Repository navigation
Add release notes for ML.NET 0.4 - #656
Conversation
| Today we are releasing ML.NET 0.4. During this release we have started | ||
| exploring new APIs for ML.NET that enable functionality that is missing from | ||
| the current APIs. We welcome feedback and contributions to the | ||
| [discussion](https://github.com/dotnet/machinelearning/projects/4). While the |
There was a problem hiding this comment.
discussion [](start = 1, length = 10)
Is it discussion or project? the link opens as a project view. #Closed
There was a problem hiding this comment.
The project is a collection of the relevant issues so it might be a good pointer. Should we change to something else? #Closed
There was a problem hiding this comment.
I'll update to make it clearer. #Closed
| SymSGD can still be helpful in scenarios where you want to try many | ||
| different learners and limit each of them to a single thread. | ||
| * Documentation can be found | ||
| [here](https://docs.microsoft.com/en-us/dotnet/api/microsoft.ml.trainers.symsgdbinaryclassifier?view=ml-dotnet) |
There was a problem hiding this comment.
https://docs.microsoft.com/en-us/dotnet/api/microsoft.ml.trainers.symsgdbinaryclassifier?view=ml-dotnet [](start = 13, length = 103)
currently not available, perhaps needs to be released together? #Closed
There was a problem hiding this comment.
Yep, mentioned in the PR description that the docs.microsoft.com links will be available after the official release.
| [fastText](https://en.wikipedia.org/wiki/FastText), and | ||
| [SSWE](http://anthology.aclweb.org/P/P14/P14-1146.pdf). | ||
| * Documentation can be found | ||
| [here](https://docs.microsoft.com/en-us/dotnet/api/microsoft.ml.transforms.wordembeddings?view=ml-dotnet). |
There was a problem hiding this comment.
(https://docs.microsoft.com/en-us/dotnet/api/microsoft.ml.transforms.wordembeddings?view=ml-dotnet [](start = 12, length = 98)
also not currently available
There was a problem hiding this comment.
This link will be available after the official release.
| * Added Word Embeddings Transform for text scenarios | ||
| ([#545](https://github.com/dotnet/machinelearning/pull/545)) | ||
|
|
||
| * [Word embeddings](https://en.wikipedia.org/wiki/Word_embedding) is a |
There was a problem hiding this comment.
I'd say "words or phrases".
- Word embeddings is a
technique for mapping words or phases to numeric vectors of relatively low
dimension (in comparison with the high dimensional n-gram extraction).
These numeric vectors are intended to capture some of the meaning of the
words so they can be used for training a better model. As an example,
SSWE (Sentiment-Specific Word Embedding) can be useful for sentiment
related tasks. #resolved
There was a problem hiding this comment.
Thanks. Addressed in next commit.
|
Should we call out that symSGD requires a seperate NuGet Package? |
| words so they can be used for training a better model. As an example, | ||
| SSWE (Sentiment-Specific Word Embedding) can be useful for sentiment | ||
| related tasks. | ||
| * This transform enables using pretrained models to get the embeddings |
There was a problem hiding this comment.
We can note the word embeddings are automatically downloaded on first use. (So users don't worry about downloading them).
We may also want to state:
"The transform supports custom pre-trained models in common text formats (Word2vec, fastText, GloVe, etc), allowing users to train word embedding models on their data and use in ML.NET."
There was a problem hiding this comment.
Yes I'll add this note. Regarding custom pretrained models, let's make sure we cover that in the documentation instead as it is a more advanced scenario.
There was a problem hiding this comment.
Good point. We should write-up a "how to train" a custom model section in the docs.
| From package manager: | ||
| ``` | ||
| Install-Package Microsoft.ML | ||
| ``` |
There was a problem hiding this comment.
Should we link to how to build your own copy?
There was a problem hiding this comment.
I don't think this is necessary for the release notes (it's more about getting started with using the package as opposed to building it).
|
Do we want to call out the F# and property support that went in? #616 ? |
|
I was thinking about it but wondering if we wanted to add some more docs / tests before including it in the release notes. What do you think? |
|
We have F# tests in the repo: https://github.com/dotnet/machinelearning/tree/master/test/Microsoft.ML.FSharp.Tests Also, @dsyme has a PR for the samples repo to add F# samples: dotnet/machinelearning-samples#36 We can hold off if you want, but it seems like mentioning that we are adding supporting for F# record types in 0.4 would be valuable. |
|
Sounds good. I've added "improved F# support" to the notes. |
ghost
locked as resolved and limited conversation to collaborators
This adds release notes for ML.NET 0.4.
Note that some of the documentation links are not available yet. They will start working after the official release.