Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[SYSTEMML-540] [SYSTEMML-515] Allow an expression for sparsity #351

Closed
wants to merge 2 commits into from
Closed

[SYSTEMML-540] [SYSTEMML-515] Allow an expression for sparsity #351

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7d724b8
[SYSTEMML-540] [SYSTEMML-515] Allow an expression for sparsity
Jan 19, 2017
b1b0b66
Incorporated Matthias and Mike's comments
Jan 19, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
10 changes: 8 additions & 2 deletions scripts/staging/SystemML-NN/nn/layers/dropout.dml
View file
Open in desktop
Expand Up @@ -42,10 +42,16 @@ forward = function(matrix[double] X, double p, int seed)
* - out: Ouptuts, of same shape as X.
* - mask: Dropout mask used to compute the output.
*/
# Normally, we might use something like
# `mask = rand(rows=nrow(X), cols=ncol(X), min=0, max=1, seed=seed) <= p`
# to create a dropout mask. Fortunately, SystemML has a `sparsity` parameter on
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

These three lines need to be added back so that a user can indicate that the dropout layer should use a random seed.

# the `rand` function that allows use to create a mask directly.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we add a comment here explaining how we are creating the mask with the sparsity flag, since that is not a common construct? I.e. add the following comment:

# Normally, we might use something like
#    `mask = rand(rows=nrow(X), cols=ncol(X), min=0, max=1, seed=seed) <= p`
# to create a dropout mask.  Fortunately, SystemML has a `sparsity` parameter on
# the `rand` function that allows use to create a mask directly.
mask = rand(rows=nrow(X), cols=ncol(X), min=1, max=1, sparsity=p, seed=seed)

if (seed == -1) {
seed = as.integer(floor(as.scalar(rand(rows=1, cols=1, min=1, max=100000))))
mask = rand(rows=nrow(X), cols=ncol(X), min=1, max=1, sparsity=p)
}
else {
mask = rand(rows=nrow(X), cols=ncol(X), min=1, max=1, sparsity=p, seed=seed)
}
mask = rand(rows=nrow(X), cols=ncol(X), min=0, max=1, seed=seed) <= p
out = X * mask / p
}

Expand Down
12 changes: 8 additions & 4 deletions src/main/java/org/apache/sysml/hops/DataGenOp.java
View file
Open in desktop
Expand Up @@ -100,8 +100,10 @@ public DataGenOp(DataGenMethod mthd, DataIdentifier id, HashMap<String, Hop> inp
_paramIndexMap.put(s, index);
index++;
}
if ( mthd == DataGenMethod.RAND )
_sparsity = Double.valueOf(((LiteralOp)inputParameters.get(DataExpression.RAND_SPARSITY)).getName());

Hop sparsityOp = inputParameters.get(DataExpression.RAND_SPARSITY);
if ( mthd == DataGenMethod.RAND && sparsityOp instanceof LiteralOp)
_sparsity = Double.valueOf(((LiteralOp)sparsityOp).getName());

//generate base dir
String scratch = ConfigurationManager.getScratchSpace();
Expand Down Expand Up @@ -199,7 +201,7 @@ protected double computeOutputMemEstimate( long dim1, long dim2, long nnz )
{
double ret = 0;

if ( _op == DataGenMethod.RAND ) {
if ( _op == DataGenMethod.RAND && _sparsity != -1 ) {
if( hasConstantValue(0.0) ) { //if empty block
ret = OptimizerUtils.estimateSizeEmptyBlock(dim1, dim2);
}
Expand Down Expand Up @@ -237,7 +239,7 @@ protected long[] inferOutputCharacteristics( MemoTable memo )
{
long dim1 = computeDimParameterInformation(getInput().get(_paramIndexMap.get(DataExpression.RAND_ROWS)), memo);
long dim2 = computeDimParameterInformation(getInput().get(_paramIndexMap.get(DataExpression.RAND_COLS)), memo);
long nnz = (long)(_sparsity * dim1 * dim2);
long nnz = _sparsity >= 0 ? (long)(_sparsity * dim1 * dim2) : -1;
if( dim1>0 && dim2>0 )
return new long[]{ dim1, dim2, nnz };
}
Expand Down Expand Up @@ -355,6 +357,8 @@ else if (_op == DataGenMethod.SEQ )
_nnz = 0;
else if ( dimsKnown() && _sparsity>=0 ) //general case
_nnz = (long) (_sparsity * _dim1 * _dim2);
else
_nnz = -1;
}


