Understanding the parameters of `em` and `mv` function

[vinhqdang]

Hi there

It is difficult for me to map the parameters of two functions em and mv

em(t, t_max, volume_support, s_unif, s_X, n_generated)

mv(axis_alpha, volume_support, s_unif, s_X, n_generated)

to the description in the paper.

Might you give some hints?

Thanks

[ngoix]

Hi,

sorry for the delay, and for this old unmaintained uncommented unclean code (I had not much experience in python at the time of writing)

• t is an array of levels, on which we want to evaluate EM_s(t) on samples X from an underlying density f.
• s_unif stands for s, evaluated on a uniform sample generated on a rectangle containing all the data X. This is used to estimate Leb(s>t)
• s_X stands for s(X), i.e s evaluated on a sample from the underlying density f. This is used to estimate P(s>t)
• n_generated is just the size of the uniform sample.
• volume_support is the volume of the rectangle containing all the data X.
• t_max should be called EM_min instead. We are only interested in the beginning of the EM curve.

hope that helps!

[vinhqdang]

Hi Nicolas

Many thanks.

Mind you answer one more question - if you don't mind.

Suppose I have a dataframe (df) that contains the data points, and a list (abnormal_score) that contains abnormal scores calculated by a model (One-class SVM for instance). Given that how should I call the function em or mv to calculate the score?

Thanks

[ngoix]

You need more information about the one-class SVM decision_function. More precisely, you have a dataframe representing decision_function(X), and you also need the same dataframe but for uniform samples on the rectangle containing your data X, i.e. decision_function(uniform_sample).

With this two dataframes you should be able to compute em and mv as done in em_bench.py for numpy arrays.