kz930 opened a new issue, #6790:
URL: https://github.com/apache/texera/issues/6790
### What happened?
`MachineLearningScorerOpDesc.getSelectedMetrics()` returns the chosen metric
names as one comma-separated, already-quoted fragment, and its return type is
**`EncodableString`**:
`common/workflow-operator/src/main/scala/org/apache/texera/amber/operator/machineLearning/Scorer/MachineLearningScorerOpDesc.scala`
```scala
private def getSelectedMetrics(): EncodableString = {
val metric = if (isRegression) regressionMetrics else classificationMetrics
metric.map(metric => getMetricName(metric)).mkString("'", "','", "'") //
-> 'Accuracy','F1 Score'
}
```
It is spliced into the generated Python as a list:
```
metric_list = [${getSelectedMetrics()}]
```
Because the value is an `EncodableString`, the Python template builder
**re-encodes the whole fragment as one Python string value** (rendered via
`self.decode_python_template(...)`) instead of splicing it verbatim. The
generated code becomes:
```python
metric_list = ["'Accuracy','F1 Score'"] # ONE element: the literal string,
incl. inner quotes
```
instead of the intended:
```python
metric_list = ['Accuracy', 'F1 Score']
```
So every downstream `if 'X' in metric_list` and `for metric in metric_list`
/ `metrics_func[metric]` operates on that single malformed element — the
selected metrics are never matched, and the classification path hits a
`KeyError` on the bogus key. The Scorer produces wrong/empty output for both
the classification and regression paths.
**Expected:** `metric_list` is a proper list of metric names and each
selected metric is computed.
### How to reproduce?
**A — user-facing:** Add a **Machine Learning Scorer** operator, select one
or more metrics (e.g. Accuracy, F1 Score), give it prediction/label columns,
and run. The selected metrics are not computed correctly (empty result /
`KeyError`).
**B — confirmed against `main`'s operator directly:** construct
`MachineLearningScorerOpDesc` with `isRegression = false` and
`classificationMetrics = List(accuracy, f1Score)`, then call
`generatePythonCode()`. The emitted line is:
```
metric_list = [self.decode_python_template('J0FjY3VyYWN5JywnRjEgU2NvcmUn')]
```
The base64 `J0FjY3VyYWN5JywnRjEgU2NvcmUn` decodes to `'Accuracy','F1 Score'`
— i.e. the entire quoted list collapses into a single decoded string element,
so `metric_list` has one malformed entry rather than two metric names.
**Proposed fix:** change `getSelectedMetrics()`'s return type from
`EncodableString` to plain `String`, so the builder splices it verbatim into
`metric_list = ['Accuracy','F1 Score']`. The method body is unchanged.
### Version/Branch
`main` (1.3.0-incubating-SNAPSHOT)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
