Two questions about distfit

[erdogant]

Two questions about distfit

[erdogant]

Dear lesshaste, Thank you for your feedback!
You are right about levy_l (and also stable), these are very slow so I removed them from the full list.
The loc parameter is automatically determined. There is no input parameter to force this to be a specified value.
Can you describe why you want to fixate this parameter?

try to pip install the new version and check the version. It should be 1.1.5.

pip install -U distfit
import distfit
print(distfit.__version__)

[lesshaste]

Thank you for the reply. In my case I know the distribution is symmetric with mean (and median) 0. So it makes sense to use that information I was thinking.

[erdogant]

I you have sufficient data, your mean value will be towards zero. Maybe not exactly 0 but very close to it. If your data is symmetric, it will try to fit a distribution that shows the best fit. The fun part is that you do not have to worry about these parameters but are estimated for you.

For example a normal distribution with 10000 data points with mean ~0

X = np.random.normal(0, 2, 10000)

from distfit import distfit
dist = distfit()
model = dist.fit_transform(X)
dist.plot()

This gives the following output:

[distfit] >[norm ] [RSS: 0.0009400] [loc=0.017 scale=1.993]
[distfit] >[expon ] [RSS: 0.3300147] [loc=-8.140 scale=8.157]
[distfit] >[pareto ] [RSS: 0.3251750] [loc=-406937692.614 scale=406937684.474]
[distfit] >[dweibull ] [RSS: 0.0070768] [loc=0.011 scale=1.698]
[distfit] >[t ] [RSS: 0.0009299] [loc=0.017 scale=1.988]
[distfit] >[genextreme] [RSS: 0.0028985] [loc=-0.759 scale=1.961]
[distfit] >[gamma ] [RSS: 0.0008770] [loc=-85.198 scale=0.047]
[distfit] >[lognorm ] [RSS: 0.0009010] [loc=-74.206 scale=74.190]
[distfit] >[beta ] [RSS: 0.0008786] [loc=-82.021 scale=25037346.492]
[distfit] >[uniform ] [RSS: 0.2463460] [loc=-8.140 scale=16.556]

In case of a normal distribution the location (loc) keyword specifies the mean. The scale (scale) keyword specifies the standard deviation. But in different distribution these can have different properties. Mostly to shift and/or scale the distribution. So it may not be so straightforward to set these manually.

https://docs.scipy.org/doc/scipy/reference/stats.html