar', function () { var a = tf.scalar(-2); var result = tf.logSigmoid(a); var expected = [Math.log(1 / (1 + Math.exp(2)))]; test_util_1.expectArraysClose(result, expected); }); it('tensor2D', function () { var values = [1, 2, -3, 5]; var a = tf.tensor2d(values, [2, 2]); var result = tf.logSigmoid(a); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = Math.log(1 / (1 + Math.exp(-values[i]))); } test_util_1.expectArraysClose(result, expected); }); it('larger magnitude negative inputs', function () { var values = [-100, -200, -3000, -50000]; var a = tf.tensor1d(values); var result = tf.logSigmoid(a); var expected = [-100, -200, -3000, -50000]; test_util_1.expectArraysClose(result, expected); }); it('larger magnitude positive inputs', function () { var values = [100, 200, 3000, 50000]; var a = tf.tensor1d(values); var result = tf.logSigmoid(a); var expected = [0, 0, 0, 0]; test_util_1.expectArraysClose(result, expected); }); it('propagates NaNs', function () { var a = tf.tensor1d([3, NaN]); var res = tf.logSigmoid(a); test_util_1.expectArraysClose(res, [Math.log(1 / (1 + Math.exp(-3))), NaN]); }); it('gradients: Scalar', function () { var a = tf.scalar(3); var dy = tf.scalar(4); var da = tf.grad(function (a) { return tf.logSigmoid(a); })(a, dy).get(); var y = 1 / (1 + Math.exp(a.get())); test_util_1.expectNumbersClose(da, dy.get() * y); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([1, 2, -3, 5]); var dy = tf.tensor1d([1, 2, 3, 4]); var da = tf.grad(function (a) { return tf.logSigmoid(a); })(a, dy); var expected = []; for (var i = 0; i < a.size; i++) { var y = 1 / (1 + Math.exp(a.get(i))); expected[i] = dy.get(i) * y; } test_util_1.expectArraysClose(da, expected); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([1, 2, -3, 5], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var da = tf.grad(function (a) { return tf.logSigmoid(a); })(a, dy); var expected = []; var aVals = a.dataSync(); var dyVals = dy.dataSync(); for (var i = 0; i < a.size; i++) { var y = 1 / (1 + Math.exp(aVals[i])); expected[i] = dyVals[i] * y; } test_util_1.expectArraysClose(da, expected); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.logSigmoid({}); }) .toThrowError(/Argument 'x' passed to 'logSigmoid' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('softplus', test_util_1.ALL_ENVS, function () { it('basic', function () { var values = [1, -3, 2, 7, -4]; var a = tf.tensor1d(values); var result = tf.softplus(a); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = Math.log((1 + Math.exp(values[i]))); } test_util_1.expectArraysClose(result, expected); }); it('scalar', function () { var a = tf.scalar(-2); var result = tf.softplus(a); var expected = [Math.log((1 + Math.exp(-2)))]; test_util_1.expectArraysClose(result, expected); }); it('tensor2D', function () { var values = [1, 2, -3, 5]; var a = tf.tensor2d(values, [2, 2]); var result = tf.softplus(a); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = Math.log((1 + Math.exp(values[i]))); } test_util_1.expectArraysClose(result, expected); }); it('larger magnitude negative inputs', function () { var values = [-100, -200, -3000, -50000]; var a = tf.tensor1d(values); var result = tf.softplus(a); var expected = [0, 0, 0, 0]; test_util_1.expectArraysClose(result, expected); }); it('larger magnitude positive inputs', function () { var values = [100, 200, 3000, 50000]; var a = tf.tensor1d(values); var result = tf.softplus(a); var expected = [100, 200, 3000, 50000]; test_util_1.expectArraysClose(result, expected); }); it('propagates NaNs', function () { var a = tf.tensor1d([3, NaN]); var res = tf.softplus(a); test_util_1.expectArraysClose(res, [Math.log((1 + Math.exp(3))), NaN]); }); it('gradients: Scalar', function () { var a = tf.scalar(3); var dy = tf.scalar(4); var da = tf.grad(function (a) { return tf.softplus(a); })(a, dy); var y = 1 / (1 + Math.exp(-a.get())); test_util_1.expectNumbersClose(da.get(), dy.get() * y); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([1, 2, -3, 5]); var dy = tf.tensor1d([1, 2, 3, 4]); var da = tf.grad(function (a) { return tf.softplus(a); })(a, dy); var expected = []; for (var i = 0; i < a.size; i++) { var y = 1 / (1 + Math.exp(-a.get(i))); expected[i] = dy.get(i) * y; } test_util_1.expectArraysClose(da, expected); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([1, 2, -3, 5], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var da = tf.grad(function (a) { return tf.softplus(a); })(a, dy); var expected = []; var aVals = a.dataSync(); var dyVals = dy.dataSync(); for (var i = 0; i < a.size; i++) { var y = 1 / (1 + Math.exp(-aVals[i])); expected[i] = dyVals[i] * y; } test_util_1.expectArraysClose(da, expected); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.softplus({}); }) .toThrowError(/Argument 'x' passed to 'softplus' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('sqrt', test_util_1.ALL_ENVS, function () { it('sqrt', function () { var a = tf.tensor1d([2, 4]); var r = tf.sqrt(a); test_util_1.expectNumbersClose(r.get(0), Math.sqrt(2)); test_util_1.expectNumbersClose(r.get(1), Math.sqrt(4)); }); it('sqrt propagates NaNs', function () { var a = tf.tensor1d([1, NaN]); var r = tf.sqrt(a); test_util_1.expectArraysClose(r, [Math.sqrt(1), NaN]); }); it('gradients: Scalar', function () { var a = tf.scalar(4); var dy = tf.scalar(8); var da = tf.grad(function (a) { return tf.sqrt(a); })(a, dy); expect(da.shape).toEqual(a.shape); expect(da.dtype).toEqual('float32'); test_util_1.expectArraysClose(da, [8 / (2 * Math.sqrt(4))]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([1, 2, 3, 5]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.sqrt(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [ 1 / (2 * Math.sqrt(1)), 2 / (2 * Math.sqrt(2)), 3 / (2 * Math.sqrt(3)), 4 / (2 * Math.sqrt(5)) ], 1e-1); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([3, 1, 2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.sqrt(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [ 1 / (2 * Math.sqrt(3)), 2 / (2 * Math.sqrt(1)), 3 / (2 * Math.sqrt(2)), 4 / (2 * Math.sqrt(3)) ], 1e-1); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.sqrt({}); }) .toThrowError(/Argument 'x' passed to 'sqrt' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('rsqrt', test_util_1.ALL_ENVS, function () { it('rsqrt', function () { var a = tf.tensor1d([2, 4]); var r = tf.rsqrt(a); test_util_1.expectNumbersClose(r.get(0), 1 / Math.sqrt(2)); test_util_1.expectNumbersClose(r.get(1), 1 / Math.sqrt(4)); }); it('rsqrt propagates NaNs', function () { var a = tf.tensor1d([1, NaN]); var r = tf.rsqrt(a); test_util_1.expectArraysClose(r, [1 / Math.sqrt(1), NaN]); }); it('gradients: Scalar', function () { var a = tf.scalar(4); var dy = tf.scalar(8); var da = tf.grad(function (a) { return tf.rsqrt(a); })(a, dy); expect(da.shape).toEqual(a.shape); expect(da.dtype).toEqual('float32'); test_util_1.expectArraysClose(da, [(-1 * 8) / (2 * Math.pow(4, 1.5))]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([1, 2, 3, 5]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.rsqrt(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [ -1 * 1 / (2 * Math.pow(1, 1.5)), -1 * 2 / (2 * Math.pow(2, 1.5)), -1 * 3 / (2 * Math.pow(3, 1.5)), -1 * 4 / (2 * Math.pow(5, 1.5)) ], 1e-1); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([3, 1, 2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.rsqrt(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [ -1 * 1 / (2 * Math.pow(3, 1.5)), -1 * 2 / (2 * Math.pow(1, 1.5)), -1 * 3 / (2 * Math.pow(2, 1.5)), -1 * 4 / (2 * Math.pow(3, 1.5)) ], 1e-1); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.rsqrt({}); }) .toThrowError(/Argument 'x' passed to 'rsqrt' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('square', test_util_1.ALL_ENVS, function () { it('1D array', function () { var a = tf.tensor1d([2, 