Package utils

Crawl recursively an MDP Node looking for arrays.

Return (dictionary, string), where the dictionary is: { attribute_name: (size_in_bytes, array_reference)} and string is a nice string representation of it.

Update __module__ attribute and remove old_modules from namespace

When classes are imported from implementation modules into the package exporting them, the __module__ attribute reflects the place of definition. Splitting the code into separate files (and thus modules) makes the implementation managable. Nevertheless, we do not want the implementation modules to be visible and delete their names from the package's namespace. This causes some problems: when looking at the exported classes and other objects, their __module__ attribute points to something non-importable, repr output and documentation do not show the module from which they are supposed to be imported. The documentation generators like epydoc and sphinx are also confused. To alleviate those problems, the __module__ attributes of all exported classes defined in a "private" module and then exported elsewhere are changed to the latter.

For each name in names, if <mname>.<name> is accessible, and if its __module__ attribute is equal to one of the names in old_modules, it is changed to "<mname>". In other words, all the __module__ attributes of objects exported from module <mname> are updated, iff they used to point to one of the "private" modules in old_modules.

This operation is performed not only for classes, but actually for all objects with the __module__ attribute, following the rules stated above. The operation is also performed recursively, not only for names in names, but also for methods, inner classes, and other attributes. This recursive invocation is necessary because all the problems affecting top-level exported classes also affect their attributes visible for the user, and especially documented functions.

If names is None, all public names in module <mname> (not starting with '_') are affected.

After the __module__ attributes are changed, "private" modules given in old_modules, except for the ones in keep_modules, are deleted from the namespace of <mname> module.

Return a 2D array containing a Gabor wavelet.

Input arguments:
size -- (height, width) (pixels)
alpha -- orientation (rad)
phi -- phase (rad)
freq -- frequency (cycles/deg)
sgm -- (sigma_x, sigma_y) standard deviation along the axis
       of the gaussian ellipse (pixel)
x0 -- (x,y) coordinates of the center of the wavelet (pixel)
      Default: None, meaning the center of the array
res -- spatial resolution (deg/pixel)
       Default: 1, so that 'freq' is measured in cycles/pixel
ampl -- constant multiplying the result
        Default: 1.

Return the list of dtypes corresponding to the set of typecodes defined in numpy.typecodes[typecodes_key]. E.g., get_dtypes('Float') = [dtype('f'), dtype('d'), dtype('g')].

If _safe is True (default), we remove large floating point types if the numerical backend does not support them.

Return node total byte-size using cPickle with protocol=2.

The byte-size is related to the memory needed by the node).

Return a string with the JS and HTML code for an image slideshow.

Note that the CSS code for the slideshow is not included, so you should
add SLIDESHOW_STYLE or a custom style to your CSS code.

filenames -- Sequence of the image filenames.
image_size -- Tuple (x,y) with the original image size, or enter
    a different size to force scaling.
title -- Optional slideshow title (for default None not title is shown).
section_ids -- List with the section id for each slide index. The id
        can be a string or a number. Default value None disables the
        section feature.

For additional keyword arguments see the ImageHTMLSlideShow class.

Use nearest neighbour resampling in Firefox 3.6+ and IE.

Webkit (Chrome, Safari) does not support this yet. (see http://code.google.com/p/chromium/issues/detail?id=1502)

Same as izip, except that for convenience non-iterables are repeated ad infinitum.

This is useful when trying to zip input data with respective labels
and allows for having a single label for all data, as well as for
havning a list of labels for each data vector.
Note that this will take strings as an iterable (of course), so
strings acting as a single value need to be wrapped in a repeat
statement of their own.

Thus,
>>> for zipped in izip_stretched([1, 2, 3], -1):
        print zipped
(1, -1)
(2, -1)
(3, -1)

is equivalent to
>>> for zipped in izip([1, 2, 3], [-1] * 3):
        print zipped
(1, -1)
(2, -1)
(3, -1)

Return alpha*(a*b) + beta*c.
a,b,c : matrices
alpha, beta: scalars
trans_a : 0 (a not transposed), 1 (a transposed),
          2 (a conjugate transposed)
trans_b : 0 (b not transposed), 1 (b transposed),
          2 (b conjugate transposed)

Dot product of two arrays.

For 2-D arrays it is equivalent to matrix multiplication, and for 1-D
arrays to inner product of vectors (without complex conjugation). For
N dimensions it is a sum product over the last axis of `a` and
the second-to-last of `b`::

    dot(a, b)[i,j,k,m] = sum(a[i,j,:] * b[k,:,m])

Parameters
----------
a : array_like
    First argument.
b : array_like
    Second argument.
out : ndarray, optional
    Output argument. This must have the exact kind that would be returned
    if it was not used. In particular, it must have the right type, must be
    C-contiguous, and its dtype must be the dtype that would be returned
    for `dot(a,b)`. This is a performance feature. Therefore, if these
    conditions are not met, an exception is raised, instead of attempting
    to be flexible.

Returns
-------
output : ndarray
    Returns the dot product of `a` and `b`.  If `a` and `b` are both
    scalars or both 1-D arrays then a scalar is returned; otherwise
    an array is returned.
    If `out` is given, then it is returned.

Raises
------
ValueError
    If the last dimension of `a` is not the same size as
    the second-to-last dimension of `b`.

See Also
--------
vdot : Complex-conjugating dot product.
tensordot : Sum products over arbitrary axes.
einsum : Einstein summation convention.
matmul : '@' operator as method with out parameter.

