blent-ai / pycopula Goto Github PK

The pdf (method pdf_param) of Archimedian copulas seems to be incorrect if the dimension is 3 or higher, since I am getting negative and complex results.

I see that for n>2 a numeric differentiation with scipy is used.

I am a researcher on hydrology. In the field of hydrology, the marginal distribution is known as Pearson-III. So my question is how can I got the joint distributions by pycopulas with the known marginal distribution.

Is the data the value of Probability density function ,or the value of random variable？

Need to improve nearest Positive Definite Matrix function with BLAS subroutines.

(1001, 2)
Fitting Student copula.

UnboundLocalError Traceback (most recent call last)
in ()
7 cop = StudentCopula(dim=2)
8 print(X.shape)
----> 9 cop.fit(X, method='clme')
10
11 # Visualization of CDF and PDF

/usr/local/lib/python3.6/dist-packages/pycopula/copula.py in fit(self, X, method, df_fixed, verbose, **kwargs)
809 rho = fitted_params
810 else:
--> 811 nu = fitted_params[0]
812 rho = fitted_params[1:]
813

UnboundLocalError: local variable 'fitted_params' referenced before assignment

Hi, I'm curious that why GaussianCopula can't do with IFM estimation and StudentCopula can't do with MLE and IFM estimation.
I am very appreciate that you build up this package but just can't understand how it works?
Is there any detailed document?

Thanks

Hello!
I think there is some problems with freshness of documentation, because code from examples did not work. So, for example, I am trying to run this code: https://blent-ai.github.io/pycopula/build/html/examples.html#sampling
And I get the following errors:

---------------------------------------------------------------------------
TypeError                                 
Traceback (most recent call last)
<ipython-input-40-bb6934b931b1> in <module>()
----> 1 gaussian = GaussianCopula(dim=2, sigma=[[1, 0.8], [0.8, 1]])
 
TypeError: __init__() got an unexpected keyword argument 'sigma'

And after removing 'sigma' from arguments I've got a different error:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-38-a14416189e8d> in <module>()
----> 1 sim = simulate(archimedean, 100)

/usr/local/lib/python3.5/dist-packages/pycopula/simulation.py in simulate(copula, n)
     19                 The size of the sample.
     20 	"""
---> 21         d = copula.getDimension()
     22 
     23         X = []

AttributeError: 'GaussianCopula' object has no attribute 'getDimension'

I'm using Python 3.5.2 and downloaded PyCopula using pip.
If you need any help, please, let me know.

Hi there, your package works well in Python 3.7.1 on Windows.

For plotting, instead of giving examples in the documentation page, perhaps put plotting tools inside the visualization module / library.

For instance,

def plot_copula_3d(copula,plot_type='pdf',subplot=None):
    """Plots a pycopula copula object's pdf or cdf.

        Usage: 
            plt.figure()
            plot_copula_3d(copula,'pdf')
            plt.figure()
            plt.subplot(2,1,1)
            plot_copula_3d(copula, 'pdf', (2,1,1))
            plt.subplot(2,1,2)
            plot_copula_3d(copula, 'cdf', (2,1,2))
            plt.show()
    """
    if subplot is None: ax = plt.gca(projection='3d')
    else: # Overwrites the entire figure with single plot
        fig = plt.gcf()
        ax = fig.add_subplot(*subplot,projection='3d')
    if plot_type=='pdf':
        u,v,C = pycopula.visualization.pdf_2d(copula)
    elif plot_type=='cdf':
        u,v,C = pycopula.visualization.cdf_2d(copula)
    X, Y = np.meshgrid(u,v)
    ax.plot_surface(X,Y,C)
    ax.plot_wireframe(X,Y,C,color='black',alpha=0.3)

This gives an error if the number of hyperparameters that require estimation exceeds the number of dimensions (d). I think this might be an easy fix:

h=0
for dic in hyper_param:
h+=len(dic)

start_vector = np.repeat(1, h + thetaOffset)