Beautiful. Thanks! I love how you find interesting ways to find outliers! I would recommend mentioning in this context cumulative distribution function (CDF) plots. They are a good replacement for histograms and have no Hyperparameters.
Vincent here. CDFs certainly have benefits, but how might you use them to detect outliers in a multi modal distribution in higher dimensions? The density estimator also has some benefits in this domain and I worry that CDFs may fall short here.
@@probabl_ai I originally meant that it's worth mentioning in the context of showing the distribution of the data without HPs (bins, kernels). And now, since you asked about outliers, an idea came up - maybe it's interesting to look for points that live in an area of the cdf where there is a sudden "jump" upward on the y-axis. This is equivalent to points being in an area that is less populated on the histogram.
Your explanations are very clear. Congratulations!
Thank you Vincent. Learning new useful things from you every time. Really interesting.
Happy to hear it!
Beautiful. Thanks! I love how you find interesting ways to find outliers!
I would recommend mentioning in this context cumulative distribution function (CDF) plots. They are a good replacement for histograms and have no Hyperparameters.
Vincent here.
CDFs certainly have benefits, but how might you use them to detect outliers in a multi modal distribution in higher dimensions? The density estimator also has some benefits in this domain and I worry that CDFs may fall short here.
@@probabl_ai I originally meant that it's worth mentioning in the context of showing the distribution of the data without HPs (bins, kernels).
And now, since you asked about outliers, an idea came up - maybe it's interesting to look for points that live in an area of the cdf where there is a sudden "jump" upward on the y-axis. This is equivalent to points being in an area that is less populated on the histogram.