Examples
--------
>>> np.dot(3, 4)
12

Neither argument is complex-conjugated:

>>> np.dot([2j, 3j], [2j, 3j])
(-13+0j)

For 2-D arrays it is the matrix product:

>>> a = [[1, 0], [0, 1]]
>>> b = [[4, 1], [2, 2]]
>>> np.dot(a, b)
array([[4, 1],
       [2, 2]])

>>> a = np.arange(3*4*5*6).reshape((3,4,5,6))
>>> b = np.arange(3*4*5*6)[::-1].reshape((5,4,6,3))
>>> np.dot(a, b)[2,3,2,1,2,2]
499128
>>> sum(a[2,3,2,:] * b[1,2,:,2])
499128

Multiply a full matrix by a diagonal matrix. This function should always be faster than dot.

Input:

d -- 1D (N,) array (contains the diagonal elements) mtx -- 2D (N,N) array

Output:

mult_diag(d, mts, left=True) == dot(diag(d), mtx) mult_diag(d, mts, left=False) == dot(mtx, diag(d))

Takes a dictionary with lists as keys and returns all permutations of these list elements in new dicts.

This function is useful, when a method with several arguments shall be tested and all of the arguments can take several values.

The order is not defined, therefore the elements should be orthogonal to each other.

>>> for i in orthogonal_permutations({'a': [1,2,3], 'b': [4,5]}):
        print i
{'a': 1, 'b': 4}
{'a': 1, 'b': 5}
{'a': 2, 'b': 4}
{'a': 2, 'b': 5}
{'a': 3, 'b': 4}
{'a': 3, 'b': 5}

A fully configurable text-mode progress info box tailored to the
  command-line die-hards.

  To get a progress info box for your loops use it like this:

     >>> for i in progressinfo(sequence):
     ...     do_something(i)

 You can also use it with generators, files or any other iterable object,
 but in this case you have to specify the total length of the sequence:

     >>> for line in progressinfo(open_file, nlines):
     ...     do_something(line)

 If the number of iterations is not known in advance, you may prefer
 to iterate on the items directly. This can be useful for example if
 you are downloading a big file in a subprocess and want to monitor
 the progress. If the file to be downloaded is TOTAL bytes large and
 you are downloading it on local:
     >>> def done():
     ...     yield os.path.getsize(localfile)
     >>> for bytes in progressinfo(done(), -TOTAL)
     ...     time.sleep(1)
     ...     if download_process_has_finished():
     ...         break


Arguments:

sequence    - if it is a Python container object (list,
              dict, string, etc...) and it supports the
              __len__ method call, the length argument can
              be omitted. If it is an iterator (generators,
              file objects, etc...) the length argument must
              be specified.

Keyword arguments:

length     - length of the sequence. Automatically set
             if `sequence' has the __len__ method. If length is
             negative, iterate on items.

style      - If style == 'bar', display a progress bar. The
             default layout is:

             [===========60%===>.........]

             If style == 'timer', display a time elapsed / time
             remaining info box. The default layout is:

             23% [02:01:28] - [00:12:37]

             where fields have the following meaning:

             percent_done% [time_elapsed] - [time_remaining]

custom     - a dictionary for customizing the layout.
             Default layout for the 'bar' style:
              custom = { 'indent': '',
                         'width' : terminal_width - 1,
                         'position' : 'middle',
                         'delimiters' : '[]',
                         'char1' : '=',
                         'char2' : '>',
                         'char3' : '.' }


             Default layout for the 'timer' style:
              custom = { 'speed': 'mean',
                         'indent': '',
                         'position' : 'left',
                         'delimiters' : '[]',
                         'separator' : ' - ' }

             Description:
               speed = completion time estimation method, must be one of
                       ['mean', 'last']. 'mean' uses average speed, 'last'
                       uses last step speed.
               indent = string used for indenting the progress info box
               position = position of the percent done string,
                          must be one out of ['left', 'middle', 'right']


Note 1: by default sys.stdout is flushed each time a new box is drawn.
        If you need to rely on buffered stdout you'd better not use this
        (any?) progress info box.
Note 2: progressinfo slows down your loops. Always profile your scripts
        and check that you are not wasting 99% of the time in drawing
        the progress info box.

Rotate in-place data matrix (NxM) in the plane defined by the columns=[i,j] when observation are stored on rows. Observations are rotated counterclockwise. This corresponds to the following matrix-multiplication for each data-point (unchanged elements omitted):

[ cos(angle) -sin(angle) [ x_i ]

sin(angle) cos(angle) ] * [ x_j ]

If M=2, columns=[0,1].

Return False for each negative value, else True.

The value for 0 is specified with 'zero'.

Return a random symmetric (Hermitian) matrix.

If 'dim_or_eigv' is an integer N, return a NxN matrix, with eigenvalues

uniformly distributed on (-1,1).

If 'dim_or_eigv' is 1-D real array 'a', return a matrix whose

eigenvalues are 'a'.

Returns a key from a dictionary based on the weight that the value suggests. If 'normalize' is False, it is assumed the weights sum up to unity. Otherwise, the algorithm will take care of normalising.

Example: >>> d = {'a': 0.1, 'b': 0.5, 'c': 0.4} >>> weighted_choice(d) # draws 'b':'c':'a' with 5:4:1 probability

TODO: It might be good to either shuffle the order or explicitely specify it, before walking through the items, to minimise possible degeneration.