This is an automated email from the ASF dual-hosted git repository.
mboehm7 pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/systemds.git
The following commit(s) were added to refs/heads/master by this push:
new eeffef6 [SYSTEMDS-2812] Added conv1d layer to nn library (via conv2d)
eeffef6 is described below
commit eeffef63cd8e2a5563c31f067cfe21e73a08e32d
Author: sebwrede <[email protected]>
AuthorDate: Sun Jan 31 00:28:02 2021 +0100
[SYSTEMDS-2812] Added conv1d layer to nn library (via conv2d)
Closes #1174.
---
scripts/nn/layers/conv1d.dml | 100 +++++++++
.../sysds/test/functions/dnn/Conv1DTest.java | 247 +++++++++++++++++++++
.../functions/tensor/Conv1DBackwardDataTest.dml | 52 +++++
.../functions/tensor/Conv1DBackwardFilterTest.dml | 52 +++++
src/test/scripts/functions/tensor/Conv1DTest.dml | 53 +++++
5 files changed, 504 insertions(+)
diff --git a/scripts/nn/layers/conv1d.dml b/scripts/nn/layers/conv1d.dml
new file mode 100644
index 0000000..f1beeeb
--- /dev/null
+++ b/scripts/nn/layers/conv1d.dml
@@ -0,0 +1,100 @@
+#-------------------------------------------------------------
+#
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements. See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership. The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied. See the License for the
+# specific language governing permissions and limitations
+# under the License.
+#
+#-------------------------------------------------------------
+
+source("scripts/nn/util.dml") as util
+
+forward = function(matrix[double] input, matrix[double] filter, int pad, int
stride,
+ int numInput, int numChannels, int inputWidth, int numFilters, int
filterSize)
+ return (matrix[double] out)
+{
+ /*
+ * Computes the forward pass for a 1D spatial convolutional layer
+ * by reshaping the input to fit conv2d.
+ *
+ * Inputs:
+ * - input: Inputs, of shape (N, C*W).
+ * - filter: Weights, of shape (F, C*W).
+ * - pad: Padding for left and right sides of input elements.
+ * - stride: Stride for moving filter.
+ * - numInput: Number of input elements N
+ * - numChannels: Number of input channels (dimensionality of input depth).
+ * - inputWidth: Input width.
+ * - numFilters: Number of filters F
+ * - filterSize: Filter width.
+ *
+ * Outputs:
+ * - out: Outputs, of shape (N, F*Wout).
+ */
+ out = conv2d(input, filter, padding=[0,pad], stride=[1, stride],
+ input_shape=[numInput,numChannels,1,inputWidth],
filter_shape=[numFilters,numChannels,1,filterSize])
+}
+
+backward_data = function(matrix[double] filter, matrix[double] dout, int pad,
int stride,
+ int numInput, int numChannels, int inputWidth, int numFilters, int
filterSize)
+ return (matrix[double] out)
+{
+ /*
+ * Computes the backward pass regarding the input data for a 1D spatial
convolutional layer
+ * by reshaping the input to fit conv2d backward data pass.
+ *
+ * Inputs:
+ * - filter: Weights, of shape (F, C*W).
+ * - dout: Output of the forward pass
+ * - pad: Padding for left and right sides of input elements.
+ * - stride: Stride for moving filter.
+ * - numInput: Number of input elements N
+ * - numChannels: Number of input channels (dimensionality of input depth).
+ * - inputWidth: Input width.
+ * - numFilters: Number of filters F
+ * - filterSize: Filter width.
+ *
+ * Outputs:
+ * - out: gradients based on the input data of the convolution.
+ */
+ out = conv2d_backward_data(filter, dout, stride=[1,stride], padding=[0,pad],
+ input_shape=[numInput,numChannels,1,inputWidth],
filter_shape=[numFilters,numChannels,1,filterSize])
+}
+
+backward_filter = function(matrix[double] input, matrix[double] dout, int pad,
int stride,
+ int numInput, int numChannels, int inputWidth, int numFilters, int
filterSize)
+ return (matrix[double] out)
+{
+ /*
+ * Computes the backward pass regarding the filter for a 1D spatial
convolutional layer
+ * by reshaping the input to fit conv2d backward data pass.
