gibsramen / evident_ Goto Github PK

At the moment, power analysis calculations assume that the same number of samples will be used in each group. Could be useful to allow the user to pass a list of proportions corresponding to the ratio of each group to the whole.

At the moment I think it's unclear whether or not an EffectSizeResults artifact was generated via whole group or pairwise comparisons. I think we could do something like EffectSizeResults[Group] and EffectSizeResults[Pairwise] to better delineate these two.

Seems like this line fails when you pass in a np.array

https://github.com/gibsramen/evident/blob/f1a0fd65100e09f46c3e20c1e343d32f6e3e5943/evident/diversity_handler.py#L119

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Input In [180], in <module>
----> 1 [[1, 2, 3], np.linspace(10, 30, 5), None].count(None)

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

https://pingouin-stats.org/

Seems like this package might be easier to work with for power analysis & effect size calculations.

Idea courtesy of @cameronmartino

Something where you can maybe slide different values of alpha, total observations, etc. and see how the power changes. I think something Plotly or Bokeh might be able to do this but I want to write as little JavaScript as possible.

Hello,

Sorry for the novice question. I noticed that I need to input metadata.qza when running commands such as "qiime evident univariate-power-analysis ".

I have only ever used metadata as a csv file. How should I convert it to qza format?

Thank you.

I don't think these need to be separate. Or at least we should also include the group level comparisons in EffectSizeResult. This would also benefit from being able to assign sign to effect sizes which currently is not supported.

We could probably parallelize certain parts of the effect size/power calculations.

We currently allow difference to be passed such that difference/pooled_std_dev is used as the Cohen's d effect size. It's not immediately clear to me that this works for Cohen's f when there are more than two groups - something to look into.

https://github.com/gibsramen/evident/blob/d1a3fe636764f47278d4cbade05534a2b57d1b87/evident/_exceptions.py#L61

If NaN exists, this line will fail because the number of "unique" levels > 1

Would be useful for, as an example, running effect size calculations on multiple columns for debugging purposes.