Introduce examples for pipeline api. #677
Conversation
| var testDataPath = GetDataPath(SentimentDataPath); | ||
| var pipeline = new LearningPipeline(); | ||
|
|
||
| pipeline.Add(MakeSentimentTextLoader(dataPath)); |
There was a problem hiding this comment.
MakeSentimentTextLoader [](start = 25, length = 23)
can you not use TextLoader somehow? #Closed
There was a problem hiding this comment.
| void CrossValidation() | ||
| { | ||
| var dataPath = GetDataPath(SentimentDataPath); | ||
| var testDataPath = GetDataPath(SentimentDataPath); |
There was a problem hiding this comment.
testDataPath [](start = 16, length = 12)
remove line #Closed
| var testDataPath = GetDataPath(SentimentDataPath); | ||
| var pipeline = new LearningPipeline(); | ||
|
|
||
| pipeline.Add(MakeSentimentTextLoader(dataPath)); |
There was a problem hiding this comment.
Add [](start = 21, length = 3)
TextLoader ? #Closed
| pipeline.Add(new ColumnConcatenator(outputColumn: "Features", | ||
| "SepalLength", "SepalWidth", "PetalLength", "PetalWidth")); | ||
|
|
||
| pipeline.Add(OneVersusAll.With(new StochasticDualCoordinateAscentBinaryClassifier())); |
There was a problem hiding this comment.
With [](start = 38, length = 4)
does it accept multiclass? #Closed
There was a problem hiding this comment.
write a comment that this is checked at Train time, so technically this example is not possible with LP
In reply to: 210040514 [](ancestors = 210040514)
| /// <summary> | ||
| /// This is a trivial implementation of a thread-safe prediction engine is just guarded by a lock. | ||
| /// </summary> | ||
| private sealed class LockBasedPredictionEngine |
There was a problem hiding this comment.
class [](start = 23, length = 5)
remove custom classes #Closed
| var model = pipeline.Train<SentimentData, SentimentPrediction>(); | ||
| var modelName = "multithreadModel.zip"; | ||
| DeleteOutputPath(modelName); | ||
| model.WriteAsync(modelName); |
There was a problem hiding this comment.
model [](start = 12, length = 5)
await? #Closed
|
|
||
| pipeline.Add(new PredictedLabelColumnOriginalValueConverter() { PredictedLabelColumn = "PredictedLabel" }); | ||
| var model = pipeline.Train<SentimentData, SentimentPrediction>(); | ||
| var testData = MakeSentimentTextLoader(testDataPath); |
There was a problem hiding this comment.
MakeSentimentTextLoader [](start = 27, length = 23)
yuck #Closed
There was a problem hiding this comment.
rename testData to testHandle?.. testSomething. It's not data
In reply to: 210042025 [](ancestors = 210042025)
| var loadedModel = await PredictionModel.ReadAsync<SentimentData, SentimentPrediction>(modelName); | ||
| var testData = MakeSentimentTextLoader(testDataPath); | ||
| var evaluator = new BinaryClassificationEvaluator(); | ||
| var metrics = evaluator.Evaluate(loadedModel, testData); |
There was a problem hiding this comment.
remove from this test #Closed
There was a problem hiding this comment.
| var testData = MakeSentimentTextLoader(testDataPath); | ||
| var evaluator = new BinaryClassificationEvaluator(); | ||
| var metrics = evaluator.Evaluate(loadedModel, testData); | ||
| var singlePrediction = loadedModel.Predict(new SentimentData() { SentimentText = "Not big fan of this." }); |
There was a problem hiding this comment.
SentimentData [](start = 59, length = 13)
try to do the same with Iris, and check if loading a model with IrisDataNoLabel works, but Predict fails. If that's the case, then there's no decomposable piope. #ByDesign
There was a problem hiding this comment.
| var metrics = evaluator.Evaluate(model, testLearningPipelineItem); | ||
|
|
||
| var singlePrediction = model.Predict(new SentimentData() { SentimentText = "Not big fan of this." }); | ||
| Assert.True(singlePrediction.Sentiment); |
| pipeline.Add(new ColumnConcatenator(outputColumn: "Features", | ||
| "SepalLength", "SepalWidth", "PetalLength", "PetalWidth")); | ||
|
|
||
| // this will throw exception during training time if you specify any other than binary classifier. |
There was a problem hiding this comment.
t [](start = 15, length = 1)
T #Resolved
|
@codemzs @eerhardt I'm trying to introduce examples for #584, and right now I have Can you confirm or tell me I'm wrong for some cases. |
|
I can only answer a few. Maybe @codemzs can answer the others.
I don't see any Entry Point for these scenarios, let alone a LearningPipeline API...
Isn't this |
In reply to: 413201180 [](ancestors = 413201180) |
|
|
||
| Parallel.ForEach(collection, (input) => | ||
| { | ||
| lock (model) |
There was a problem hiding this comment.
I would put a comment here why we would need to lock.
examples for #584