Let's say I have a column x with uniform distributed values. To these values, I applied a cdf-function.

Now I want to calculate the Gaussian Copula, but I can't find the function in python. I read already, that Gaussian Copula is something like the "inverse of the cdf function".

The reason why I'm doing it comes from this paragraph:

A visual depiction of applying the Gaussian Copula process to normalize an observation by applying ๐‘› = Phi^-1(๐น(๐‘ฅ)). Calculating ๐น(๐‘ฅ) yields a value ๐‘ข โˆˆ [0, 1] representing the proportion of shaded area at the left. Then Phi^โˆ’1(๐‘ข) yields a value ๐‘› by matching the shaded area in a Gaussian distribution.

I need your help, does everyone has an idea how to calculate that?

I have 2 ideas so far:

1) gauss = 1/(sqrt(2*pi)*s)*e**(-0.5*(float(x-m)/s)**2) --> so transform all the values with this to a new value

2) norm.ppf(array,loc,scale) --> So give the ppf function the mean and the std and the array and it will calculate me the inverse of the CDF... But I doubt #2

The thing is


Is not what I want. The idea why I'm doing it, is transforming a not normal/gaussian distribution to a normal distribution.

Like here:

Transform from a non gaussian distribution to a gaussian distribution with Gaussian Copula

Any other ideas or tipps?

Thank you very much :)


I found 2 links which are quite usefull: 1. How to transform an arcsine distribution to a normal distribution? 2. Transformation Chi-squared to Normal distribution

In this posts its said that you have to

All the details are in the answer already - you take your random variable, and transform it by its own cdf ..... yielding a uniform result.

Thats true for me. If I take a random distirbution and apply the norm.cdf(data, mean,std) function, I get a uniform distributed cdf

Compare: import pandas as pd data = [1.5]*7 + [2.5]*2 + [3.5]*8 + [4.5]*3 + [5.5]*1 + [6.5]*8 cdf=n.cdf(data, n.mean(data),n.std(data)) print cdf

But How can I do the

You then transform again, applying the quantile function (inverse cdf) of the desired distribution (in this case by the standard normal quantile function /inverse of the normal cdf, producing a variable with a standard normal distribution).

Because when I use f.e. the norm.ppf function, the values are not reasonable

EDIT: This link could be useful too enter link description here


  1. inverse transform method
  2. Box-Mueller Method
  3. invgauss
  4. ppf of cdf
  5. gauss = 1/(sqrt(2*pi)*s)*e**(-0.5*(float(x-m)/s)**2)
  • 1
    $\begingroup$ You want the inverse CDF of a standard normal distribution. See here for an implementation. $\endgroup$
    – Cliff AB
    Nov 18, 2017 at 16:37
  • $\begingroup$ In fact, norm.ppf should give you the result you want, but for some reason you think the values are not reasonable. Can you explain why you think they are unreasonable? $\endgroup$
    – Cliff AB
    Nov 18, 2017 at 16:41

2 Answers 2


Based on the answer provided by the OP, I believe the issue is not with the transformation into the copula space (i.e. applying the inverse CDF of a standard normal), but rather the transformation into uniform random variables from the raw data.

Recall that with copula models, there are two parts. First is modeling the marginal distributions for each variable and the second is modeling the copula, which defines the joint CDF of the transformed values.

From what I understand about your code, it appears that the marginal distributions are very poor fits. In particular, I believe that

cdf = n.cdf(data, mean, sd)

is modeling the marginal as a normal distribution. However, we can see from the plot that the data does not look normal at all!

Assuming you are dealing with a continuous variable (discrete copulas are a bit trickier), one of the easiest methods is to use the empirical distribution function, which essentially assigns equal probability to each observed value to model the CDF. It is worth noting that for copulas, you will need a special form (see alternative definition in link) of it to make sure all points are mapped to (0,1) rather than (0,1]. By using the EDF to model a variable's CDF, you are guaranteed that the observed values will be mapped to a uniform distribution when plugged into the EDF function, where as if you were to use a marginal distribution that was a poor fit, this would not happen.

For rather technical reasons that I will skip for now for the sake of brevity, this method should only be used for continuous (or at least nearly continuous, i.e. not too much probability mass should be assigned to any one value) variable.

Here's a demonstration in R on how using the wrong CDF can lead to very non-normal data after the transformations and how using the EDF can fix that up.

# Simulating highly bimodal data
raw_data = c(rnorm(n = 500, mean = 0, sd = 1), 
             rnorm(n = 500, mean = 6, sd = 2))
hist(raw_data, main = 'Raw Data')


# Fitting marginal as if normal
# (pnorm = cdf of a normal)
norm_cdf = pnorm(raw_data, 
                 mean = mean(raw_data), 
                 sd = sd(raw_data))
# We see that the CDF values are very non-uniform!
hist(norm_cdf, main = "CDF values from Normal marginal") 


# As a consequence, our copula values are very non-normal
# (qnorm = quantiles of a normal)
cop_vals = qnorm(norm_cdf)
hist(cop_vals, main = "Copula values from Normal marginal")


# Now using the EDF instead of normal cdf
# Using version that ensures all cdf 
# values will be in (0,1), not (0, 1]
edf = rank(raw_data) / (length(raw_data) + 1)
# By definition, these values will be uniform
hist(edf, main = "CDF values from EDF marginal")


# And so the copula values will be normal
cop_vals = qnorm(edf)
hist(cop_vals, main = "Copula values from EDF marginal")

enter image description here


I have to answer like that, because I can't answer on @CliffAB (I dont have 50 reputation)

The reason why I think its not working is easy to show on an example:

data = [1.5]*7 + [2.5]*2 + [3.5]*8 + [4.5]*3 + [5.5]*1 + [6.5]*8
cdf = n.cdf(data, mean, std)
ppf = n.ppf(cdf)
df = pd.DataFrame(ppf)

When I plot the ppf, the data looks exactly the same as the original data. But the original data is not normal distributed, so I expected it to look more like

enter link description here

In fact, the ppf function just makes the original data again. But I want to normalize the original data (which is not normal distributed) to a normal distributed one. (according to the picture). According to other post, if you do a cdf of a dataset and then use the normal invers of cdf, the values should be normal distributed, what is not the case

Maybe I explain myself wrong, if you have questions, feel free to ask....

The plot of my original Data:

enter image description here

Plot of my ppf Data... Same shape, but I want a normal distribution enter image description here


Your Answer

By clicking โ€œPost Your Answerโ€, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.