#acl All:read = NumPy/SciPy Cheat Sheet = This cheat sheet is a quick reference for !NumPy / !SciPy beginners and gives an overview about the most important commands and functions of !NumPy and !SciPy that you might need on solving the exercise sheets about Linear Algebra in Information Retrieval. It doesn't claim to be complete and will be extended continuously. If you think that some important thing is missing or if you find any errors, please let us know. <> == General == === What is NumPy? === A library that allows to work with arrays and matrices in Python. === What is SciPy? === Another library built upon !NumPy that provides advanced Linear Algebra stuff. == Install == The routine to install !NumPy and !SciPy depends on your operating system. === Linux (Ubuntu, Debian) === {{{ apt-get install python-numpy python-scipy }}} === Other systems (Windows, Mac, etc.) === For all other systems (Windows, Mac, etc.) see the instructions given on the offical [[https://scipy.org/install.html|SciPy website]]. ------ == Matrix construction == {{{#!html We distinguish between dense matrices and sparse matrices (Note: This color code will be used conistently throughout this cheat sheet). }}} Dense matrices store every entry in the matrix, while sparse matrices only store the non-zero entries (together with their row and column index). Dense matrices are more feature-rich, but may consume more memory space than sparse matrices (in particular if most of the entries in a matrix are zero). === Dense matrices === In !NumPy, there are two concepts of dense matrices: '''matrices''' and '''arrays'''. Matrices are strictly 2-dimensional, while arrays are n-dimensional (the term ''array'' is a bit misleading here). Construct a matrix: {{{#!html
Dense
numpy.matrix(arg, dtype=None)

arg:
   The data to construct the matrix from, given as
     * a standard Python array; or
     * a string with columns separated by commas or spaces and rows separated by semicolons.
dtype (str, optional):
   The type of the entries in the matrix (e.g., 'integer', 'float', 'string', etc.).

----------
Examples:

>>> numpy.matrix("1 2; 3 4")
[[1 2]
 [3 4]]

>>> numpy.matrix([[1, 2], [3, 4]], dtype='float')
[[1.0 2.0]
 [3.0 4.0]]
}}} [[https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matrix.html|numpy.matrix]] <
><
> Construct an array: {{{#!html
Dense
numpy.array(arg, dtype=None, ndmin=0)

arg:
   The data to construct the matrix from, given as
      * a standard array; or
      * a function that returns an array.
dtype (str, optional):
   The type of the entries in the matrix ('integer', 'float', 'string', etc.).
ndmin (int, optional):
   The minimum number of dimensions that the array should have.

----------
Examples:

>>> numpy.array([[1, 2], [3, 4]])
[[1 2]
 [3 4]]

>>> numpy.array([[1, 2], [3, 4]], dtype='float')
[[1.0 2.0]
 [3.0 4.0]]

>>> numpy.array([[1, 2], [3, 4]], ndmin=3)
[[[1 2]
  [3 4]]]
}}} [[https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.array.html|numpy.array]] <
><
> === Sparse matrices === There are two principle concepts of sparse matrices: * ''Compressed Sparse Row'' matrix (CSR matrix): entries are stored row by row (sorted by row index first) * ''Compressed Sparse Column'' matrix (CSC matrix): entries are stored column by column (sorted by column index first) Construct a CSR/CSC matrix: {{{#!html
scipy.sparse.csr_matrix(arg, shape=None, dtype=None)
scipy.sparse.csc_matrix(arg, shape=None, dtype=None)

arg:
   The data to create the CSR matrix from, given as 
     * a dense matrix; or
     * another sparse matrix; or
     * a tuple (m, n), to construct an empty matrix with shape (n, m); or
     * a tuple (data, (rows, cols), to construct a matrix A where A[rows[k], cols[k]] = data[k]; or
     * a tuple (data, indices, indptr)
shape (int or sequence of ints):
   The dimensions of the matrix to create.
dtype (str, optional):
   The type of the entries in the matrix ('integer', 'float', 'string', etc.).

----------
Examples:

>>> scipy.sparse.csr_matrix([[1, 2, 3], [0, 0, 1], [0, 1, 3]])
[[1 2 3]
 [0 0 1]
 [0 1 3]]  # (transformed to a dense matrix for visualization).

>>> scipy.sparse.csc_matrix([[1, 2, 3], [0, 0, 1], [0, 1, 3]])
[[1 2 3]
 [0 0 1]
 [0 1 3]]  # (transformed to a dense matrix for visualization).

