Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add ML.NET notes from session 5 #81

Merged
merged 1 commit into from
Jan 23, 2019
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 40 additions & 0 deletions 2018/Microsoft.ML.Core/Session5.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,40 @@
# Microsoft.ML

Status: **Needs more work**

## Notes

* `MLContext`
- Solves two problems: the chaining for `IHostEnvironment` and the
discoverability for finding implementations of common abstraction.
- They are solved by offering extension methods
- We should align the naming of `XxxCatalog` with `XxxContext`

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

- It's a bit odd that the constructor uses nullability to indicate default,
but sentinel values seem more problem. Overloads don't work very well
because the types are the same and it will force people to specify value
of the other. Easiest solution might be to make both nullable. Still not
idiomatic API design, but at least consistent.
* [Typical example](https://dotnet.microsoft.com/learn/machinelearning-ai/ml-dotnet-get-started-tutorial)
- Most of the methods aren't verbs; instead of `TextReader()` it should be
`CreateTextReader()`
- We should have a pattern for the arguments:
- Simple set of arguments should just be parameters

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is captured in existing issues but to reinforce also mentioned it as a subitem in a new issue as well.

- More complicated arguments should be an option class that is mutable

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Entered as this issue, this was a fairly lengthy discussion that went into more things than are captured in this paragraph so it's possible I forgot something in the issue, let me know if so.

- Some types probably want to offer discrete parameters for convenience
while also being able to accept more complex options. The overloads
should never mix them as this would create conflicts which value
"wins".
- Example:
```C#
var options = new TextFileSource.Options();
options.Separator = ",",
options.HasHeader = true,
options.Column.Add("SepalLength", DataKind.R4, 0);
options.Column.Add("SepalWidth", DataKind.R4, 1);
options.Column.Add("PetalLength", DataKind.R4, 2);
options.Column.Add("PetalWidth", DataKind.R4, 3);
options.Column.Add("Label", DataKind.Text, 4);

var reader = mlContext.Data.CreateTextFileSource(options);
```
- Tom check what you were plan on checking :-)

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💯

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It turns out the answer was, "we did both." 😛 Issue here.