Parameter name: labelCount (Microsoft.ML.) To build a multiclass classification model, the number of classes needs to be 2 or greater
MODEL BUILDER ITERATE ROW SELECTION TRIAL
17:23:01.2188 INFO Trial 0 encounter error with message: Only one class found in the Label column. Since this exception occurs in a Task (I think it's part of an async path?) it doesn't end up appearing in the resultant stack trace displayed by VS, but it does appear in the log:
MODEL BUILDER ITERATE ROW SELECTION CODE
Finally, in InitializeTrainingGraph, the code calls GetKe圜ount() to find out how many keys (labels) were identified, then calls Contracts.CheckParam to ensure that it was greater than 1. This calls into ImageClassificationTrainer.TrainModelCore, which passes to InitializeTrainingGraph. After this occurs, a value to key mapping is performed via Microsoft.ML.Transforms.ValueToKeyMappingTransformer, which produces only one label. In a dataset where the first label has more then 10000 items (rows), this results in all of the initial sample containing just one label. In Microsoft.ML.Propose() there is a call to HyperParameterProposer.Propose(.), which sets an initial sample size of 10000 based on the initializeSampleSize field. When the model builder creates a new model, it proposes a set of initial parameters. I believe I have discovered the underlying bug. I investigated by throwing the DLLs into dotPeek and debugging .圆4.exe while it attempted to perform the training. I initially thought that this was related to #2046, since the exact same error and stacktrace was displayed, however on further investigation the stacktrace in the log is different to the one provided in that issue. Uninstalling/reinstalling the extensions, clearing temp files, killing processes, rebooting, etc. End of stack trace from previous location where exception was thrown -Īt .Throw()Īt .HandleNonSuccessAndDebuggerNotification(Task task)Īt Microsoft.ML._21.MoveNext() in /_/src/Microsoft.ML.ModelBuilder.AutoMLService/AutoMLEngineService/AutoMLEngine.cs:line 160 (Microsoft.ML.) At Microsoft.ML.(IExceptionContext ctx, String s, String paramName)Īt Microsoft.ML.ModelOperationsCatalog.Load(String filePath, DataViewSchema& inputSchema)Īt Microsoft.ML.d_15.MoveNext() in /_/src/Microsoft.ML.ModelBuilder.AutoMLService/Experiments/LocalAutoMLExperiment.cs:line 231
0 kommentar(er)