>>> values = [1, 2, 3]
>>> rows   = [0, 0, 1]
>>> cols   = [0, 1, 3]
>>> scipy.sparse.csr_matrix((values, (rows, columns)), shape=[5, 5], dtype=int)
[[1 1 0 0]
 [0 0 0 3]
 [0 0 0 0]
 [0 0 0 0]]  # (transformed to a dense matrix for visualization).

>>> values = [1, 2, 3]
>>> rows   = [0, 0, 1]
>>> cols   = [0, 1, 3]
>>> scipy.sparse.csc_matrix((values, (rows, columns)), shape=[5, 5], dtype=int)
[[1 1 0 0]
 [0 0 0 3]
 [0 0 0 0]
 [0 0 0 0]]  # (transformed to a dense matrix for visualization).
}}} [[https://docs.scipy.org/doc/scipy-0.14.0/reference/generated/scipy.sparse.csr_matrix.html|scipy.sparse.csr_matrix]]<
> [[https://docs.scipy.org/doc/scipy-0.14.0/reference/generated/scipy.sparse.csc_matrix.html|scipy.sparse.csc_matrix]]<
><
> === Special matrices === There are some utility functions to create special matrices/arrays: (1) Construct an '''empty array''', without initializing the entries (an array with random entries): {{{#!html
numpy.empty(shape, dtype=float)

shape (int or sequence of ints):
   The dimensions of the array to create.
dtype (str, optional):
   The type of the entries in the matrix ('integer', 'float', 'string', etc.).

----------
Examples:

>>> numpy.empty(3)
[6.95052181e-310 1.74512682e-316 1.58101007e-322]

>>> numpy.empty([3, 2], dtype='int')
[[140045355821992 140045355821992]
 [140045136216840 140045136244784]
 [140045125643544 140045153116544]]
}}} [[https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.empty.html|numpy.empty]] <
><
> (2) Construct an '''array filled with zeros''': {{{#!html
numpy.zeros(shape, dtype=float)

shape (int or sequence of ints):
   The dimensions of the array to create.
dtype (str, optional):
   The type of the entries in the matrix ('integer', 'float', 'string', etc.).

----------
Examples:

>>> numpy.zeros(3)
[0.0, 0.0, 0.0]

>>> numpy.zeros([3, 2], dtype='int')
[[0 0]
 [0 0]
 [0 0]]
}}} [[https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.zeros.html|numpy.zeros]] <
><
> (3) Construct an '''array filled with ones''': {{{#!html
numpy.ones(shape, dtype=float)

shape (int or sequence of ints):
   The dimensions of the array to create.
dtype (str, optional):
   The type of the entries in the matrix ('integer', 'float', 'string', etc.).

----------
Examples:

>>> numpy.ones(3)
[1.0, 1.0, 1.0]

>>> numpy.ones([3, 2], dtype='int')
[[1 1]
 [1 1]
 [1 1]]
}}} [[https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.ones.html|numpy.ones]] <
><
> (4) Construct a '''diagonal array''', a (usually square) array in which all entries are 0, except on the main diagonal: {{{#!html
numpy.diag(arg, k=0)

arg (1-dim array):
   The entries of the diagonal.
k (int, optional):
   The diagonal in question. Use k > 0 for diagonals above the main diagonal, and k < 0 for diagonals below the main diagonal. 

----------
Examples:

>>> numpy.diag([1, 2, 3])
[[1 0 0]
 [0 2 0]
 [0 0 3]]

>>> numpy.diag([1, 2, 3], k=1)
[[0 1 0 0]
 [0 0 2 0]
 [0 0 0 3]
 [0 0 0 0]]

>>> numpy.diag([1, 2, 3], k=-1)
[[0 0 0 0]
 [1 0 0 0]
 [0 2 0 0]
 [0 0 3 0]]
}}} [[https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.diag.html|numpy.diag]] {{{#!html
scipy.sparse.diags(diagonals, offsets=0, dtype=None)

diagonals (sequence of arrays):
   The entries of the matrix diagonals.
offsets (sequence of ints or int, optional):
   The diagonals in question. k = 0 is the main diagonal; k > 0 is the k-th upper diagonal; k < 0 is the k-th lower diagonal
dtype (str, optional):
   The type of the entries in the matrix ('integer', 'float', 'string', etc.).

----------
Examples:

>>> scipy.sparse.diags([1, 2, 3])
[[1.0 0.0 0.0]
 [0.0 2.0 0.0]
 [0.0 0.0 3.0]]  # (transformed to a dense matrix for visualization).

