Skip to content
This repository was archived by the owner on Aug 15, 2019. It is now read-only.

Commit 25a532b

Browse files
oveddanoveddan
committed
Per feedback, changed order of parameters to match
Tensorflow api. Added dataFormat parameter to conv1d, conv2d, and depthwiseConv2d, but did not yet implement that parameter; it defaults to a value and cannot be a different value until the functionality is implemented
1 parent 321481d commit 25a532b

File tree

5 files changed

+106
-61
lines changed

5 files changed

+106
-61
lines changed

‎src/math.ts

Lines changed: 9 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -344,32 +344,35 @@ export class NDArrayMath {
344344
/** @deprecated */
345345
conv1d<T extends Tensor2D|Tensor3D>(
346346
input: T, filter: Tensor3D, bias: Tensor1D|null, stride: number,
347-
dilation: number, pad: 'valid'|'same'|number,
347+
pad: 'valid'|'same'|number, dataFormat: 'NWC'|'NCW' = 'NWC', dilation = 1,
348348
dimRoundingMode?: 'floor'|'round'|'ceil'): T {
349349
if (bias != null) {
350350
util.assert(
351351
bias.rank === 1,
352352
`Error in conv1d: bias must be rank 1, but got rank ` +
353353
`${bias.rank}.`);
354354
}
355-
const res =
356-
ops.conv1d(input, filter, stride, dilation, pad, dimRoundingMode);
355+
const res = ops.conv1d(
356+
input, filter, stride, pad, dataFormat, dilation, dimRoundingMode);
357357
return res.add(bias) as T;
358358
}
359359

360360
/** @deprecated */
361361
conv2d<T extends Tensor3D|Tensor4D>(
362362
x: T, filter: Tensor4D, bias: Tensor1D|null,
363-
strides: [number, number]|number, dilations: [number, number]|number,
364-
pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {
363+
strides: [number, number]|number, pad: 'valid'|'same'|number,
364+
dataFormat: 'NHWC'|'NCHW' = 'NHWC',
365+
dilations: [number, number]|number = [1, 1],
366+
dimRoundingMode?: 'floor'|'round'|'ceil'): T {
365367
if (bias != null) {
366368
util.assert(
367369
bias.rank === 1,
368370
`Error in conv2d: bias must be rank 1, but got rank ` +
369371
`${bias.rank}.`);
370372
}
371373

372-
const res = ops.conv2d(x, filter, strides, dilations, pad, dimRoundingMode);
374+
const res = ops.conv2d(
375+
x, filter, strides, pad, dataFormat, dilations, dimRoundingMode);
373376
return res.add(bias) as T;
374377
}
375378

‎src/ops/conv.ts

Lines changed: 37 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -40,15 +40,22 @@ export class ConvOps {
4040
* - For more info, see this guide:
4141
* [https://www.tensorflow.org/api_guides/python/nn#Convolution](
4242
* https://www.tensorflow.org/api_guides/python/nn#Convolution)
43+
* @param dataFormat An optional string from "NWC", "NCW". Defaults to "NWC",
44+
* the data is stored in the order of [batch, in_width, in_channels]. Only
45+
* "NWC" is currently supported.
46+
* @param dilation The dilation rate in which we sample input values in
47+
* atrous convolution. Defaults to `1`. If it is greater than 1, then
48+
* stride must be `1`.
4349
* @param dimRoundingMode The rounding mode used when computing output
4450
* dimensions if pad is a number. If none is provided, it will not round
4551
* and error if the output is of fractional size.
4652
*/
4753
@doc({heading: 'Operations', subheading: 'Convolution'})
4854
@operation
4955
static conv1d<T extends Tensor2D|Tensor3D>(
50-
input: T, filter: Tensor3D, stride: number, dilation: number,
51-
pad: 'valid'|'same'|number, dimRoundingMode?: 'floor'|'round'|'ceil'): T {
56+
input: T, filter: Tensor3D, stride: number, pad: 'valid'|'same'|number,
57+
dataFormat: 'NWC'|'NCW' = 'NWC', dilation = 1,
58+
dimRoundingMode?: 'floor'|'round'|'ceil'): T {
5259
let input3D = input as Tensor3D;
5360
let reshapedTo3D = false;
5461
if (input.rank === 2) {
@@ -78,6 +85,10 @@ export class ConvOps {
7885
eitherStridesOrDilationsAreOne(stride, dilation),
7986
'Error in conv1D: Either stride or dilation must be 1.' +
8087
`Got stride ${stride} and dilation '${dilation}'`);
88+
util.assert(
89+
dataFormat === 'NWC',
90+
`Error in conv1d: got dataFormat of ${
91+
dataFormat} but only NWC is currently supported.`);
8192

