Boxplots do not show values if there is only a single point

E.g.,

I would use barplot but I bar plot stacks the data over instead of computing means and do not see any possibility to compute means using weave.

Hi Wojciech,

Thanks for writing in! Would you mind sharing a link to this chart and so I can have a look at it? If I use boxplot, when hovering over my mouse I can see max/min and median, the extra feature that you would like to have is being able to see also the mean, is it?

Also, would you like to see data when there is only one point?

Best,
Luis

Exactly, mean on box plots would be a plus. But median would be enough if it is also shown for a single point data.

Box plots are the only way I found in wand to do something that I would do in Pandas using sth like:

runs.summary.groupby(‘dataset_name’).median(‘accuracy’)

Is there a way to do the above using weave? I cannot find anything like that in the documentation.

Hi Wojciech,

Thanks for sharing this! I can create a new request for this feature, I would really appreciate if you could share with me your use-case and if this is blocking you in any way.

In terms of finding a workaround for this, I am not sure if I fully understand what you are trying to do. If there is a single point, then the box-plot would not show the median since this has to be calculated with at least two points, so would you mind explaining me what would be your ideal behaviour? Also, it would be great if you could share a link to the Workspace where you have this data and so I can have a look at it.

Best,
Luis

Hi Wojciech,

I just wanted to follow up here, could you share with the case in which you are finding this feature useful? Also, would you mind explaining a little bit more about your use-case and so I can help you finding a workaround for this?

Best,
Luis

Use case: I test a number of algorithms on a set of datasets (or seeds). Thus, naturally, I’d like to group by algorithms and see a distribution of performance for the algorithms. Some algorithms take a lot of time, so I cannot repeat them more than once thus, instead of having a distribution, I’d like to have a single point shown on the plot.

As to the groupby. Let me show you an exemplary board with a few runs: Weights & Biases. There is a panel with summary table. I’d like to do some groupby on the summary to calculate mean over the groups (algorithms). I am aware of Weights & Biases but I don’t get how to use the groupby on run.summary and run.history. Not enough examples and documentation.

Use case: I test a number of algorithms on a set of datasets (or seeds). Thus, naturally, I’d like to group by algorithms and see a distribution of performance for the algorithms. Some algorithms take a lot of time, so I cannot repeat them more than once thus, instead of having a distribution, I’d like to have a single point shown on the plot.

As to the groupby. Let me show you an exemplary board with a few runs: Weights & Biases. There is a panel with summary table. I’d like to do some groupby on the summary to calculate mean over the groups (algorithms). I am aware of Weights & Biases but I don’t get how to use the groupby on run.summary and run.history. Not enough examples and documentation.

[Discourse post]

Oh. I c&p and the link to the example did not get through. Let me share it again: Weights & Biases

Hi Wojciech,

Thanks a lot for sharing this! I have been able to explore this in depth. Weave gets the data from our Database meanwhile operations in the Workspace (like group your runs) are not affecting the database, that is why you cannot see your metrics grouped in the Weave expression even when there is some kind of group in the Workspace. There is not any groupby option in Weave. I have already reported your feedback about boxplots, would you like me to also file a request for having groupby as an option in Weave expressions?

Best,
Luis

To me, weave without groupby is not useful.