+ *
+ * Inputs:
+ * - input: Inputs, of shape (N, C*W).
+ * - dout: Output of the forward pass
+ * - pad: Padding for left and right sides of input elements.
+ * - stride: Stride for moving filter.
+ * - numInput: Number of input elements N
+ * - numChannels: Number of input channels (dimensionality of input depth).
+ * - inputWidth: Input width.
+ * - numFilters: Number of filters F
+ * - filterSize: Filter width.
+ *
+ * Outputs:
+ * - out: gradients bsaed on the filter of the convolution.
+ */
+ out = conv2d_backward_filter(input, dout, stride=[1,stride],
padding=[0,pad],
+ input_shape=[numInput,numChannels,1,inputWidth],
filter_shape=[numFilters,numChannels,1,filterSize])
+}
diff --git a/src/test/java/org/apache/sysds/test/functions/dnn/Conv1DTest.java
b/src/test/java/org/apache/sysds/test/functions/dnn/Conv1DTest.java
new file mode 100644
index 0000000..9f5add5
--- /dev/null
+++ b/src/test/java/org/apache/sysds/test/functions/dnn/Conv1DTest.java
@@ -0,0 +1,247 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+package org.apache.sysds.test.functions.dnn;
+
+import java.util.HashMap;
+import java.util.stream.IntStream;
+
+import org.junit.Test;
+import org.apache.sysds.api.DMLScript;
+import org.apache.sysds.common.Types.ExecMode;
+import org.apache.sysds.lops.LopProperties.ExecType;
+import org.apache.sysds.runtime.matrix.data.MatrixValue.CellIndex;
+import org.apache.sysds.test.AutomatedTestBase;
+import org.apache.sysds.test.TestConfiguration;
+import org.apache.sysds.test.TestUtils;
+
+public class Conv1DTest extends AutomatedTestBase
+{
+ private final static String TEST_NAME_1 = "Conv1DTest";
+ private final static String TEST_NAME_2 = "Conv1DBackwardDataTest";
+ private final static String TEST_NAME_3 = "Conv1DBackwardFilterTest";
+ private final static String TEST_DIR = "functions/tensor/";
+ private final static String TEST_CLASS_DIR = TEST_DIR +
Conv1DTest.class.getSimpleName() + "/";
+ private final static double epsilon=0.0000000001;
+
+ @Override
+ public void setUp() {
+ TestUtils.clearAssertionInformation();
+ addTestConfiguration(TEST_NAME_1, new
TestConfiguration(TEST_CLASS_DIR, TEST_NAME_1, new String[] {"output"}));
+ addTestConfiguration(TEST_NAME_2, new
TestConfiguration(TEST_CLASS_DIR, TEST_NAME_2, new String[] {"output"}));
+ addTestConfiguration(TEST_NAME_3, new
TestConfiguration(TEST_CLASS_DIR, TEST_NAME_3, new String[] {"output"}));
+ }
+
+ @Test
+ public void testSimpleConv1DDenseSingleBatchSingleChannelSingleFilter(){
+ int numImg = 1; int imgSize = 4; int numChannels = 1; int
numFilters = 1; int filterSize = 2; int stride = 1; int pad = 0;
+ HashMap<CellIndex, Double> expected = new HashMap<>();
+ expected.put(new CellIndex(1,1), 3.0);
+ expected.put(new CellIndex(1,2), 5.0);
+ expected.put(new CellIndex(1,3), 7.0);
+ runConv1DTest(ExecType.CP, imgSize, numImg, numChannels,
numFilters, filterSize, stride, pad, expected, TEST_NAME_1);
+ }
+
+ @Test
+ public void testConv1DDense1() {
+ int numImg = 5; int imgSize = 3; int numChannels = 3; int
numFilters = 3; int filterSize = 2; int stride = 1; int pad = 0;
+ HashMap<CellIndex, Double> expected = new HashMap<>();
+ fillExpected(expected, 1, 6, 21.0, 39.0);
+ fillExpected(expected, 2, 6, 75.0, 93.0);
+ fillExpected(expected, 3, 6, 129.0, 147.0);
+ fillExpected(expected, 4, 6, 183.0, 201.0);
