wittawatj / kernel-gof Goto Github PK

View Code? Open in Web Editor NEW

65.0 6.0 16.0 3.56 MB

NeurIPS 2017 best paper. An interpretable linear-time kernel goodness-of-fit test.

License: MIT License

Jupyter Notebook 46.07% Shell 0.73% Python 53.10% Dockerfile 0.10%

machine-learning kernel-methods goodness-of-fit statistical-tests python

kernel-gof's People

Contributors

Stargazers

Watchers

Forkers

botev oztc sweatyrichard afcarl djsutherland lebrave2015 thanaphon prakhar-agarwal codelearning123 andreas-koukorinis kronerte ccheng2021 darsh-patel ahmad-abdellatif notthatanonymous

kernel-gof's Issues

Dockerfile buid failed

add --user to the end of github url as below:
RUN apt-get update --allow-releaseinfo-change && apt-get install -y gcc && pip install jupyter && pip install git+https://github.com/wittawatj/kernel-gof.git --user

can this algorithm compare differences in two matrices

hi @wittawatj !!

i hope your well.

I was wondering if you have a matrix of returns, then new data comes in and you have a second rolling matrix of returns, can the test tell if the distribution has changed from one matrix to the next?

Kind regards,
Andrew

When I try to work with 1D arrays - i get IndexError: too many indices for array

Am I doing something wrong, or code just designed for 2D arrays only?

Here is the code and the error

X = np.random.poisson(2, 400)
tr, te = train_test_split(X, test_size=0.8)
V_opt, gw_opt, opt_info = gof.GaussFSSD.optimize_auto_init(p, data.Data(tr), J, **opts)

IndexError                                Traceback (most recent call last)
<ipython-input-40-380dbe5289fc> in <module>
     14 # make sure to give tr (NOT te).
     15 # do the optimization with the options in opts.
---> 16 V_opt, gw_opt, opt_info = gof.GaussFSSD.optimize_auto_init(p, data.Data(tr), J, **opts)

~\AppData\Local\Continuum\anaconda3\envs\ml_python36\lib\site-packages\kgof\goftest.py in optimize_auto_init(p, dat, J, **ops)
    499         n_gwidth_cand = 5
    500         gwidth_factors = 2.0**np.linspace(-3, 3, n_gwidth_cand)
--> 501         med2 = util.meddistance(X, 1000)**2
    502 
    503         k = kernel.KGauss(med2*2)

~\AppData\Local\Continuum\anaconda3\envs\ml_python36\lib\site-packages\kgof\util.py in meddistance(X, subsample, mean_on_fail)
    146         np.random.set_state(rand_state)
    147         # recursion just one
--> 148         return meddistance(X[ind, :], None, mean_on_fail)
    149 
    150 

IndexError: too many indices for array

Gamma Distribution when k=1

Hi, I tried to test gamma distribution with your test.
I used a gradient log to specify p as below:

def gamma_grad_log(X):
    return ((k-1)/X - 1/theta)*(X>0)

When k=2 and theta=2, and build a sample data from same gamma distribution as below

k_t = 2
theta_t = 2
n=500
seed = 4
np.random.seed(seed)
X = np.random.gamma(k_t, theta_t,(n,1))

In this case, the test works fine - not reject H_0.
{'alpha': 0.01,
'h0_rejected': False,
'n_simulate': 3000,
'pvalue': 0.31133333333333335,
'test_stat': 0.0003191977834490073,
'time_secs': 0.001219034194946289}

But if k=1, the test will reject H_0.
{'alpha': 0.01,
'h0_rejected': True,
'n_simulate': 3000,
'pvalue': 0.0,
'test_stat': 100.18839042139058,
'time_secs': 0.0012459754943847656}

wittawatj / kernel-gof Goto Github PK

kernel-gof's People

Contributors

Stargazers

Watchers

Forkers

kernel-gof's Issues

Dockerfile buid failed

can this algorithm compare differences in two matrices

When I try to work with 1D arrays - i get IndexError: too many indices for array

Gamma Distribution when k=1

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent