System.InvalidOperationException:异步操作尚未完成

Question

在Ml.net中拟合导致异常的管道。Fit不可用，我很困惑为什么会发生这种情况。任何帮助都是可观的

var model = pipeline.Fit(dataView);

堆栈跟踪

System.Threading.Channels.AsyncOperation.SthrowIncompleteOperationException((在System.Threading.Channels.SyncOperation1.GetResult(Int16 token) at Microsoft.ML.Transforms.RowShufflingTransformer.Cursor.MoveNextCore() at Microsoft.ML.Data.RootCursorBase.MoveNext() at Microsoft.ML.Trainers.TrainingCursorBase.MoveNext() at Microsoft.ML.Trainers.SdcaTrainerBase3.TrainCore(IChannel ch，RoleMappedData数据，LinearModelParameters预测器，Int32weightSetCount(Microsoft.ML.Trainers.StochasticTrainerBase2.TrainModelCore(TrainContext context) at Microsoft.ML.Trainers.TrainerEstimatorBase2.TrainTransformer(IDataViewtrainSet、IDataView验证集、IPredctor initPredictor(位于的Microsoft.ML.Trainers.TrainerEstimatorBase2.Fit(IDataView input) at Microsoft.ML.Data.EstimatorChain1.Fit(IDataView输入(中的ML.DetectFakeJobPosts.Analyzer.Train((D： \Sources\code everything\CodeItHere\ML.DetectFakeJobPosts\Program.cs:line75，位于中的ML.DetectFakeJobPosts.Program.Main(String[]args(D： \Sources\code everything\CodeItHere\ML.DetectFakeJobPosts\Program.cs:line13

管道

var pipeline = _context.Transforms.Categorical.OneHotEncoding("ec_title", "title")
.Append(_context.Transforms.Categorical.OneHotEncoding("ec_location", "location"))
.Append(_context.Transforms.Categorical.OneHotEncoding("ec_department", "department"))
.Append(_context.Transforms.Categorical.OneHotEncoding("ec_salary_range", "salary_range"))
.Append(_context.Transforms.Text.FeaturizeText("ec_company_profile", "company_profile"))
.Append(_context.Transforms.Text.FeaturizeText("ec_description", "description"))
.Append(_context.Transforms.Text.FeaturizeText("ec_requirements", "requirements"))
.Append(_context.Transforms.Text.FeaturizeText("ec_benefits", "benefits"))
.Append(_context.Transforms.Categorical.OneHotEncoding("ec_employment_type", "employment_type"))
.Append(_context.Transforms.Categorical.OneHotEncoding("ec_required_experience", "required_experience"))
.Append(_context.Transforms.Categorical.OneHotEncoding("ec_required_education", "required_education"))
.Append(_context.Transforms.Categorical.OneHotEncoding("ec_industry", "industry"))
.Append(_context.Transforms.Categorical.OneHotEncoding("ec_function", "function"))
//drop unnecessary columns from view
.Append(_context.Transforms.DropColumns("title", "location", "department", "salary_range", "company_profile", "description", "requirements", "benefits", "employment_type", "required_experience", "required_education", "industry", "function", "telecommuting", "has_company_logo", "has_questions"))
//concate features
.Append(_context.Transforms.Concatenate("Features", "ec_title", "ec_location", "ec_department", "ec_salary_range", "ec_company_profile", "ec_description", "ec_requirements", "ec_benefits", "ec_employment_type", "ec_required_experience", "ec_required_education", "ec_industry", "ec_function"))
//set label/prediction column
.Append(_context.Transforms.Conversion.ConvertType("Label", "fraudulent", DataKind.Boolean))
//select a trainer
.Append(_context.BinaryClassification.Trainers.SdcaLogisticRegression());

单击此处查看数据集

版本

Ml.net：1.5.1

核心：3.1

Answer 1

这个问题来自我为更新部分代码以使用Channels而进行的PR。事实证明，在我们检查导致错误的异步调用结果之前，并不总是保证完成异步调用。

目前在每日NuGet提要中对此进行了修复，以获得最新的每日构建，直到1.5.2版本正式发布。