+ fillExpected(expected, 5, 6, 237.0, 255.0);
+
+ runConv1DTest(ExecType.CP, imgSize, numImg, numChannels,
numFilters, filterSize, stride, pad, expected, TEST_NAME_1);
+ }
+
+ @Test
+ public void testConv1DDense2() {
+ int numImg = 1; int imgSize = 10; int numChannels = 4; int
numFilters = 3; int filterSize = 4; int stride = 2; int pad = 0;
+ HashMap<CellIndex, Double> expected = new HashMap<>();
+ fillExpectedRepeated(expected, 3, new
double[]{136.,264.,392.,520.},1);
+ runConv1DTest(ExecType.CP, imgSize, numImg, numChannels,
numFilters, filterSize, stride, pad, expected, TEST_NAME_1);
+ }
+
+ @Test
+ public void testConv1DDense3() {
+ int numImg = 1; int imgSize = 10; int numChannels = 4; int
numFilters = 3; int filterSize = 4; int stride = 2; int pad = 1;
+ HashMap<CellIndex, Double> expected = new HashMap<>();
+ fillExpectedRepeated(expected, 3, new
double[]{78.,200.,328.,456.,414.},1);
+ runConv1DTest(ExecType.CP, imgSize, numImg, numChannels,
numFilters, filterSize, stride, pad, expected, TEST_NAME_1);
+ }
+
+ @Test
+ public void testConv1DDense4() {
+ int numImg = 3; int imgSize = 10; int numChannels = 1; int
numFilters = 3; int filterSize = 2; int stride = 2; int pad = 1;
+ HashMap<CellIndex, Double> expected = new HashMap<>();
+ fillExpectedRepeated(expected,3,new
double[]{1.,5.,9.,13.,17.,10.},1);
+ fillExpectedRepeated(expected,3,new
double[]{11.,25.,29.,33.,37.,20.},2);
+ fillExpectedRepeated(expected,3,new
double[]{21.,45.,49.,53.,57.,30.},3);
+ runConv1DTest(ExecType.CP, imgSize, numImg, numChannels,
numFilters, filterSize, stride, pad, expected, TEST_NAME_1);
+ }
+
+ @Test
+ public void testConv1DDense5() {
+ int numImg = 3; int imgSize = 8; int numChannels = 2; int
numFilters = 3; int filterSize = 3; int stride = 1; int pad = 2;
+ HashMap<CellIndex, Double> expected = new HashMap<>();
+ fillExpectedRepeated(expected,3,new
double[]{3.,10.,21.,33.,45.,57.,69.,81.,58.,31.},1);
+ fillExpectedRepeated(expected,3,new
double[]{35.,74.,117.,129.,141.,153.,165.,177.,122.,63.},2);
+ fillExpectedRepeated(expected,3,new
double[]{67.,138.,213.,225.,237.,249.,261.,273.,186.,95.},3);
+ runConv1DTest(ExecType.CP, imgSize, numImg, numChannels,
numFilters, filterSize, stride, pad, expected, TEST_NAME_1);
+ }
+
+ @Test
+ public void testConv1DDense6() {
+ int numImg = 1; int imgSize = 10; int numChannels = 4; int
numFilters = 3; int filterSize = 4; int stride = 1; int pad = 0;
+ HashMap<CellIndex, Double> expected = new HashMap<>();
+ fillExpectedRepeated(expected,3,new
double[]{136.,200.,264.,328.,392.,456.,520.},1);
+ runConv1DTest(ExecType.CP, imgSize, numImg, numChannels,
numFilters, filterSize, stride, pad, expected, TEST_NAME_1);
+ }
+
+ @Test
+ public void testConv1DDense7() {
+ int numImg = 3; int imgSize = 64; int numChannels = 1; int
numFilters = 3; int filterSize = 2; int stride = 1; int pad = 0;
+ HashMap<CellIndex, Double> expected = new HashMap<>();
+ double[] firstExpected = IntStream.iterate(3,n ->
n+2).limit(63).mapToDouble(i->(double)i).toArray();
+ double[] secondExpected = IntStream.iterate(131,n ->
n+2).limit(63).mapToDouble(i->(double)i).toArray();
+ double[] thirdExpected = IntStream.iterate(259,n ->
n+2).limit(63).mapToDouble(i->(double)i).toArray();
+ fillExpectedRepeated(expected,3,firstExpected,1);
+ fillExpectedRepeated(expected,3,secondExpected,2);
+ fillExpectedRepeated(expected,3,thirdExpected,3);
