ECDF Normalization for Multiple Categories

I am using plotly 5.5.0. - What I noticed is that the normalization of ecdf is not done with the number of samples per category. For example, these are number of counts:


but the normalization leads to this:

The normalization is not done with the total counts per category, but somehow with a “total number”.
In the documentation, the different number of counts per category are considered correctly.
Empirical cumulative distribution plots in Python (plotly.com)
Can somebody explain how the number of counts are normalized? What is the “total number” for normlization?

Updated to plotly 5.8.2 (Python 3.9), but I have still the same issue with normalization:

Oh, I found the issue. I have a multi-index data frame that I “melt” before handing it over to plotly. Because the data series have different length, there are empty cells, which are counted for normalization as well. Removing the empty cells solves the problem:

error_total = error.melt(ignore_index=False)
error_total.dropna(subset=["value"], inplace=True)

fig = px.ecdf(error_total, x="value", facet_col="eval_type", color="est_type", hover_name="est_type", ecdfnorm="percent")