8293
const filter4D =
8394
filter.as4D(1, filter.shape[0], filter.shape[1], filter.shape[2]);
@@ -86,8 +97,11 @@ export class ConvOps {
8697
const strides: [number, number] = [1, stride];
8798
const dilations: [number, number] = [1, dilation];
8899

100+
const conv2dDataFormat = 'NHWC';
101+
89102
const res = ConvOps.conv2d(
90-
input4D, filter4D, strides, dilations, pad, dimRoundingMode);
103+
input4D, filter4D, strides, pad, conv2dDataFormat, dilations,
104+
dimRoundingMode);
91105

92106
if (reshapedTo3D) {
93107
return res.as2D(res.shape[2], res.shape[3]) as T;
@@ -105,11 +119,6 @@ export class ConvOps {
105119
* `[filterHeight, filterWidth, inDepth, outDepth]`.
106120
* @param strides The strides of the convolution: `[strideHeight,
107121
* strideWidth]`.
108-
* @param dilations The dilation rates: `[dilationHeight, dilationWidth]`
109-
* in which we sample input values across the height and width dimensions
110-
* in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single
111-
* number, then `dilationHeight == dilationWidth`. If it is greater than
112-
* 1, then all values of `strides` must be 1.
113122
* @param pad The type of padding algorithm.
114123
* - `same` and stride 1: output will be of same size as input,
115124
* regardless of filter size.
@@ -118,6 +127,15 @@ export class ConvOps {
118127
* - For more info, see this guide:
119128
* [https://www.tensorflow.org/api_guides/python/nn#Convolution](
120129
* https://www.tensorflow.org/api_guides/python/nn#Convolution)
130+
* @param dataFormat: An optional string from: "NHWC", "NCHW". Defaults to
131+
* "NHWC". Specify the data format of the input and output data. With the
132+
* default format "NHWC", the data is stored in the order of: [batch,
133+
* height, width, channels]. Only "NHWC" is currently supported.
134+
* @param dilations The dilation rates: `[dilationHeight, dilationWidth]`
135+
* in which we sample input values across the height and width dimensions
136+
* in atrous convolution. Defaults to `[1, 1]`. If `dilations` is a single
137+
* number, then `dilationHeight == dilationWidth`. If it is greater than
138+
* 1, then all values of `strides` must be 1.
121139
* @param dimRoundingMode The rounding mode used when computing output
122140
* dimensions if pad is a number. If none is provided, it will not round
123141
* and error if the output is of fractional size.
@@ -126,7 +144,8 @@ export class ConvOps {
126144
@operation
127145
static conv2d<T extends Tensor3D|Tensor4D>(
128146
x: T, filter: Tensor4D, strides: [number, number]|number,
129-
dilations: [number, number]|number = [1, 1], pad: 'valid'|'same'|number,
147+
pad: 'valid'|'same'|number, dataFormat: 'NHWC'|'NCHW' = 'NHWC',
148+
dilations: [number, number]|number = [1, 1],
130149
dimRoundingMode?: 'floor'|'round'|'ceil'): T {
131150
let x4D = x as Tensor4D;
132151
let reshapedTo4D = false;
@@ -157,6 +176,10 @@ export class ConvOps {
157176
eitherStridesOrDilationsAreOne(strides, dilations),
158177
'Error in conv2D: Either strides or dilations must be 1.' +
159178
`Got strides ${strides} and dilations '${dilations}'`);
179+
util.assert(
180+
dataFormat === 'NHWC',
181+
`Error in conv2d: got dataFormat of ${
182+
dataFormat} but only NHWC is currently supported.`);
160183