+ runConv1DTest(ExecType.CP, imgSize, numImg, numChannels,
numFilters, filterSize, stride, pad, expected, TEST_NAME_1);
+ }
+
+ @Test
+ public void testConv1DDense1SP()
+ {
+ int numImg = 5; int imgSize = 3; int numChannels = 3; int
numFilters = 6; int filterSize = 2; int stride = 1; int pad = 0;
+ HashMap<CellIndex, Double> expected = new HashMap<>();
+ fillExpected(expected, 1, 12, 21.0, 39.0);
+ fillExpected(expected, 2, 12, 75.0, 93.0);
+ fillExpected(expected, 3, 12, 129.0, 147.0);
+ fillExpected(expected, 4, 12, 183.0, 201.0);
+ fillExpected(expected, 5, 12, 237.0, 255.0);
+ runConv1DTest(ExecType.SPARK, imgSize, numImg, numChannels,
numFilters, filterSize, stride, pad, expected, TEST_NAME_1);
+ }
+
+ @Test
+ public void testConv1DBackwardDataDense1() {
+ int numImg = 5; int imgSize = 3; int numChannels = 3; int
numFilters = 3; int filterSize = 1; int stride = 1; int pad = 0;
+ HashMap<CellIndex, Double> expected = new HashMap<>();
+ fillExpectedRepeated(expected,3,new double[]{6.,15.,24.},1);
+ fillExpectedRepeated(expected,3,new double[]{33.,42.,51.},2);
+ fillExpectedRepeated(expected,3,new double[]{60.,69.,78.},3);
+ fillExpectedRepeated(expected,3,new double[]{87.,96.,105.},4);
+ fillExpectedRepeated(expected,3,new double[]{114.,123.,132.},5);
+
+ runConv1DTest(ExecType.CP, imgSize, numImg, numChannels,
numFilters, filterSize, stride, pad, expected, TEST_NAME_2);
+ }
+
+ @Test
+ public void testConv1DBackwardFilterDense1() {
+ int numImg = 2; int imgSize = 3; int numChannels = 2; int
numFilters = 3; int filterSize = 1; int stride = 1; int pad = 0;
+ HashMap<CellIndex, Double> expected = new HashMap<>();
+ fillExpectedRepeatedCol(expected,3,new double[]{608.,686.});
+
+ runConv1DTest(ExecType.CP, imgSize, numImg, numChannels,
numFilters, filterSize, stride, pad, expected, TEST_NAME_3);
+ }
+
+ @Test
+ public void testConv1DBackwardFilterDense2() {
+ int numImg = 2; int imgSize = 3; int numChannels = 2; int
numFilters = 3; int filterSize = 2; int stride = 1; int pad = 0;
+ HashMap<CellIndex, Double> expected = new HashMap<>();
+ fillExpectedRepeatedCol(expected,3,new
double[]{680.,888.,784.,992.});
+
+ runConv1DTest(ExecType.CP, imgSize, numImg, numChannels,
numFilters, filterSize, stride, pad, expected, TEST_NAME_3);
+ }
+
+ private static void fillExpected(HashMap<CellIndex, Double> expected,
+ int rowNum, int rowLength, double value1, double value2)
+ {
+ for ( int m = 1; m <= rowLength; m+=2){
+ expected.put(new CellIndex(rowNum,m), value1);
+ expected.put(new CellIndex(rowNum,m+1), value2);
+ }
+ }
+
+ private static void fillExpectedRepeated(HashMap<CellIndex, Double>
expected,
+ int repetitionNum, double[] values, int row)
+ {
+ int colPointer = 1;
+ for (int i = 1; i <= repetitionNum;i++){
+ for(double value : values) {
+ expected.put(new CellIndex(row, colPointer),
value);
+ colPointer++;
+ }
+ }
+ }
+
+ private static void fillExpectedRepeatedCol(HashMap<CellIndex, Double>
expected,int repetitionRows, double[] values){
+ for ( int i = 1; i <= repetitionRows; i++){
+ for ( int j = 1; j <= values.length; j++ ){
+ expected.put(new CellIndex(i,j), values[j-1]);
+ }
+ }
+ }
+
+ public void runConv1DTest( ExecType et, int imgSize, int numImg, int
numChannels, int numFilters,
+ int filterSize, int stride, int pad, HashMap<CellIndex, Double>
expected, String TEST_NAME)
