Skip to content

Fixes multiclass logistic regression #4963

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
merged 3 commits into from
Mar 24, 2020
Merged

Fixes multiclass logistic regression #4963

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
65557b4
fix for logistic regression
Lynx1820 Mar 23, 2020
e9b2a42
rebase changes
Lynx1820 Mar 24, 2020
6e2bf6a
resolving comments
Lynx1820 Mar 24, 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
32 changes: 10 additions & 22 deletions ...Microsoft.ML.StandardTrainers/Standard/LogisticRegression/MulticlassLogisticRegression.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -978,41 +978,29 @@ private JToken SaveAsPfaCore(BoundPfaContext ctx, JToken input)
private bool SaveAsOnnxCore(OnnxContext ctx, string[] outputs, string featureColumn)
{
Host.CheckValue(ctx, nameof(ctx));
Host.Assert(outputs[0] == DefaultColumnNames.PredictedLabel);
Host.Assert(outputs[1] == DefaultColumnNames.Score);

string predictedLabelInt64 = null;
string predictedLabelUint32 = null;
// REVIEW: What is the right way to get the name of the predicted column?
for (int i = 0; i < outputs.Length; i++)
{
if (outputs[i] != DefaultColumnNames.PredictedLabel)
continue;
predictedLabelUint32 = DefaultColumnNames.PredictedLabel;
predictedLabelInt64 = ctx.AddIntermediateVariable(NumberDataViewType.Int64, "PredictedLabelInt64", true);
outputs[i] = predictedLabelInt64;
break;
}

Host.CheckValue(predictedLabelInt64, nameof(predictedLabelInt64));
string classifierLabelOutput = ctx.AddIntermediateVariable(NumberDataViewType.Int64, "ClassifierLabelOutput", true);

string opType = "LinearClassifier";
var node = ctx.CreateNode(opType, new[] { featureColumn }, outputs, ctx.GetNodeName(opType));
var node = ctx.CreateNode(opType, new[] { featureColumn }, new[] { classifierLabelOutput, outputs[1] }, ctx.GetNodeName(opType));
node.AddAttribute("post_transform", GetOnnxPostTransform());
node.AddAttribute("multi_class", true);
node.AddAttribute("coefficients", Weights.SelectMany(w => w.DenseValues()));
node.AddAttribute("intercepts", Biases);
node.AddAttribute("classlabels_ints", Enumerable.Range(1, NumberOfClasses).Select(x => (long)x));

opType = "Unsqueeze";
var unsqueezeOutput = ctx.AddIntermediateVariable(NumberDataViewType.Int64, "CastNodeOutput");
var unsqueezeNode = ctx.CreateNode(opType, classifierLabelOutput, unsqueezeOutput, ctx.GetNodeName(opType), "");
unsqueezeNode.AddAttribute("axes", new long[] { 1 });

// Onnx outputs an Int64, but ML.NET outputs UInt32. So cast the Onnx output here
opType = "Cast";
var castNodeOutput = ctx.AddIntermediateVariable(NumberDataViewType.UInt32, "CastNodeOutput", true);
var castNode = ctx.CreateNode(opType, predictedLabelInt64, castNodeOutput, ctx.GetNodeName(opType), "");
var castNode = ctx.CreateNode(opType, unsqueezeOutput, outputs[0], ctx.GetNodeName(opType), "");
var t = InternalDataKindExtensions.ToInternalDataKind(DataKind.UInt32).ToType();
castNode.AddAttribute("to", t);

opType = "Unsqueeze";
var unsqueezeNode = ctx.CreateNode(opType, castNodeOutput, predictedLabelUint32, ctx.GetNodeName(opType), "");
unsqueezeNode.AddAttribute("axes", new long[] { 0 });

