org.nd4j.linalg.api.ops.impl.transforms.clip.ClipByNorm Maven / Gradle / Ivy
/*******************************************************************************
* Copyright (c) 2015-2018 Skymind, Inc.
*
* This program and the accompanying materials are made available under the
* terms of the Apache License, Version 2.0 which is available at
* https://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.
*
* SPDX-License-Identifier: Apache-2.0
******************************************************************************/
package org.nd4j.linalg.api.ops.impl.transforms.clip;
import onnx.OnnxProto3;
import org.nd4j.autodiff.samediff.SDVariable;
import org.nd4j.autodiff.samediff.SameDiff;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.ops.DynamicCustomOp;
import org.nd4j.linalg.api.shape.Shape;
import org.tensorflow.framework.AttrValue;
import org.tensorflow.framework.GraphDef;
import org.tensorflow.framework.NodeDef;
import java.util.Arrays;
import java.util.Collections;
import java.util.List;
import java.util.Map;
public class ClipByNorm extends DynamicCustomOp {
private double clipValue;
public ClipByNorm() {
}
public ClipByNorm(SameDiff sameDiff, SDVariable x, double clipValue, int... dimensions) {
super(null, sameDiff, new SDVariable[]{x});
this.clipValue = clipValue;
this.dimensions = dimensions;
addIArgument(dimensions);
addTArgument(clipValue);
}
public ClipByNorm(INDArray in, INDArray out, double clipValue, int... dimensions){
super(null, new INDArray[]{in}, (out == null ? null : new INDArray[]{out}), Collections.singletonList(clipValue), dimensions);
}
@Override
public String opName() {
return "clipbynorm";
}
@Override
public void initFromTensorFlow(NodeDef nodeDef, SameDiff initWith, Map attributesForNode, GraphDef graph) {
throw new UnsupportedOperationException("Not yet implemented");
}
@Override
public void initFromOnnx(OnnxProto3.NodeProto node, SameDiff initWith, Map attributesForNode, OnnxProto3.GraphProto graph) {
throw new UnsupportedOperationException("Not yet implemented");
}
@Override
public List doDiff(List grad) {
//dOut/dIn is ??? if clipped, 1 otherwise
/*
int origRank = Shape.rankFromShape(arg().getShape());
SDVariable l2norm = f().norm2(arg(), true, dimensions);
SDVariable isClippedBC = f().gte(l2norm, clipValue);
SDVariable notClippedBC = isClippedBC.rsub(1.0);
// SDVariable dnormdx = arg().div(broadcastableNorm);
// SDVariable sqNorm = f().square(broadcastableNorm);
// SDVariable dOutdInClipped = sqNorm.rdiv(-1).mul(dnormdx).mul(arg()) //-1/(norm2(x))^2 * x/norm2(x)
// .add(broadcastableNorm.rdiv(1.0))
// .mul(clipValue);
SDVariable dOutdInClipped = f().neg(f().square(arg()).div(f().cube(l2norm))) //-x^2/(norm2(x))^3
.add(l2norm.rdiv(1.0)) //+ 1/norm(x)
.mul(clipValue).mul(isClippedBC);
SDVariable ret = notClippedBC.add(dOutdInClipped).mul(grad.get(0));
return Arrays.asList(ret);
*/
return Collections.singletonList(new ClipByNormBp(f().sameDiff(), arg(), grad.get(0), clipValue, dimensions).outputVariable());
}
}