4, Math.sqrt(2)]); var r = tf.square(a); test_util_1.expectArraysClose(r, [4, 16, 2]); }); it('2D array', function () { var a = tf.tensor2d([1, 2, Math.sqrt(2), Math.sqrt(3)], [2, 2]); var r = tf.square(a); expect(r.shape).toEqual([2, 2]); test_util_1.expectArraysClose(r, [1, 4, 2, 3]); }); it('square propagates NaNs', function () { var a = tf.tensor1d([1.5, NaN]); var r = tf.square(a); test_util_1.expectArraysClose(r, [2.25, NaN]); }); it('gradients: Scalar', function () { var a = tf.scalar(5); var dy = tf.scalar(8); var gradients = tf.grad(function (a) { return tf.square(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [2 * 5 * 8]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([-1, 2, 3, -5]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.square(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [-2, 4 * 2, 6 * 3, -10 * 4]); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([-3, 1, 2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.square(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [-6 * 1, 2 * 2, 4 * 3, 6 * 4]); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.square({}); }) .toThrowError(/Argument 'x' passed to 'square' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('reciprocal', test_util_1.ALL_ENVS, function () { it('1D array', function () { var a = tf.tensor1d([2, 3, 0, NaN]); var r = tf.reciprocal(a); test_util_1.expectArraysClose(r, [1 / 2, 1 / 3, Infinity, NaN]); }); it('2D array', function () { var a = tf.tensor2d([1, Infinity, 0, NaN], [2, 2]); var r = tf.reciprocal(a); expect(r.shape).toEqual([2, 2]); test_util_1.expectArraysClose(r, [1 / 1, 0, Infinity, NaN]); }); it('reciprocal propagates NaNs', function () { var a = tf.tensor1d([1.5, NaN]); var r = tf.reciprocal(a); test_util_1.expectArraysClose(r, [1 / 1.5, NaN]); }); it('gradients: Scalar', function () { var a = tf.scalar(5); var dy = tf.scalar(8); var gradients = tf.grad(function (a) { return tf.reciprocal(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [-1 * 8 * (1 / (5 * 5))]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([-1, 2, 3, -5]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.reciprocal(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [ -1 * 1 * (1 / (-1 * -1)), -1 * 2 * (1 / (2 * 2)), -1 * 3 * (1 / (3 * 3)), -1 * 4 * (1 / (-5 * -5)) ]); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([-1, 2, 3, -5], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.reciprocal(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [ -1 * 1 * (1 / (-1 * -1)), -1 * 2 * (1 / (2 * 2)), -1 * 3 * (1 / (3 * 3)), -1 * 4 * (1 / (-5 * -5)) ]); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.reciprocal({}); }) .toThrowError(/Argument 'x' passed to 'reciprocal' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('log', test_util_1.ALL_ENVS, function () { it('log', function () { var a = tf.tensor1d([1, 2]); var r = tf.log(a); test_util_1.expectNumbersClose(r.get(0), Math.log(1)); test_util_1.expectNumbersClose(r.get(1), Math.log(2)); }); it('log propagates NaNs', function () { var a = tf.tensor1d([1, NaN]); var r = tf.log(a); test_util_1.expectArraysClose(r, [Math.log(1), NaN]); }); it('gradients: Scalar', function () { var a = tf.scalar(5); var dy = tf.scalar(3); var gradients = tf.grad(function (a) { return tf.log(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [3 / 5]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([-1, 2, 3, -5]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.log(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [1 / -1, 2 / 2, 3 / 3, 4 / -5]); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([-3, 1, 2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.log(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [1 / -3, 2 / 1, 3 / 2, 4 / 3]); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.log({}); }) .toThrowError(/Argument 'x' passed to 'log' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('log1p', test_util_1.ALL_ENVS, function () { it('log1p', function () { var a = tf.tensor1d([1, 2]); var r = tf.log1p(a); test_util_1.expectNumbersClose(r.get(0), Math.log1p(1)); test_util_1.expectNumbersClose(r.get(1), Math.log1p(2)); }); it('log1p propagates NaNs', function () { var a = tf.tensor1d([1, NaN]); var r = tf.log1p(a); test_util_1.expectArraysClose(r, [Math.log1p(1), NaN]); }); it('gradients: Scalar', function () { var a = tf.scalar(5); var dy = tf.scalar(3); var gradients = tf.grad(function (a) { return tf.log1p(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [3 / (1 + 5)]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([-1, 2, 3, -5]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.log1p(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [Infinity, 2 / (1 + 2), 3 / (1 + 3), 4 / (1 + -5)]); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([-3, 1, 2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.log1p(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [1 / (1 + -3), 2 / (1 + 1), 3 / (1 + 2), 4 / (1 + 3)]); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.log1p({}); }) .toThrowError(/Argument 'x' passed to 'log1p' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('ceil', test_util_1.ALL_ENVS, function () { it('basic', function () { var a = tf.tensor1d([1.5, 2.1, -1.4]); var r = tf.ceil(a); test_util_1.expectNumbersClose(r.get(0), 2); test_util_1.expectNumbersClose(r.get(1), 3); test_util_1.expectNumbersClose(r.get(2), -1); }); it('propagates NaNs', function () { var a = tf.tensor1d([1.5, NaN, -1.4]); var r = tf.ceil(a); test_util_1.expectArraysClose(r, [2, NaN, -1]); }); it('gradients: Scalar', function () { var a = tf.scalar(5.2); var dy = tf.scalar(3); var gradients = tf.grad(function (a) { return tf.ceil(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [0]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([-1.1, 2.6, 3, -5.9]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.ceil(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [0, 0, 0, 0]); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([-3, 1, 2.2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.ceil(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [0, 0, 0, 0]); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.ceil({}); }) .toThrowError(/Argument 'x' passed to 'ceil' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('floor', test_util_1.ALL_ENVS, function () { it('basic', function () { var a = tf.tensor1d([1.5, 2.1, -1.4]); var r = tf.floor(a); test_util_1.expectNumbersClose(r.get(0), 1); test_util_1.expectNumbersClose(r.get(1), 2); test_util_1.expectNumbersClose(r.get(2), -2); }); it('propagates NaNs', function () { var a = tf.tensor1d([1.5, NaN, -1.4]); var r = tf.floor(a); test_util_1.expectArraysClose(r, [1, NaN, -2]); }); it('gradients: Scalar', function () { var a = tf.scalar(5.2); var dy = tf.scalar(3); var gradients = tf.grad(function (a) { return tf.floor(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [0]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([-1.1, 2.6, 3, -5.9]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.floor(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [0, 0, 0, 0]); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([-3, 1, 2.2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.floor(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [0, 0, 0, 0]); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.floor({}); }) .toThrowError(/Argument 'x' passed to 'floor' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('sign', test_util_1.ALL_ENVS, function () { it('basic', function () { var a = tf.tensor1d([1.5, 0, NaN, -1.4]); var r = tf.sign(a); test_util_1.expectNumbersClose(r.get(0), 1); test_util_1.expectNumbersClose(r.get(1), 0); test_util_1.expectNumbersClose(r.get(2), 0); test_util_1.expectNumbersClose(r.get(3), -1); }); it('propagates NaNs', function () { var a = tf.tensor1d([1.5, NaN, -1.4]); var r = tf.sign(a); test_util_1.expectArraysClose(r, [1, 0, -1]); }); it('gradients: Scalar', function () { var a = tf.scalar(5.2); var dy = tf.scalar(3); var gradients = tf.grad(function (a) { return tf.sign(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [0]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([-1.1, 2.6, 3, -5.9]); var dy = tf.tensor1d([-1, 1, 1, -1]); var gradients = tf.grad(function (a) { return tf.sign(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [0, 0, 0, 0]); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([-3, 1, 2.2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.sign(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [0, 0, 0, 0]); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.sign({}); }) .toThrowError(/Argument 'x' passed to 'sign' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('exp', test_util_1.ALL_ENVS, function () { it('exp', function () { var a = tf.tensor1d([1, 2, 0]); var r = tf.exp(a); test_util_1.expectNumbersClose(r.get(0), Math.exp(1)); test_util_1.expectNumbersClose(r.get(1), Math.exp(2)); test_util_1.expectNumbersClose(r.get(2), 1); }); it('exp propagates NaNs', function () { var a = tf.tensor1d([1, NaN, 0]); var r = tf.exp(a); test_util_1.expectArraysClose(r, [Math.exp(1), NaN, 1]); }); it('gradients: Scalar', function () { var a = tf.scalar(0.5); var dy = tf.scalar(3); var gradients = tf.grad(function (a) { return tf.exp(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [3 * Math.exp(0.5)]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([-1, 2, 3, -5]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.exp(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [1 * Math.exp(-1), 2 * Math.exp(2), 3 * Math.exp(3), 4 * Math.exp(-5)], 1e-1); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([-3, 1, 2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.exp(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [1 * Math.exp(-3), 2 * Math.exp(1), 3 * Math.exp(2), 4 * Math.exp(3)], 1e-1); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.exp({}); }) .toThrowError(/Argument 'x' passed to 'exp' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('expm1', test_util_1.ALL_ENVS, function () { it('expm1', function () { var a = tf.tensor1d([1, 2, 0]); var r = tf.expm1(a); test_util_1.expectNumbersClose(r.get(0), Math.expm1(1)); test_util_1.expectNumbersClose(r.get(1), Math.expm1(2)); test_util_1.expectNumbersClose(r.get(2), Math.expm1(0)); }); it('expm1 propagates NaNs', function () { var a = tf.tensor1d([1, NaN, 0]); var r = tf.expm1(a); test_util_1.expectArraysClose(r, [Math.expm1(1), NaN, Math.expm1(0)]); }); it('gradients: Scalar', function () { var a = tf.scalar(0.5); var dy = tf.scalar(3); var gradients = tf.grad(function (a) { return tf.expm1(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [3 * Math.exp(0.5)]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([-1, 2, 3, -5]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.expm1(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [1 * Math.exp(-1), 2 * Math.exp(2), 3 * Math.exp(3), 4 * Math.exp(-5)], 1e-1); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([-3, 1, 2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.expm1(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [1 * Math.exp(-3), 2 * Math.exp(1), 3 * Math.exp(2), 4 * Math.exp(3)], 1e-1); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.expm1({}); }) .toThrowError(/Argument 'x' passed to 'expm1' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('sin', test_util_1.ALL_ENVS, function () { it('basic', function () { var values = [1, -3, 2, 7, -4]; var a = tf.tensor1d(values); var result = tf.sin(a); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = Math.sin(values[i]); } test_util_1.expectArraysClose(result, expected); }); it('propagates NaNs', function () { var a = tf.tensor1d([4, NaN, 