Class: SVDMatrix

Inherits:
Matrix
  • Object
show all
Defined in:
lib/svd_matrix.rb

Instance Method Summary collapse

Constructor Details

#initialize(m, n) ⇒ SVDMatrix

Create a new SVD Matrix with m rows, n columns



8
9
10
11
 
# File 'lib/svd_matrix.rb', line 8

def initialize(m, n)
  @rows = Array.new(m)
  m.times {|i| @rows[i] = Array.new(n)}
end

Instance Method Details

#[]=(i, j, val) ⇒ Object

Set the value of the cell i, j



14
15
16
 
# File 'lib/svd_matrix.rb', line 14

def []=(i, j, val)
  @rows[i][j] = val
end

#decompose(reduce_dimensions_to = nil) ⇒ Object

Perform SVD and decompose the matrix into three matrices: U, W, and V. You can choose to reduce the dimensionality of the data by setting a number of diagonal cells to 0. For example, reduce_dimentions_to = 2 will set a 4x4 W matrix into:

NUM, 0, 0, 0
0, NUM, 0, 0
0, 0, 0, 0
0, 0, 0, 0 


37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
 
# File 'lib/svd_matrix.rb', line 37

def decompose(reduce_dimensions_to = nil)
  input_array = []
  @rows.each {|row| input_array += row}
  u_array, w_array, v_array = SVD.decompose(input_array, row_size, column_size)
  
  # recompose U matrix
  u = SVDMatrix.new(row_size, reduce_dimensions_to || column_size)
  row_size.times {|i| u.set_row(i, u_array.slice!(0, column_size)[0...reduce_dimensions_to])}
  
  # recompose V matric
  v = SVDMatrix.new(column_size, reduce_dimensions_to || column_size)
  column_size.times {|i| v.set_row(i, v_array.slice!(0, column_size)[0...reduce_dimensions_to])}
  
  # diagonalise W array as a matrix
  if reduce_dimensions_to
    w_array = w_array[0...reduce_dimensions_to]
  end
  w = Matrix.diagonal(*w_array)
  
  [u, w, v]
end

#inspect ⇒ Object

Nicely formatted inspect string for the matrix



24
25
26
 
# File 'lib/svd_matrix.rb', line 24

def inspect
  @rows.collect {|row| "#{row.inspect}\n"}
end

#reduce_dimensions(dimensions = 2) ⇒ Object

Reduce the number of dimensions of the data to dimensions. Returns a back a recombined matrix (conceptually the original matrix dimensionally reduced). For example Latent Semantic Analysis uses 2 dimensions, and commonly tf-idf cell data. The recombined matrix, and the 3 decomposed matrices are returned.



65
66
67
68
 
# File 'lib/svd_matrix.rb', line 65

def reduce_dimensions(dimensions = 2)
  u, w, v = self.decompose(dimensions)
  [(u * w * v.transpose), u, w, v]
end

#set_row(i, row) ⇒ Object

Set the value of a row to an array



19
20
21
 
# File 'lib/svd_matrix.rb', line 19

def set_row(i, row)
  @rows[i] = row
end