+ {
+ ExecMode platformOld = rtplatform;
+ switch( et ){
+ case SPARK: rtplatform = ExecMode.SPARK; break;
+ default: rtplatform = ExecMode.HYBRID; break;
+ }
+ boolean sparkConfigOld = DMLScript.USE_LOCAL_SPARK_CONFIG;
+ if( rtplatform == ExecMode.SPARK )
+ DMLScript.USE_LOCAL_SPARK_CONFIG = true;
+
+ try
+ {
+ getAndLoadTestConfiguration(TEST_NAME);
+
+ String SCRIPT_HOME = SCRIPT_DIR + TEST_DIR + TEST_NAME;
+ fullDMLScriptName = SCRIPT_HOME + ".dml";
+
+ programArgs = new String[] {
+ "-nvargs",
+ "imgSize=" + imgSize,
+ "numImg=" + numImg,
+ "numChannels=" + numChannels,
+ "numFilters=" + numFilters,
+ "filterSize=" + filterSize,
+ "stride=" + stride,
+ "pad=" + pad,
+ "output=" + output("output")
+ };
+
+ // Run DML
+ runTest(true, false, null, -1);
+
+ HashMap<CellIndex, Double> dmlfile =
readDMLMatrixFromOutputDir("output");
+ System.out.println(dmlfile.toString());
+ if ( expected != null)
+ TestUtils.compareMatrices(dmlfile, expected,
epsilon, "B-DML", "B-R");
+ }
+ finally {
+ rtplatform = platformOld;
+ DMLScript.USE_LOCAL_SPARK_CONFIG = sparkConfigOld;
+ }
+ }
+}
diff --git a/src/test/scripts/functions/tensor/Conv1DBackwardDataTest.dml
b/src/test/scripts/functions/tensor/Conv1DBackwardDataTest.dml
new file mode 100644
index 0000000..0a86377
--- /dev/null
+++ b/src/test/scripts/functions/tensor/Conv1DBackwardDataTest.dml
@@ -0,0 +1,52 @@
+#-------------------------------------------------------------
+#
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements. See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership. The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied. See the License for the
+# specific language governing permissions and limitations
+# under the License.
+#
+#-------------------------------------------------------------
+
+source("scripts/nn/layers/conv1d.dml") as conv1d
+
+# Build input
+rowBuild = seq(1,$numChannels*$imgSize,$numChannels)
+if ( $numChannels > 1 & $imgSize > 1 ){
+ for ( i in 2:$numChannels ){
+ rowBuild = rbind(rowBuild, seq(i,$numChannels*$imgSize,$numChannels))
+ }
+}
+
+colBuild = rowBuild
+if ( $numImg > 1 ){
+ for ( j in 2:$numImg ){
+ rowBuild =
seq((j-1)*$imgSize*$numChannels+1,j*$imgSize*$numChannels,$numChannels)
+ if ( $numChannels > 1 ){
+ for ( i in 2:$numChannels ){
+ rowBuild = rbind(rowBuild,
seq((j-1)*$imgSize*$numChannels+i,j*$numChannels*$imgSize,$numChannels))
+ }
+ }
+ colBuild = cbind(colBuild, rowBuild)
+ }
+}
+
+# Set input variables
+x = t(colBuild)
+w=matrix(1,rows=$numFilters, cols=$numChannels*$filterSize)
+
+output = conv1d::backward_data(w, x, $pad, $stride, $numImg, $numChannels,
$imgSize, $numFilters, $filterSize)
+
+#Write output
+write(output, $output)
diff --git a/src/test/scripts/functions/tensor/Conv1DBackwardFilterTest.dml
b/src/test/scripts/functions/tensor/Conv1DBackwardFilterTest.dml
new file mode 100644
index 0000000..7364e93
--- /dev/null
+++ b/src/test/scripts/functions/tensor/Conv1DBackwardFilterTest.dml
@@ -0,0 +1,52 @@
+#-------------------------------------------------------------
+#
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements. See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership. The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied. See the License for the
+# specific language governing permissions and limitations
+# under the License.
