GCCA

This repository is implementation of Generalized Canonical Correlation Analysis(GCCA). CCA can use only 2 data but GCCA can use more than 2 data.

CCA

CCA is the method to transform 2 data to one joint space. See example graph:

CCA inplementation contains PCCA (Probablistic Canonical Correlation Analysis) transformation that is assumed that there is latent space in 2 data.

GCCA

GCCA is the method to transform multiple data to one joint space. See example graph:

You can give GCCA any number of data.

Installation

You can use 'git clone' command to install

Dependencies

You have to install python dependent libraries in advance as follow:

numpy==1.9.1
scipy==0.14.1
matplotlib==1.4.2
h5py==2.4.0

Usage of CCA

from cca import CCA
import logging
import numpy as np

# set log level
logging.root.setLevel(level=logging.INFO)

# create data in advance
a = np.random.rand(50, 50)
b = np.random.rand(50, 60)

# create instance of CCA
cca = CCA()
# calculate CCA
cca.fit(a, b)
# transform
cca.transform(a, b)
# transform by PCCA
cca.ptransform(a, b)
# save
cca.save_params("save/cca.h5")
# load
cca.load_params("save/cca.h5")
# plot
cca.plot_pcca_result()

Usage of GCCA

from gcca import GCCA
import logging
import numpy as np

# set log level
logging.root.setLevel(level=logging.INFO)

# create data in advance
a = np.random.rand(50, 50)
b = np.random.rand(50, 60)
c = np.random.rand(50, 70)
d = np.random.rand(50, 80)
e = np.random.rand(50, 90)
f = np.random.rand(50, 100)
g = np.random.rand(50, 110)
h = np.random.rand(50, 120)
i = np.random.rand(50, 130)
j = np.random.rand(50, 140)
k = np.random.rand(50, 150)

# create instance of GCCA
gcca = GCCA()
# calculate GCCA
gcca.fit(a, b, c, d, e, f, g, h, i, j, k)
# transform
gcca.transform(a, b, c, d, e, f, g, h, i, j, k)
# save
gcca.save_params("save/gcca.h5")
# load
gcca.load_params("save/gcca.h5")
# plot
gcca.plot_gcca_result()

That's it!