return true;
}

Expand Down
223 changes: 142 additions & 81 deletions ...sClassification/BreastCancer/MultiClassificationLogisticRegressionSaveModelToOnnxTest.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,31 @@
"output": [
"Features0"
],
"name": "Imputer",
"opType": "Imputer",
"attribute": [
{
"name": "replaced_value_float",
"f": "NaN",
"type": "FLOAT"
},
{
"name": "imputed_value_floats",
"floats": [
0
],
"type": "FLOATS"
}
],
"domain": "ai.onnx.ml"
},
{
"input": [
"Features0"
],
"output": [
"Features1"
],
"name": "Scaler",
"opType": "Scaler",
"attribute": [
Expand Down Expand Up @@ -115,10 +140,10 @@
},
{
"input": [
"Features0"
"Features1"
],
"output": [
"PredictedLabelInt64",
"ClassifierLabelOutput",
"Score"
],
"name": "LinearClassifier",
Expand All @@ -137,102 +162,102 @@
{
"name": "coefficients",
"floats": [
-0.0871891156,
0.209310874,
0.747134566,
0.364765137,
-0.377612084,
-0.6847462,
-0.03167667,
0.143471345,
0.712303758,
0.352516025,
-0.301202744,
-0.741327,
0,
-0.5566554,
-0.3849638,
-1.29262471,
-0.50539434,
-0.495568782,
-1.18026292,
0,
0,
-0.479907274,
-0.08740093,
-0.5489706,
-0.498504966,
-0.0420997739,
-0.613307,
0,
0.630316138,
0.607902765,
0,
0.141199484,
0,
0,
0.07319626,
0.171390951,
0.6936194,
0.211812049,
0.7400253,
0,
0,
-0.6189027,
-0.6818558,
0,
-0.732489467,
-0.71812433,
0.2614429,
-0.4669126,
-0.250123739,
1.01838875,
0.7936676,
-0.551549,
-0.696967065,
0.160799474,
-0.330957055,
-0.2750144,
1.0280081,
0.88234216,
0,
0,
0.8072781,
0.6424018,
0,
0.833407462,
0.8743826,
0,
-1.67462111,
-1.19559848,
-0.553805768,
-0.5710498,
-0.7325714,
-0.5470721,
-0.7483947,
-1.70732152,
-1.191878,
-0.5418798,
-0.600155354,
-0.740290344,
-0.674283,
-0.7527394,
0,
-0.5655844,
-0.9892823,
-0.237264976,
-0.5712943,
-1.04777157,
-0.226992026,
0,
-0.81984,
-0.0930810943,
-0.4526821,
-0.706178248,
-0.145423546,
-0.465133369,
0,
0,
0,
0.726712048,
0.7297528,
0,
1.12171924,
0.323810369,
0.245762676,
0.07872447,
0.939905643,
0.923160553,
1.16632056,
0.311890721,
0.247274086,
0.102029644,
0.955604,
0.9211974,
0,
0,
1.10209334,
0.704743862,
1.15463662,
0.6863172,
0,
0.322121173,
0.5064917,
1.30212963,
0.35739857,
0.51222384,
1.30172575,
0,
0.8623323,
0.0155395176,
0.8770747,
0.0631457046,
0,
0.192209348,
0.262786478
0.184382111,
0.289009124
],
"type": "FLOATS"
},
{
"name": "intercepts",
"floats": [
1.23585367,
1.68783426,
-0.8096311,
1.35599542,
-1.59806383,
2.57355452,
1.03064489,
-1.67592752,
-1.40655541,
-2.39366078
1.24280047,
1.713405,
-0.833424449,
1.31703234,
-1.520917,
2.636808,
1.016353,
-1.69996321,
-1.4401859,
-2.43189263
],
"type": "FLOATS"
},
Expand All @@ -257,18 +282,20 @@
},
{
"input": [
"PredictedLabelInt64"
"ClassifierLabelOutput"
],
"output": [
"CastNodeOutput"
],
"name": "Cast0",
"opType": "Cast",
"name": "Unsqueeze",
"opType": "Unsqueeze",
"attribute": [
{
"name": "to",
"i": "12",
"type": "INT"
"name": "axes",
"ints": [
"1"
],
"type": "INTS"
}
]
},
Expand All @@ -279,15 +306,13 @@
"output": [
"PredictedLabel"
],
"name": "Unsqueeze",
"opType": "Unsqueeze",
"name": "Cast0",
"opType": "Cast",
"attribute": [
{
"name": "axes",
"ints": [
"0"
],
"type": "INTS"
"name": "to",
"i": "12",
"type": "INT"
}
]
},
Expand All @@ -303,7 +328,7 @@
},
{
"input": [
"Features0"
"Features1"
],
"output": [
"Features.output"
Expand Down Expand Up @@ -540,6 +565,24 @@
}
}
},
{
"name": "Features1",
"type": {
"tensorType": {
"elemType": 1,
"shape": {
"dim": [
{
"dimValue": "-1"
},
{
"dimValue": "8"
}
]
}
}
}
},
{
"name": "Label0",
"type": {
Expand Down Expand Up @@ -612,6 +655,24 @@
}
}
},
{
"name": "CastNodeOutput",
"type": {
"tensorType": {
"elemType": 7,
"shape": {
"dim": [
{
"dimValue": "-1"
},
{
"dimValue": "1"
}
]
}
}
}
},
{
"name": "Label.output",
"type": {
Expand Down
14 changes: 13 additions & 1 deletion test/Microsoft.ML.Tests/OnnxConversionTest.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -736,7 +736,8 @@ public void MulticlassLogisticRegressionOnnxConversionTest()
separatorChar: '\t',
hasHeader: true);

var pipeline = mlContext.Transforms.NormalizeMinMax("Features").
var pipeline = mlContext.Transforms.ReplaceMissingValues("Features").
Append(mlContext.Transforms.NormalizeMinMax("Features")).
Append(mlContext.Transforms.Conversion.MapValueToKey("Label")).
Append(mlContext.MulticlassClassification.Trainers.LbfgsMaximumEntropy(new LbfgsMaximumEntropyMulticlassTrainer.Options() { NumberOfThreads = 1 }));

Expand All @@ -752,6 +753,17 @@ public void MulticlassLogisticRegressionOnnxConversionTest()

SaveOnnxModel(onnxModel, onnxFilePath, onnxTextPath);

// Compare results produced by ML.NET and ONNX's runtime.
if (IsOnnxRuntimeSupported())
{
// Evaluate the saved ONNX model using the data used to train the ML.NET pipeline.
var onnxEstimator = mlContext.Transforms.ApplyOnnxModel(onnxFilePath);
var onnxTransformer = onnxEstimator.Fit(data);
var onnxResult = onnxTransformer.Transform(data);
CompareSelectedColumns<UInt32>("PredictedLabel", "PredictedLabel", transformedData, onnxResult);
CompareSelectedColumns<float>("Score", "Score", transformedData, onnxResult);
}

CheckEquality(subDir, onnxTextName, digitsOfPrecision: 2);
Done();
}
Expand Down