pca_control.py

import numpy as np

class pca:
   def __svd_thin(self, a):
      u, s, vh = np.linalg.svd(a, full_matrices=False, compute_uv=True)
      return ( u, s, vh.T )

   def __eig_symmetric(self, a):
      c    = np.dot(a.T, a) / (a.shape[0] - 1)
      e, v = np.linalg.eig(c)
      i    = e.argsort()[::-1]
      return ( e[i], v[:,i], c )

   def __fit_transform(self, svd=True):
      if svd is True:
         u, s, v = self.__svd_thin(self.__a)
         p       = np.square(s) / (self.__m - 1)
         x       = np.dot(u, np.diag(s))
         c       = v @ np.diag(p) @ v.T
         #c       = v @ np.diag(p) @ np.linalg.inv(v)
         c_verif = np.dot(self.__a.T, self.__a) / (self.__m - 1)
         return ( x, p, v, c, c_verif )
      else:
         p, v, c_verif = self.__eig_symmetric(self.__a)
         x             = np.dot(self.__a, v)
         c             = v @ np.diag(p) @ v.T
         #c             = v @ np.diag(p) @ np.linalg.inv(v)
         return ( x, p, v, c, c_verif )

   def __init__(self, a):
      self.__a           = np.array(a)
      self.__m, self.__n = self.__a.shape
      self.__a           = self.__a - self.__a.mean(axis=0)

   def fit(self, svd=True):
      return self.__fit_transform(svd)

a           = np.random.rand(1200, 3)
o           = pca(a)

print("")

x1, p, c, _ = o.fit()

print(x1)
print(p)
print(c)

print("")

x2, p, c, _ = o.fit(False)

print(x2)
print(p)
print(c)

print(np.allclose(np.abs(x1), np.abs(x2), atol=10E-05))

# EOF