If no other training data, don't add any keyterms

[johnml1135]

Resolves #569

---

This change is 

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 62.92%. Comparing base (0cfd145) to head (b0c0286).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #580      +/-   ##
==========================================
- Coverage   62.97%   62.92%   -0.06%     
==========================================
  Files         279      279              
  Lines       13984    13989       +5     
  Branches     1814     1817       +3     
==========================================
- Hits         8807     8803       -4     
- Misses       4551     4560       +9     
  Partials      626      626              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Enkidu93]

Reviewed 3 of 4 files at r1, all commit messages.
Reviewable status: 3 of 4 files reviewed, 2 unresolved discussions (waiting on @ddaspit and @johnml1135)

---

src/Machine/test/Serval.Machine.Shared.Tests/Services/PreprocessBuildJobTests.cs line 747 at r1 (raw file):

        Assert.That(
            pretranslations[2]!["translation"]!.ToString(),
            Is.EqualTo("Source one, chapter twelve, verse one.")

Is there a way to 'save' this test? Or are we happy for it just to be saved in the commit history? I think ultimately we'll need a test like this once we enable some kind of filtering and, as you can see, it's a fairly long test.

---

src/ServiceToolkit/src/SIL.ServiceToolkit/Services/ParallelCorpusPreprocessingService.cs line 89 at r1 (raw file):

            }

            if (useKeyTerms)

Why not just && parallelTrainingDataPresent?

[Enkidu93]

My only other concern is that I think the EITL team needs to 'buy in' before we do this. It's my impression that they already feel undermined in how KBT changes were done/unstable in regard to what the behavior is in Serval/SF. Even a quick email or Slack message that has an asking-not-telling tone would do it. Not sure if that's already been done.

Reviewable status: 3 of 4 files reviewed, 2 unresolved discussions (waiting on @ddaspit and @johnml1135)

[ddaspit]

I agree. We need buy-in from the EITL team before we proceed.

Reviewed 4 of 4 files at r1, all commit messages.
Reviewable status: all files reviewed, 2 unresolved discussions (waiting on @johnml1135)

[johnml1135]

src/Machine/test/Serval.Machine.Shared.Tests/Services/PreprocessBuildJobTests.cs line 747 at r1 (raw file):

Previously, Enkidu93 (Eli C. Lowry) wrote…

Is there a way to 'save' this test? Or are we happy for it just to be saved in the commit history? I think ultimately we'll need a test like this once we enable some kind of filtering and, as you can see, it's a fairly long test.

We could "ignore" or otherwise pass over the test. That would be about equivalent as to saving it in commit history, but just more visible. I am fine either way. Also, if we actually implement this, we will likely want to change it to "filter on the pretranslations" rather than on the training data.

[johnml1135]

src/ServiceToolkit/src/SIL.ServiceToolkit/Services/ParallelCorpusPreprocessingService.cs line 89 at r1 (raw file):

Previously, Enkidu93 (Eli C. Lowry) wrote…

Why not just && parallelTrainingDataPresent?

Because we won't know until every corpus has been run through whether there is any data at all. The simplest solution that I could think of was to collect them all and if there was data, add them when all the corpora are looped through. Otherwise, we would have to loop through the corpora twice.

[johnml1135]

Request for input made.

[Enkidu93]

In the meeting that you had to leave, @johnml1135, I got the go-ahead from the EITL team for this PR.

[Enkidu93]

Reviewable status: all files reviewed, 2 unresolved discussions (waiting on @johnml1135)