Add release notes for ML.NET 0.6 #1102
Conversation
|
For type system you can add issue #673 that contains the spec and benchmarks. |
|
Should we specifically call out the changes to which data-types are allowed/disallowed to have missing values? I think it's uncommon for users to have NA values in data types other than R4 as most pipelines are mostly R4 for Features & Key/R4/BL for Label. #RESOLVED |
| * New APIs for ML.NET | ||
|
|
||
| * While the `LearningPipeline` APIs that were released with ML.NET 0.1 were easy to get started with, they had obvious limitations in functionality. Certain tasks that were possible with the internal version of ML.NET like inspecting model weights, creating a transform-only pipeline, and training from an initial predictor could not be done with `LearningPipeline`. | ||
| * The important concepts for understanding the new API are introduced [here](https://github.com/dotnet/machinelearning/blob/3cdd3c8b32705e91dcf46c429ee34196163af6da/docs/code/MlNetHighLevelConcepts.md). |
There was a problem hiding this comment.
3cdd3c8 [](start = 128, length = 40)
now that it's merged, you should use https://github.com/dotnet/machinelearning/blob/master/docs/code/MlNetHighLevelConcepts.md etc.
There was a problem hiding this comment.
This is still a link to master but it is a permalink (in case the content of the file changes or it gets moved).
| * Replaced ML.NET's Dv type system with .NET's standard type system ([#863](https://github.com/dotnet/machinelearning/pull/863)) | ||
|
|
||
| * ML.NET previously had its own type system which helped it more efficiently deal with things like missing values (a common case in ML). This type system required users to work with types like `DvText`, `DvBool`, `DvInt4`, etc. | ||
| * This update replaces the Dv type system with .NET's standard type system to make ML.NET easier to use and to take advantage of innovation in .NET. |
There was a problem hiding this comment.
easier to use [](start = 94, length = 13)
probably mention the fact that now only floats and doubles have missing values, represented with NaN?
There was a problem hiding this comment.
@GalOshri can you also add pointer to sample for ONNX scoring and if we have any data re: the perf speedup for prediction can we share? I feel uncomfortable explaining a big perf speedup with no backing evidence :(
|
|
||
| * Up to ~200x speedup in prediction engine performance ([#973](https://github.com/dotnet/machinelearning/pull/973)) | ||
|
|
||
| * This improvement leads to a significant speedup when making predictions for single records. |
There was a problem hiding this comment.
This improvement leads to a significant speedup [](start = 6, length = 47)
LOL. First rule of tautology club is the first rule of tautology club.
| * The important concepts for understanding the new API are introduced [here](https://github.com/dotnet/machinelearning/blob/3cdd3c8b32705e91dcf46c429ee34196163af6da/docs/code/MlNetHighLevelConcepts.md). | ||
| * A cookbook that shows how to use these APIs for a variety of existing and new scenarios can be found [here](https://github.com/dotnet/machinelearning/blob/3cdd3c8b32705e91dcf46c429ee34196163af6da/docs/code/MlNetCookBook.md). | ||
| * These APIs are still evolving, so we would love to hear any feedback or questions. | ||
| * The `LearningPipeline` APIs have moved to the `Microsoft.ML.Legacy` namespace. |
There was a problem hiding this comment.
I would mention that we plan on removing the LearningPipeline APIs in the next 0.7 release.
ghost
locked as resolved and limited conversation to collaborators
7 participants
This adds release notes for ML.NET 0.6.