>>> scipy.sparse.diags([[1, 2, 3], [4, 5, 6]], offsets=[0, 1])
[[1.0 4.0 0.0]
 [0.0 2.0 5.0]
 [0.0 0.0 3.0]]  # (transformed to a dense matrix for visualization).
}}} [[https://docs.scipy.org/doc/scipy/reference/generated/scipy.sparse.diags.html|scipy.sparse.diags]] <
><
> (5) Construct an '''identity array''', a square array in which all entries on the main diagonal are 1 and all other entries are 0: {{{#!html
numpy.identity(n, dtype=float)

n (int):
   The dimension of the array to create (the output is a n x n array).
dtype (str, optional):
   The type of the entries in the matrix ('integer', 'float', 'string', etc.).

----------
Examples:

>>> numpy.identity(3)
[[1.0, 0.0, 0.0]
 [0.0, 1.0, 0.0]
 [0.0, 0.0, 1.0]]

>>> numpy.identity(3, dtype=int)
[[1, 0, 0]
 [0, 1, 0]
 [0, 0, 1]]
}}} [[https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.identity.html|numpy.identity]] {{{#!html
scipy.sparse.identity(n, dtype=float, format="csr")

n (int):
   The dimension of the array to create.
dtype (str, optional):
   The type of the entries in the matrix ('integer', 'float', 'string', etc.).
format (str, optional)
   The sparse format of the array, e.g. "csr" or "csc".

----------
Examples:

>>> scipy.sparse.identity(3)
[[1.0, 0.0, 0.0]
 [0.0, 1.0, 0.0]
 [0.0, 0.0, 1.0]]  # (transformed to a dense matrix for visualization).

>>> scipy.sparse.identity(3, dtype=int)
[[1, 0, 0]
 [0, 1, 0]
 [0, 0, 1]]  # (transformed to a dense matrix for visualization).
}}} [[https://docs.scipy.org/doc/scipy-0.14.0/reference/generated/scipy.sparse.identity.html|scipy.sparse.identity]]<
><
> (6) Construct an '''triangular array''', a square array in which all entries below (upper triangle) or above (lower triangle) the main diagonal are zero: {{{#!html
numpy.triu(arg, k=0)  # Zero entries in the upper triangle of an array.
numpy.tril(arg, k=0)  # Zero entries in the lower triangle of an array.

arg (array):
   The original array.
k (int, optional):
   Diagonal above which to zero entries. k = 0 is the main diagonal, k < 0 is below it and k > 0 is above.

----------
Examples:

>>> numpy.triu([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
[[1 2 3]
 [0 5 6]
 [0 0 9]]

>>> numpy.triu([[1, 2, 3], [4, 5, 6], [7, 8, 9]], k=1)
[[0 2 3]
 [0 0 6]
 [0 0 0]]

>>> numpy.tril([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
[[1 0 0]
 [4 5 0]
 [7 8 9]]

>>> numpy.tril([[1, 2, 3], [4, 5, 6], [7, 8, 9]], k=-1)
[[0 0 0]
 [4 0 0]
 [7 8 0]]
}}} [[https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.triu.html|numpy.triu]] <
> [[https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.tril.html|numpy.tril]] <
><
> {{{#!html
scipy.sparse.triu(arg, k=0, format="csr")  # Zero entries in the upper triangle of an array.
scipy.sparse.tril(arg, k=0, format="csr")  # Zero entries in the lower triangle of an array.

arg (array):
   The original array.
k (int, optional):
   Diagonal above which to zero entries. k = 0 is the main diagonal, k < 0 is below it and k > 0 is above.
format (str, optional)
   The sparse format of the array, e.g. "csr" or "csc".

----------
Examples:

>>> scipy.sparse.triu([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
[[1 2 3]
 [0 5 6]
 [0 0 9]]  # (transformed to a dense matrix for visualization).

>>> scipy.sparse.triu([[1, 2, 3], [4, 5, 6], [7, 8, 9]], k=1)
[[0 2 3]
 [0 0 6]
 [0 0 0]]  # (transformed to a dense matrix for visualization).

>>> scipy.sparse.tril([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
[[1 0 0]
 [4 5 0]
 [7 8 9]]  # (transformed to a dense matrix for visualization).

>>> scipy.sparse.tril([[1, 2, 3], [4, 5, 6], [7, 8, 9]], k=-1)
[[0 0 0]
 [4 0 0]
 [7 8 0]]  # (transformed to a dense matrix for visualization).
}}} [[https://docs.scipy.org/doc/scipy-0.14.0/reference/generated/scipy.sparse.tril.html|scipy.sparse.triu]] <
> [[https://docs.scipy.org/doc/scipy-0.14.0/reference/generated/scipy.sparse.triu.html|scipy.sparse.tril]] <
><
> ------ == Accessing elements == TODO: crazy element access magic, single elements, entire rows, sub-matrices ------ == Matrix operations == TODO ------ == Useful methods == TODO