Title
#plugins-integrations
Galileo-Galilei

Galileo-Galilei

02/14/2022, 10:06 PM
@User 1. You can try @User suggestion which is a really good idea! 2. You do not need a node for each metric, you can return a tuple for a node, e.g. :
python
# in node.py
def compute_metrics (y_true, y_preds):
    <compute>
    return metric1, metric2, [metric3_step0, metric3_step1, metric3_step2, metric3_step3]

# in pipeline_registry.py

Pipeline([
..., 
node(compute_metrics, ["y_true", "y_preds"], [my_metric1, my_metric2, my_metric3]),
...
])

# in catalog.yml
my_metric1:
    type: kedro_mlflow.io.metrics.MlflowMetricDataSet

my_metric2:
    type: kedro_mlflow.io.metrics.MlflowMetricDataSet

my_metric3:
    type: kedro_mlflow.io.metrics.MlflowMetricHistoryDataSet
This is easier than returning the complex format above 3. Unfortunately this is not linked to kedro-mlflow, I don't modify the UI. I know that mlflow often changes its UI and it slightly varies across versions. Feel free to [open an issue to my repo](https://github.com/Galileo-Galilei/kedro-mlflow/issues), I'll add it to my backlog to log it as a tag in the future so we can have consistent access across mlflow versions.