Expand Down
15 changes: 7 additions & 8 deletions src/main/java/org/apache/sysml/lops/DataGen.java
View file
Open in desktop
Expand Up @@ -199,12 +199,11 @@ private String getCPInstruction_Rand(String output)
sb.append(iLop.prepScalarLabel());
sb.append(OPERAND_DELIMITOR);

iLop = _inputParams.get(DataExpression.RAND_SPARSITY.toString()); //no variable support
iLop = _inputParams.get(DataExpression.RAND_SPARSITY.toString());
if (iLop.isVariable())
throw new LopsException(printErrorLocation()
+ "Parameter " + DataExpression.RAND_SPARSITY
+ " must be a literal for a Rand operation.");
sb.append(iLop.getOutputParameters().getLabel());
sb.append(iLop.prepScalarLabel());
else
sb.append(iLop.getOutputParameters().getLabel());
sb.append(OPERAND_DELIMITOR);

iLop = _inputParams.get(DataExpression.RAND_SEED.toString());
Expand Down Expand Up @@ -442,9 +441,9 @@ private String getMRInstruction_Rand(int inputIndex, int outputIndex)

iLop = _inputParams.get(DataExpression.RAND_SPARSITY.toString()); //no variable support
if (iLop.isVariable())
throw new LopsException(this.printErrorLocation() + "Parameter "
+ DataExpression.RAND_SPARSITY + " must be a literal for a Rand operation.");
sb.append( iLop.getOutputParameters().getLabel() );
sb.append(iLop.prepScalarLabel());
else
sb.append( iLop.getOutputParameters().getLabel() );
sb.append( OPERAND_DELIMITOR );

iLop = _inputParams.get(DataExpression.RAND_SEED.toString());
Expand Down
5 changes: 1 addition & 4 deletions src/main/java/org/apache/sysml/parser/DataExpression.java
View file
Open in desktop
Expand Up @@ -1162,10 +1162,7 @@ else if (dataParam instanceof StringIdentifier) {
raiseValidateError("for Rand statement " + RAND_MIN + " has incorrect value type", conditional);
}

//parameters w/o support for variable inputs (requires double/int or string constants)
if (!(getVarParam(RAND_SPARSITY) instanceof DoubleIdentifier || getVarParam(RAND_SPARSITY) instanceof IntIdentifier)) {
raiseValidateError("for Rand statement " + RAND_SPARSITY + " has incorrect value type", conditional);
}
// Since sparsity can be arbitrary expression (SYSTEMML-515), no validation check for DoubleIdentifier/IntIdentifier required.

if (!(getVarParam(RAND_PDF) instanceof StringIdentifier)) {
raiseValidateError("for Rand statement " + RAND_PDF + " has incorrect value type", conditional);
Expand Down
5 changes: 4 additions & 1 deletion src/main/java/org/apache/sysml/runtime/instructions/cp/DataGenCPInstruction.java
View file
Open in desktop
Expand Up @@ -212,7 +212,10 @@ else if ( opcode.equalsIgnoreCase(DataGen.SAMPLE_OPCODE) ) {
maxValue = Double.valueOf(s[6]).doubleValue();
}

double sparsity = Double.parseDouble(s[7]);
double sparsity = -1;
if (!s[7].contains( Lop.VARIABLE_NAME_PLACEHOLDER)) {
sparsity = Double.valueOf(s[7]);
}

long seed = DataGenOp.UNSPECIFIED_SEED;
if( !s[8].contains( Lop.VARIABLE_NAME_PLACEHOLDER)){
Expand Down
5 changes: 4 additions & 1 deletion src/main/java/org/apache/sysml/runtime/instructions/spark/RandSPInstruction.java
View file
Open in desktop
Expand Up @@ -245,7 +245,10 @@ else if ( opcode.equalsIgnoreCase(DataGen.SAMPLE_OPCODE) ) {
maxValue = Double.valueOf(s[6]).doubleValue();
}

double sparsity = Double.parseDouble(s[7]);
double sparsity = -1;
if (!s[7].contains( Lop.VARIABLE_NAME_PLACEHOLDER)) {
sparsity = Double.valueOf(s[7]);
}

long seed = DataGenOp.UNSPECIFIED_SEED;
if (!s[8].contains( Lop.VARIABLE_NAME_PLACEHOLDER)) {
Expand Down