Seaborn allows the user to input "estimator" variable in its various plots, for example: https://seaborn.pydata.org/generated/seaborn.catplot.html
I understand that the estimator should be a function of "vector" to "scalar" i.e. "mean" takes vector to and outputs a scalar (it's average value).
So far, I found "mean" and "sum" to be valid estimators, while "median" is not. Are there any other?
Many numpy functions can be used directly, for example np.median. mean is supported as np.mean. len, sum, max and min are built-in python functions and work without being imported from a library.
You can also provide your own function. For example estimator=lambda x: sum(xi*xi for xi in x). Or lambda x: np.percentile(x, 95).
Further note that many seaborn functions draw a confidence interval, calculated via "bootstrapping", which doesn't play well with all functions. You can set errorbar=None (in versions before 0.13: ci=None) to suppress the confidence interval.