0]); var res = tf.sin(a); test_util_1.expectArraysClose(res, [Math.sin(4), NaN, Math.sin(0)]); }); it('gradients: Scalar', function () { var a = tf.scalar(5); var dy = tf.scalar(8); var gradients = tf.grad(function (a) { return tf.sin(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [8 * Math.cos(5)]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([-1, 2, 3, -5]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.sin(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [1 * Math.cos(-1), 2 * Math.cos(2), 3 * Math.cos(3), 4 * Math.cos(-5)], 1e-1); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([-3, 1, 2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.sin(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [1 * Math.cos(-3), 2 * Math.cos(1), 3 * Math.cos(2), 4 * Math.cos(3)], 1e-1); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.sin({}); }) .toThrowError(/Argument 'x' passed to 'sin' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('cos', test_util_1.ALL_ENVS, function () { it('basic', function () { var values = [1, -3, 2, 7, -4]; var a = tf.tensor1d(values); var result = tf.cos(a); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = Math.cos(values[i]); } test_util_1.expectArraysClose(result, expected); }); it('propagates NaNs', function () { var a = tf.tensor1d([4, NaN, 0]); var res = tf.cos(a); test_util_1.expectArraysClose(res, [Math.cos(4), NaN, Math.cos(0)]); }); it('gradients: Scalar', function () { var a = tf.scalar(5); var dy = tf.scalar(8); var gradients = tf.grad(function (a) { return tf.cos(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [8 * Math.sin(5) * -1]); }); it('gradients: Tensor1D', function () { var a = tf.tensor1d([-1, 2, 3, -5]); var dy = tf.tensor1d([1, 2, 3, 4]); var gradients = tf.grad(function (a) { return tf.cos(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [ 1 * Math.sin(-1) * -1, 2 * Math.sin(2) * -1, 3 * Math.sin(3) * -1, 4 * Math.sin(-5) * -1 ], 1e-1); }); it('gradients: Tensor2D', function () { var a = tf.tensor2d([-3, 1, 2, 3], [2, 2]); var dy = tf.tensor2d([1, 2, 3, 4], [2, 2]); var gradients = tf.grad(function (a) { return tf.cos(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [ 1 * Math.sin(-3) * -1, 2 * Math.sin(1) * -1, 3 * Math.sin(2) * -1, 4 * Math.sin(3) * -1 ], 1e-1); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.cos({}); }) .toThrowError(/Argument 'x' passed to 'cos' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('tan', test_util_1.ALL_ENVS, function () { it('basic', function () { var values = [1, -3, 2, 7, -4]; var a = tf.tensor1d(values); var result = tf.tan(a); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = Math.tan(values[i]); } test_util_1.expectArraysClose(result, expected); }); it('propagates NaNs', function () { var a = tf.tensor1d([4, NaN, 0]); var res = tf.tan(a); test_util_1.expectArraysClose(res, [Math.tan(4), NaN, Math.tan(0)]); }); it('gradients: Scalar', function () { var a = tf.scalar(0.5); var dy = tf.scalar(8); var gradients = tf.grad(function (a) { return tf.tan(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [8 / (Math.cos(0.5) * Math.cos(0.5))]); }); it('gradients: Tensor1D', function () { var aValues = [-1, 2, 3, -5]; var dyValues = [1, 2, 3, 4]; var a = tf.tensor1d(aValues); var dy = tf.tensor1d(dyValues); var gradients = tf.grad(function (a) { return tf.tan(a); })(a, dy); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = dyValues[i] / (Math.cos(aValues[i]) * Math.cos(aValues[i])); } expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, expected); }); it('gradients: Tensor2D', function () { var aValues = [-3, 1, 2, 3]; var dyValues = [1, 2, 3, 4]; var a = tf.tensor2d(aValues, [2, 2]); var dy = tf.tensor2d(dyValues, [2, 2]); var gradients = tf.grad(function (a) { return