+#
+#-------------------------------------------------------------
+
+source("scripts/nn/layers/conv1d.dml") as conv1d
+
+# Build input
+rowBuild = seq(1,$numChannels*$imgSize,$numChannels)
+if ( $numChannels > 1 & $imgSize > 1 ){
+ for ( i in 2:$numChannels ){
+ rowBuild = rbind(rowBuild, seq(i,$numChannels*$imgSize,$numChannels))
+ }
+}
+
+colBuild = rowBuild
+if ( $numImg > 1 ){
+ for ( j in 2:$numImg ){
+ rowBuild =
seq((j-1)*$imgSize*$numChannels+1,j*$imgSize*$numChannels,$numChannels)
+ if ( $numChannels > 1 ){
+ for ( i in 2:$numChannels ){
+ rowBuild = rbind(rowBuild,
seq((j-1)*$imgSize*$numChannels+i,j*$numChannels*$imgSize,$numChannels))
+ }
+ }
+ colBuild = cbind(colBuild, rowBuild)
+ }
+}
+
+# Set input variables
+x = t(colBuild)
+w=matrix(1,rows=$numFilters, cols=$numChannels*$filterSize)
+dout = conv1d::forward(x, w, $pad, $stride, $numImg, $numChannels, $imgSize,
$numFilters, $filterSize)
+
+output = conv1d::backward_filter(x, dout, $pad, $stride, $numImg,
$numChannels, $imgSize, $numFilters, $filterSize)
+#Write output
+write(output, $output)
diff --git a/src/test/scripts/functions/tensor/Conv1DTest.dml
b/src/test/scripts/functions/tensor/Conv1DTest.dml
new file mode 100644
index 0000000..5a2f3a7
--- /dev/null
+++ b/src/test/scripts/functions/tensor/Conv1DTest.dml
@@ -0,0 +1,53 @@
+#-------------------------------------------------------------
+#
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements. See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership. The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License. You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied. See the License for the
+# specific language governing permissions and limitations
+# under the License.
+#
+#-------------------------------------------------------------
+
+source("scripts/nn/layers/conv1d.dml") as conv1d
+
+# Build input
+rowBuild = seq(1,$numChannels*$imgSize,$numChannels)
+if ( $numChannels > 1 & $imgSize > 1 ){
+ for ( i in 2:$numChannels ){
+ rowBuild = rbind(rowBuild, seq(i,$numChannels*$imgSize,$numChannels))
+ }
+}
+
+colBuild = rowBuild
+if ( $numImg > 1 ){
+ for ( j in 2:$numImg ){
+ rowBuild =
seq((j-1)*$imgSize*$numChannels+1,j*$imgSize*$numChannels,$numChannels)
+ if ( $numChannels > 1 ){
+ for ( i in 2:$numChannels ){
+ rowBuild = rbind(rowBuild,
seq((j-1)*$imgSize*$numChannels+i,j*$numChannels*$imgSize,$numChannels))
+ }
+ }
+ colBuild = cbind(colBuild, rowBuild)
+ }
+}
+
+# Set input variables
+x = t(colBuild)
+w=matrix(1,rows=$numFilters, cols=$numChannels*$filterSize)
+
+# Call function
+output = conv1d::forward(x, w, $pad, $stride, $numImg, $numChannels, $imgSize,
$numFilters, $filterSize)
+
+#Write output
+write(output, $output)
