Hello all ! I'm trying to incorporate plotly visualizations in my mlflow. I've created a simple function:

def make_plotly():
    import plotly.express as px
    df = px.data.iris()
    fig = px.scatter(df, x="sepal_width", y="sepal_length", color="species",
                    size='petal_length', hover_data=['petal_width'])
    fig.show()
    return fig

And created a pipeline where the output of the associated node is stored in the catalog

plotly_fig:
  type: kedro_mlflow.io.artifacts.MlflowArtifactDataset
  dataset:
    type: plotly.PlotlyDataset
    filepath: data/08_reporting/plotly.json
    plotly_args:
      type: scatter

After running the pipeline, the fig properly shows up in a tab, then is stored in a json format. However, in mlflow ui, it only appears as the json file (see attached image). • Is there a way to show the plotly fig in mlflow ? Do I miss something ? • Also, what args would you recommend to save and load the plotly fig ?

Someone will reply to you shortly. In the meantime, this might help:

hey @Philippe Martin it seems like MLflow provides

mlflow.log_artifact

. it is stored in the artifact repository associated with the current MLflow run. These artifacts can then be viewed and downloaded from the MLflow UI. Perhaps this can help to visualise your plotly in MLFlow? https://mlflow.org/docs/latest/traditional-ml/hyperparameter-tuning-with-child-runs/notebooks/logging-plots-in-mlflow.html

Unfortunatly no.. the artifacts is well saved as a json file in the artificat repository, but it doesn't show as a plotly figure in the MLflow UI, it shows as a json file. Maybe the error comes from how I stored the figure in the catalog ?

plotly_fig:
  type: kedro_mlflow.io.artifacts.MlflowArtifactDataset
  dataset:
    type: plotly.PlotlyDataset
    filepath: data/08_reporting/plotly.json
    plotly_args:
      type: scatter

Thank you for your help

Hi @Philippe Martin It is possible to log plotly figure with mlflow so that they show in mlflow UI. The artifact needs to be stored as HTML. Look into mlflow.log_figure. If I recall properly, it should be called that way:

mlflow.log_figure(figure=figure, artifact_file=f"{name}.html")

where figure is your plotly figure. I am not just sure how it would work with the plotly dataset.

The plotly dataset in kedro-datasets returns a figure object, which works well with Kedro-Viz. However, MLflow doesn't support this json format. A great idea would be to create an

plotly.MLFlowDataset

dataset that would return the html as you said. We are already have a plotly.HTMLDataset

Amazing, thank you !!