-
Notifications
You must be signed in to change notification settings - Fork 231
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Skewed Lognormal Distributions #584
Comments
Just to check, are you specifying the relevant factors as having a log-normal distribution? See: https://salib.readthedocs.io/en/latest/user_guide/advanced.html#generating-alternate-distributions Otherwise, I suggest trying to transform the data to a gaussian distribution. |
@ConnectedSystems yes, I went through the referred page and for the given data I calculated the ln-spaced mean and standard deviation. However, for skewed lognormal distributions these two parameters are not sufficient to describe the distribution. Regarding the transformation: I am not quite sure how this will solve the problem. I tried to fit a normal distribution to the data in the original space as well, however, this (similar to fitting a lognormal distribution) did not give a good fit. Thus, I believe that there needs to be a way to include the skewness of the distribution as well in the input parameters. In scipy this is done by giving the three parameters location, scale and shape (https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.lognorm.html). However, I do not see any equivalent in the docs for SA-lib to include that. |
Sorry, just to check again - when you say you are trying to conduct a Sobol' sensitivity analysis, is this data set you're referring to the output from the Sobol' sampling method as provided in SALib? Just want to confirm this, as what you've written makes it sound like you already have some data on hand. For clarity, the Sobol' method requires a specific sampling scheme, although SALib does offer given-data approaches such as Delta, PAWN, and HDMR. The new discrepancy analysis method is said to be comparable to Sobol' total order indices as well.
I was suggesting that you modify the data set so that it follows a normal distribution. https://scikit-learn.org/stable/auto_examples/preprocessing/plot_map_data_to_normal.html |
Hi everyone,
I am trying to run a sobol analysis on a dataset which is distributed lognormally but with a skewness in it. Consequently, I am not able to represent the data accurately with the ln-spaced mean and standard deviation.
This is how the data looks like in the original feature-space:
This is how the data looks like once I apply ln to the x-axis:
One can clearly see that the data has some skewness in it. If I simply take the mean and standard deviation of the log-transformed data and try to sample from the distribution I don't get any good samples. I also tried calculating the mean and standard deviation using the location, scale and shape parameter which I got from from scipy.stats.lognorm.fit(). Am I missing something here, would be great if someone is able to help.
Best
The text was updated successfully, but these errors were encountered: