thenmf/bignmf

A python package for performing single NMF and joint NMF algorithms

BigNmf

BigNmf (Big Data NMF) is a python 3 package for conducting analysis using NMF algorithms.

NMF Introduction

NMF (Non-negative matrix factorization) factorizes a non-negative input matrix into non-negative factors. The algorithm has an inherent clustering property and has been gaining attention in various fields especially in biological data analysis.

Brunet et al in their paper demonstrated NMF's superior capability in clustering the leukemia dataset compared to standard clustering algorithms like Hierarchial clustering and Self-organizeing maps.

Available algorithms

The following are the algorithms currently available. If you would like to know more about the algorithm, the links below lead to their papers of origin.

• Single NMF
  1. Standard Single NMF
  2. Sparse NMF
• Joint NMF
  1. Standard Joint NMF
  2. Integrative NMF

Installation

This package is available on the PyPi repository. Therefore you can install, by running the following.

pip3 install bignmf

Usage

The following examples illustrate typical usage of the algorithm.

1. Single NMF

from bignmf.datasets.datasets import Datasets
from bignmf.models.snmf.standard import StandardNmf

Datasets.list_all()
data=Datasets.read("SimulatedX1")
k = 3
iter =100
trials = 50

model = StandardNmf(data,k)

# Runs the model
model.run(trials, iter, verbose=0)
print(model.error)

# Clusters the data
model.cluster_data()
print(model.h_cluster)

#Calculates the consensus matrices
model.calc_consensus_matrices() 
print(model.consensus_matrix_w)

2. Joint NMF

from bignmf.models.jnmf.integrative import IntegrativeJnmf
from bignmf.datasets.datasets import Datasets

Datasets.list_all()
data_dict = {}
data_dict["sim1"] = Datasets.read("SimulatedX1")
data_dict["sim2"] = Datasets.read("SimulatedX2")

k = 3
iter =100
trials = 50
lamb = 0.1

model = IntegrativeJnmf(data_dict, k, lamb)
# Runs the model
model.run(trials, iter, verbose=0)
print(model.error)

# Clusters the data
model.cluster_data()
print(model.h_cluster)

#Calculates the consensus matrices
model.calc_consensus_matrices() 
print(model.consensus_matrix_w)

Here is the extensive documentation for more details.

bignmf's Issues

ValueError: Operands are not aligned. Do `left, right = left.align(right, axis=1, copy=False)` before operating.

Running the first example in the README finished with this error:

ValueError: Operands are not aligned. Do left, right = left.align(right, axis=1, copy=False) before operating.

ValueError during run

When I run the code:

nmf_cm = StandardNmf(pd.DataFrame(data_tfidf), k = 20)
nmf_cm.run(iterations = 1000, trials = 20)

I get the following error:

ValueError
30 nmf_cm = StandardNmf(pd.DataFrame(data_tfidf), k = k)
---> 31 nmf_cm.run(iterations = 100, trials = 20, verbose = 0)
32 nmf_cm.calc_consensus_matrices()
33 nmf_cm.calc_cophenetic_correlation()

File ...\bignmf\models\snmf\snmf_base.py:52, in SnmfBase.run(self, trials, iterations, verbose)
50 print("\tTrial: %i completed with Error: %f " % (i, self.error))
51 # Normalization
---> 52 self.consensus_matrix_w = self.reorder_consensus_matrix(self.consensus_matrix_w / trials)
53 self.consensus_matrix_h = self.reorder_consensus_matrix(self.consensus_matrix_h / trials)
54 # Converting values to DataFrames

File ...\bignmf\models\nmf.py:69, in NmfBase.reorder_consensus_matrix(M)
67 M = pd.DataFrame(M)
68 Y = 1 - M
---> 69 Z = linkage(squareform(Y), method='average')
70 ivl = leaves_list(Z)
71 ivl = ivl[::-1]

File ...\scipy\spatial\distance.py:2354, in squareform(X, force, checks)
2352 raise ValueError('The matrix argument must be square.')
...
2436 'be zero.') % name)
2437 else:
2438 raise ValueError('Distance matrix diagonal must be zero.')

ValueError: Distance matrix 'X' diagonal must be zero.

My input is TFIDF data and there are no negative elements in it. Is there something wrong in the implementation of the function or am I making a mistake?