Boxplots No Aggregate

[vincent-reau]

Hi everyone,

I come with a tricky question that I stumbled upon recently.
I want to display in my boxplot raw values that are already scaled (0 to 1) by variable but it seems I can't do boxplots without aggregating.

Here is the error :
Error: column "v_measurements_scaled.value_scaled" must appear in the GROUP BY clause or be used in an aggregate function LINE 4: value_scaled AS value_scaled ^

My dataset schema :

• datetime_value : timestamp
• id_measure: int
• variable_name: string
• value_scaled: float

How could I display this boxplots without an average or a count aggregation ?

Any help is appreciated,
Thanks

[dosubot[bot]]

Hi @vincent-reau! I'm Dosu and I’m helping the superset team.

Superset's boxplot visualization requires aggregation because it computes statistics like mean, median, quartiles, min, max, and outliers for each group using SQL queries that group and aggregate your data. This is why you see the error about needing to use your column in a GROUP BY clause or aggregate function—the boxplot chart expects grouped and aggregated data to calculate its distribution metrics, not raw values directly. There isn't a built-in way to display boxplots of raw (non-aggregated) values in Superset; aggregation is a core part of how boxplots are implemented in the platform see source.

If you want to visualize raw scaled values without aggregation, you might consider using a different chart type, like a scatter plot or strip plot, which can show individual data points. Alternatively, you could export your data and use an external tool that supports raw-value boxplots. Superset does not currently support plugins or alternative chart types for boxplots without aggregation see source.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[vincent-reau]

Alright, I understand a bit more the problem, here are some more information about it :

In python, by using the same dataset, I can display a boxplot that have the correct values (code below).
It seems that superset boxplot pre-aggregates data and doesn't take raw inputs like in python. How could I adapt it in superset boxplot or even create a new box plot chart in superset github ?

summary = (
    df_f.groupby("variable_name_1")["value_scaled"]
      .agg(
          min="min",
          Q1=lambda x: x.quantile(0.25),
          median="median",
          mean="mean",
          Q3=lambda x: x.quantile(0.75),
          max="max"
      )
      .reset_index()
)

[dosubot[bot]]

Superset's boxplot chart requires aggregation because it computes statistics (like min, Q1, median, mean, Q3, max, count, outliers) for each group using SQL queries and post-processing, not from raw values directly. This is different from pandas, which can compute boxplot stats from raw data in memory. The frontend boxplot plugin expects these pre-aggregated statistics and cannot process raw data points as input see code.

To mimic pandas' behavior in Superset, you would need to create a new chart plugin that accepts raw data from the backend and computes boxplot statistics in the frontend. This would require significant changes: bypassing the enforced aggregation in both the backend query builder and the frontend chart logic, and ensuring the new plugin can handle large volumes of raw data efficiently see buildQuery. There are community resources on building custom visualization plugins in Superset, such as this blog post and video.

Currently, there is no built-in way to plot boxplots from raw, unaggregated data in Superset, and this feature would require a custom development effort see discussion.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[jpatelarkimetrix]

Hi @vincent-reau, Let me know if you were able to find a workaround of this?
I am struggling with the same issue.
Thanks