161184
const convInfo = conv_util.computeConv2DInfo(
162185
x4D.shape, filter.shape, strides, dilations, pad, dimRoundingMode);
@@ -391,6 +414,10 @@ export class ConvOps {
391414
* in atrous convolution. Defaults to `[1, 1]`. If `rate` is a single
392415
* number, then `dilationHeight == dilationWidth`. If it is greater than
393416
* 1, then all values of `strides` must be 1.
417+
* @param dataFormat: An optional string from: "NHWC", "NCHW". Defaults to
418+
* "NHWC". Specify the data format of the input and output data. With the
419+
* default format "NHWC", the data is stored in the order of: [batch,
420+
* height, width, channels]. Only "NHWC" is currently supported.
394421
* @param dimRoundingMode The rounding mode used when computing output
395422
* dimensions if pad is a number. If none is provided, it will not round
396423
* and error if the output is of fractional size.
@@ -400,6 +427,7 @@ export class ConvOps {
400427
static depthwiseConv2d<T extends Tensor3D|Tensor4D>(
401428
input: T, filter: Tensor4D, strides: [number, number]|number,
402429
pad: 'valid'|'same'|number, dilations: [number, number]|number = [1, 1],
430+
dataFormat: 'NHWC'|'NCHW' = 'NHWC',
403431
dimRoundingMode?: 'floor'|'round'|'ceil'): T {
404432
let input4D = input as Tensor4D;
405433
let reshapedTo4D = false;

‎src/ops/conv1d_test.ts

Lines changed: 28 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -28,12 +28,13 @@ describeWithFlags('conv1d', ALL_ENVS, () => {
2828
const fSize = 1;
2929
const pad = 'same';
3030
const stride = 1;
31+
const dataFormat = 'NWC';
3132
const dilation = 1;
3233

3334
const x = dl.tensor3d([1, 2, 3, 4], inputShape);
3435
const w = dl.tensor3d([3], [fSize, inputDepth, outputDepth]);
3536

36-
const result = dl.conv1d(x, w, stride, dilation, pad);
37+
const result = dl.conv1d(x, w, stride, pad, dataFormat, dilation);
3738

3839
expect(result.shape).toEqual([2, 2, 1]);
3940
expectArraysClose(result, [3, 6, 9, 12]);
@@ -46,12 +47,13 @@ describeWithFlags('conv1d', ALL_ENVS, () => {
4647
const fSize = 2;
4748
const pad = 'valid';
4849
const stride = 1;
50+
const dataFormat = 'NWC';
4951
const dilation = 1;
5052

5153
const x = dl.tensor2d([1, 2, 3, 4], inputShape);
5254
const w = dl.tensor3d([2, 1], [fSize, inputDepth, outputDepth]);
5355

54-
const result = dl.conv1d(x, w, stride, dilation, pad);
56+
const result = dl.conv1d(x, w, stride, pad, dataFormat, dilation);
5557

5658
expect(result.shape).toEqual([3, 1]);
5759
expectArraysClose(result, [4, 7, 10]);
@@ -65,19 +67,20 @@ describeWithFlags('conv1d', ALL_ENVS, () => {
6567
const fSizeDilated = 3;
6668
const pad = 'valid';
6769
const stride = 1;
70+
const dataFormat = 'NWC';
6871
const dilation = 2;
6972
const dilationWEffective = 1;
7073

7174
const x = dl.tensor2d([1, 2, 3, 4], inputShape);
7275
const w = dl.tensor3d([2, 1], [fSize, inputDepth, outputDepth]);
7376
// adding a dilation rate is equivalent to using a filter
7477
// with 0s for the dilation rate
75-
const wDilated = dl.tensor3d(
76-
[2, 0, 1], [fSizeDilated, inputDepth, outputDepth]);
78+
const wDilated =
79+
dl.tensor3d([2, 0, 1], [fSizeDilated, inputDepth, outputDepth]);
7780

78-
const result = dl.conv1d(x, w, stride, dilation, pad);
79-
const expectedResult = dl.conv1d(
80-
x, wDilated, stride, dilationWEffective, pad);
81+
const result = dl.conv1d(x, w, stride, pad, dataFormat, dilation);
82+
const expectedResult =
83+
dl.conv1d(x, wDilated, stride, pad, dataFormat, dilationWEffective);
8184

8285
expect(result.shape).toEqual(expectedResult.shape);
8386
expectArraysClose(result, expectedResult);
@@ -91,20 +94,21 @@ describeWithFlags('conv1d', ALL_ENVS, () => {
9194
const fSizeDilated = 7;
9295
const pad = 'valid';
9396
const stride = 1;
97+
const dataFormat = 'NWC';
9498
const dilation = 3;
9599
const dilationWEffective = 1;
96100