tf.tan(a); })(a, dy); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = dyValues[i] / (Math.cos(aValues[i]) * Math.cos(aValues[i])); } expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, expected); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.tan({}); }) .toThrowError(/Argument 'x' passed to 'tan' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('asin', test_util_1.ALL_ENVS, function () { it('basic', function () { var values = [.1, -3, 2, 7, -4]; var a = tf.tensor1d(values); var result = tf.asin(a); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = Math.asin(values[i]); } test_util_1.expectArraysClose(result, expected); }); it('propagates NaNs', function () { var a = tf.tensor1d([4, NaN, 0]); var res = tf.asin(a); test_util_1.expectArraysClose(res, [Math.asin(4), NaN, Math.asin(0)]); }); it('gradients: Scalar', function () { var a = tf.scalar(0.5); var dy = tf.scalar(8); var gradients = tf.grad(function (a) { return tf.asin(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [8 / Math.sqrt(1 - (0.5 * 0.5))]); }); it('gradients: Tensor1D', function () { var aValues = [-0.1, 0.2, 0.3, -0.5]; var dyValues = [1, 2, 3, 4]; var a = tf.tensor1d(aValues); var dy = tf.tensor1d(dyValues); var gradients = tf.grad(function (a) { return tf.asin(a); })(a, dy); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = dyValues[i] / Math.sqrt(1 - (aValues[i] * aValues[i])); } expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, expected); }); it('gradients: Tensor2D', function () { var aValues = [-0.3, 0.1, 0.2, 0.3]; var dyValues = [1, 2, 3, 4]; var a = tf.tensor2d(aValues, [2, 2]); var dy = tf.tensor2d(dyValues, [2, 2]); var gradients = tf.grad(function (a) { return tf.asin(a); })(a, dy); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = dyValues[i] / Math.sqrt(1 - (aValues[i] * aValues[i])); } expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, expected); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.asin({}); }) .toThrowError(/Argument 'x' passed to 'asin' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('acos', test_util_1.ALL_ENVS, function () { it('basic', function () { var values = [.1, -3, 2, 7, -4]; var a = tf.tensor1d(values); var result = tf.acos(a); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = Math.acos(values[i]); } test_util_1.expectArraysClose(result, expected); }); it('propagates NaNs', function () { var a = tf.tensor1d([4, NaN, 0]); var res = tf.acos(a); test_util_1.expectArraysClose(res, [Math.acos(4), NaN, Math.acos(0)]); }); it('gradients: Scalar', function () { var a = tf.scalar(0.5); var dy = tf.scalar(8); var gradients = tf.grad(function (a) { return tf.acos(a); })(a, dy); expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, [(-1 * 8) / Math.sqrt(1 - (0.5 * 0.5))]); }); it('gradients: Tensor1D', function () { var aValues = [-0.1, 0.2, 0.3, -0.5]; var dyValues = [1, 2, 3, 4]; var a = tf.tensor1d(aValues); var dy = tf.tensor1d(dyValues); var gradients = tf.grad(function (a) { return tf.acos(a); })(a, dy); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = (-1 * dyValues[i]) / Math.sqrt(1 - (aValues[i] * aValues[i])); } expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, expected); }); it('gradients: Tensor2D', function () { var aValues = [-0.3, 0.1, 0.2, 0.3]; var dyValues = [1, 2, 3, 4]; var a = tf.tensor2d(aValues, [2, 2]); var dy = tf.tensor2d(dyValues, [2, 2]); var gradients = tf.grad(function (a) { return tf.acos(a); })(a, dy); var expected = []; for (var i = 0; i < a.size; i++) { expected[i] = (-1 * dyValues[i]) / Math.sqrt(1 - (aValues[i] * aValues[i])); } expect(gradients.shape).toEqual(a.shape); expect(gradients.dtype).toEqual('float32'); test_util_1.expectArraysClose(gradients, expected); }); it('throws when passed a non-tensor', function () { expect(function () { return tf.acos({}); }) .toThrowError(/Argument 'x' passed to 'acos' must be a Tensor/); }); }); jasmine_util_1.describeWithFlags('atan', test_util_1.ALL_ENVS, function () { it('basic', function () { var values = [1, -3, 2, 7, -4];