NMF_Base.py

# Special Thanks to Satwik Bhattamishra
# License: MIT

"""
Base class used all the methods
"""
import numpy as np
import scipy.sparse
import numpy.linalg as LA
from scipy.stats import entropy
from sys import exit

class NMFBase():

	def __init__(self, X, rank=10, **kwargs):

		self.X = X
		self._rank = rank


		self.X_dim, self._samples = self.X.shape

	def frobenius_norm(self):
		""" Euclidean error between X and W*H """

		if hasattr(self,'H') and hasattr(self,'W'):
			error = LA.norm(self.X - np.dot(self.W, self.H))
		else:
			error = None

		return error

	def kl_divergence(self):
		""" KL Divergence between X and W*H """

		if hasattr(self,'H') and hasattr(self,'W'):
			V = np.dot(self.W, self.H)
			error = entropy(self.X, V).sum()
		else:
			error = None

		return error

	def initialize_w(self):
		""" Initalize W to random values [0,1]."""

		self.W = np.random.random((self.X_dim, self._rank))

	def initialize_h(self):
		""" Initalize H to random values [0,1]."""

		self.H = np.random.random((self._rank, self._samples))

	def update_h(self):
		"Override in subclasses"
		pass

	def update_w(self):
		"Override in subclasses"
		pass


	def check_non_negativity(self):

		if self.X.min()<0:
			return 0
		else:
			return 1

	def compute_factors(self, max_iter=100):
		if self.check_non_negativity():
			pass
		else:
			print("The given matrix contains negative values")
			exit()

		if not hasattr(self,'W'):
			self.initialize_w()

		if not hasattr(self,'H'):
			self.initialize_h()


		self.frob_error = np.zeros(max_iter)

		for i in range(max_iter):

			self.update_w()

			self.update_h()

			self.frob_error[i] = self.frobenius_norm()