97101
const x = dl.tensor2d(
98-
[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14], inputShape);
102+
[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14], inputShape);
99103
const w = dl.tensor3d([3, 2, 1], [fSize, inputDepth, outputDepth]);
100104
// adding a dilation rate is equivalent to using a filter
101105
// with 0s for the dilation rate
102106
const wDilated = dl.tensor3d(
103-
[3, 0, 0, 2, 0, 0, 1], [fSizeDilated, inputDepth, outputDepth]);
107+
[3, 0, 0, 2, 0, 0, 1], [fSizeDilated, inputDepth, outputDepth]);
104108

105-
const result = dl.conv1d(x, w, stride, dilation, pad);
106-
const expectedResult = dl.conv1d(
107-
x, wDilated, stride, dilationWEffective, pad);
109+
const result = dl.conv1d(x, w, stride, pad, dataFormat, dilation);
110+
const expectedResult =
111+
dl.conv1d(x, wDilated, stride, pad, dataFormat, dilationWEffective);
108112

109113
expect(result.shape).toEqual(expectedResult.shape);
110114
expectArraysClose(result, expectedResult);
@@ -116,27 +120,31 @@ describeWithFlags('conv1d', ALL_ENVS, () => {
116120
const fSize = 2;
117121
const pad = 0;
118122
const stride = 1;
123+
const dataFormat = 'NWC';
119124
const dilation = 1;
120125

121126
// tslint:disable-next-line:no-any
122127
const x: any = dl.tensor2d([1, 2, 3, 4], [2, 2]);
123128
const w = dl.tensor3d([3, 1], [fSize, inputDepth, outputDepth]);
124129

125-
expect(() => dl.conv1d(x, w, stride, dilation, pad)).toThrowError();
130+
expect(() => dl.conv1d(x, w, stride, pad, dataFormat, dilation))
131+
.toThrowError();
126132
});
127133

128134
it('throws when weights is not rank 3', () => {
129135
const inputDepth = 1;
130136
const inputShape: [number, number, number] = [2, 2, inputDepth];
131137
const pad = 0;
132138
const stride = 1;
139+
const dataFormat = 'NWC';
133140
const dilation = 1;
134141

135142
const x = dl.tensor3d([1, 2, 3, 4], inputShape);
136143
// tslint:disable-next-line:no-any
137144
const w: any = dl.tensor4d([3, 1, 5, 0], [2, 2, 1, 1]);
138145

139-
expect(() => dl.conv1d(x, w, stride, dilation, pad)).toThrowError();
146+
expect(() => dl.conv1d(x, w, stride, pad, dataFormat, dilation))
147+
.toThrowError();
140148
});
141149

142150
it('throws when x depth does not match weight depth', () => {
@@ -147,12 +155,14 @@ describeWithFlags('conv1d', ALL_ENVS, () => {
147155
const fSize = 2;
148156
const pad = 0;
149157
const stride = 1;
158+
const dataFormat = 'NWC';
150159
const dilation = 1;
151160

152161
const x = dl.tensor3d([1, 2, 3, 4], inputShape);
153162
const w = dl.randomNormal<Rank.R3>([fSize, wrongInputDepth, outputDepth]);
154163

155-
expect(() => dl.conv1d(x, w, stride, dilation, pad)).toThrowError();
164+
expect(() => dl.conv1d(x, w, stride, pad, dataFormat, dilation))
165+
.toThrowError();
156166
});
157167

158168
it('throws when both stride and dilation are greater than 1', () => {
@@ -162,11 +172,13 @@ describeWithFlags('conv1d', ALL_ENVS, () => {
162172
const fSize = 1;
163173
const pad = 'same';
164174
const stride = 2;
175+
const dataFormat = 'NWC';
165176
const dilation = 2;
166177

167178
const x = dl.tensor3d([1, 2, 3, 4], inputShape);
168179
const w = dl.tensor3d([3], [fSize, inputDepth, outputDepth]);
169180

170-
expect(() => dl.conv1d(x, w, stride, dilation, pad)).toThrowError();
181+
expect(() => dl.conv1d(x, w, stride, pad, dataFormat, dilation))
182+
.toThrowError();
171183
});
172184
});

0 commit comments

Comments
 (0)