Skip to content

Fixes OneHotEncoding Issue #4974

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

Merged
ganik merged 8 commits into from
Mar 26, 2020
Merged

Fixes OneHotEncoding Issue #4974

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
65557b4
fix for logistic regression
kere-nel Mar 23, 2020
e9b2a42
rebase changes
kere-nel Mar 24, 2020
6e2bf6a
resolving comments
kere-nel Mar 24, 2020
30258b6
Merge remote-tracking branch 'upstream/master'
kere-nel Mar 25, 2020
bad6d80
onehotencoding dimension fix
kere-nel Mar 25, 2020
24ee576
adding shapes to nodes
kere-nel Mar 25, 2020
c434392
updating baseline
kere-nel Mar 25, 2020
d3e063d
updating baseline
kere-nel Mar 26, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
25 changes: 9 additions & 16 deletions src/Microsoft.ML.Data/Transforms/KeyToVector.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -694,29 +694,22 @@ private void SaveAsOnnxCore(OnnxContext ctx, int iinfo, ColInfo info, string src
// because inputs and outputs of a transform are declared with shapes.
Contracts.CheckValue(shape, nameof(shape));

// If Bag is true, the output of ONNX LabelEncoder needs to be fed into ONNX ReduceSum because
// default ONNX LabelEncoder just matches the behavior of Bag=false.
var encodedVariableName = _parent._columns[iinfo].OutputCountVector ? ctx.AddIntermediateVariable(null, "encoded", true) : dstVariableName;

string opType = "Cast";
var castOutput = ctx.AddIntermediateVariable(info.TypeSrc, opType, true);
var castOutput = ctx.AddIntermediateVariable(NumberDataViewType.Int64, opType);
var castNode = ctx.CreateNode(opType, srcVariableName, castOutput, ctx.GetNodeName(opType), "");
castNode.AddAttribute("to", typeof(long));

opType = "OneHotEncoder";
var categoryRange = info.TypeSrc.GetItemType().GetKeyCountAsInt32(Host);
var encodedVariableName = ctx.AddIntermediateVariable(new VectorDataViewType(NumberDataViewType.Single, 1, categoryRange), "encoded");
var node = ctx.CreateNode(opType, castOutput, encodedVariableName, ctx.GetNodeName(opType));
node.AddAttribute("cats_int64s", Enumerable.Range(1, info.TypeSrc.GetItemType().GetKeyCountAsInt32(Host)).Select(x => (long)x));
node.AddAttribute("cats_int64s", Enumerable.Range(1, categoryRange).Select(x => (long)x));
node.AddAttribute("zeros", true);
if (_parent._columns[iinfo].OutputCountVector)
{
// If input shape is [1, 3], then OneHotEncoder may produce a 3-D tensor. Thus, we need to do a
// reduction along the second last axis to merge the one-hot vectors produced by all input features.
// Note that one input feature got expended to an one-hot vector.
opType = "ReduceSum";
var reduceNode = ctx.CreateNode(opType, encodedVariableName, dstVariableName, ctx.GetNodeName(opType), "");
reduceNode.AddAttribute("axes", new long[] { shape.Count - 1 });
reduceNode.AddAttribute("keepdims", 0);
}

// OneHotEncoder adds one additional dimension, so we remove it below
opType = "Squeeze";
var reduceNode = ctx.CreateNode(opType, encodedVariableName, dstVariableName, ctx.GetNodeName(opType), "");
reduceNode.AddAttribute("axes", new long[] { shape.Count - 1 });
}
}
}
Expand Down
47 changes: 41 additions & 6 deletions ...Output/Common/Onnx/BinaryClassification/BreastCancer/ExcludeVariablesInOnnxConversion.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -130,19 +130,15 @@
"output": [
"F21"
],
"name": "ReduceSum",
"opType": "ReduceSum",
"name": "Squeeze",
"opType": "Squeeze",
"attribute": [
{
"name": "axes",
"ints": [
"1"
],
"type": "INTS"
},
{
"name": "keepdims",
"type": "INT"
}
]
},
Expand Down Expand Up @@ -659,6 +655,45 @@
}
}
},
{
"name": "Cast",
"type": {
"tensorType": {
"elemType": 7,
"shape": {
"dim": [
{
"dimValue": "-1"
},
{
"dimValue": "1"
}
]
}
}
}
},
{
"name": "encoded",
"type": {
"tensorType": {
"elemType": 1,
"shape": {
"dim": [
{
"dimValue": "-1"
},
{
"dimValue": "1"
},
{
"dimValue": "10"
}
]
}
}
}
},
{
"name": "F22",
"type": {
Expand Down
60 changes: 59 additions & 1 deletion test/BaselineOutput/Common/Onnx/BinaryClassification/BreastCancer/ModelWithLessIO.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,7 +92,7 @@
"Cast"
],
"output": [
"F21"
"encoded"
],
"name": "OneHotEncoder",
"opType": "OneHotEncoder",
Expand Down Expand Up @@ -120,6 +120,25 @@
],
"domain": "ai.onnx.ml"
},
{
"input": [
"encoded"
],
"output": [
"F21"
],
"name": "Squeeze",
"opType": "Squeeze",
"attribute": [
{
"name": "axes",
"ints": [
"1"
],
"type": "INTS"
}
]
},
{
"input": [
"F1",
Expand Down Expand Up @@ -985,6 +1004,45 @@
}
}
},
{
"name": "Cast",
"type": {
"tensorType": {
"elemType": 7,
"shape": {
"dim": [
{
"dimValue": "-1"
},
{
"dimValue": "1"
}
]
}
}
}
},
{
"name": "encoded",
"type": {
"tensorType": {
"elemType": 1,
"shape": {
"dim": [
{
"dimValue": "-1"
},
{
"dimValue": "1"
},
{
"dimValue": "9"
}
]
}
}
}
},
{
"name": "VectorFeaturizerOutput",
"type": {
Expand Down
Loading