API Reference

• Optimizees
  • Optimizee Base Module
  • Implemented examples
    • Optimizee for simple functions
      • FunctionGeneratorOptimizee
      • l2l.optimizees.functions.tools
    • Optimizee for MNIST
      • MNISTOptimizee
      • MNISTOptimizeeParameters
• Optimizers
  • Optimizer Base Module
  • Implemented Examples
    • Optimizer using Cross Entropy
      • CrossEntropyOptimizer
      • CrossEntropyParameters
      • Distributions
    • Optimizer using FACE
      • FACEOptimizer
      • FACEParameters
    • Optimizer using Gradient Descent
      • GradientDescentOptimizer
      • ClassicGDParameters
      • StochasticGDParameters
      • AdamParameters
      • RMSPropParameters
    • Optimizer using Grid Search
      • GridSearchOptimizer
      • GridSearchParameters
    • Optimizer using Evolutionary Algorithm
      • GeneticAlgorithmOptimizer
      • GeneticAlgorithmParameters
    • Optimizer using Simulated Annealing
      • SimulatedAnnealingOptimizer
      • SimulatedAnnealingParameters
    • Optimizer using Evolution Strategies
      • EvolutionStrategiesOptimizer
      • EvolutionStrategiesParameters
    • Optimizer using Natural Evolution Strategies
      • NaturalEvolutionStrategiesOptimizer
      • NaturalEvolutionStrategiesParameters
• Simulation control
  • Trajectory
  • Individual
  • ParamterGroup
  • Experiment
• Logging Tools
  • Module Functions

Other module functions

class l2l.sdict(obj)

Bases: l2l.sdictm

Immutable version of sdictm

update(**kwargs)

Update the dictionary with the values given in the function (only goes one level down) :param kwargs: :return:

apply(fn)

Recursively apply fn on all leaf key, value pairs :param fn: :return:

class l2l.sdictm(obj)

Bases: object

A dictionary which allows accessing it’s values using a dot notation. i.e. d[‘a’] can be accessed as d.a Mutable version

get(key, default_value)

keys()

todict()

copy()

Return a copy of the class. The copy is deep. :return:

update(**kwargs)

Update the dictionary with the values given in the function (only goes one level down) :param kwargs: :return:

apply(fn)

Recursively apply fn on all leaf key, value pairs :param fn: :return:

frozen()

class l2l.DictEntryType(value)

Bases: enum.Enum

An enumeration.

Sequence = 0

Scalar = 1

l2l.list_to_dict(input_list, dict_spec)

This function converts a list back into a dict.

Parameters

• input_list (list) – This is the list to be converted into a dict.

• dict_spec (tuple) – This is the dict specification that conveys information regarding the dict to be converted to. See dict_to_list() for information about the dict specification.

Returns

A Dictionary representing the list. Note that for valid behaviour, the list should have been generated by dict_to_list(). Moreover, This operation is not a perfect inverse of list_to_dict(). While the types of individual elements are preserved, the types of the iterables inside which they reside are not. list_to_dict() creates all iterables as numpy arrays.

l2l.dict_to_list(input_dict, get_dict_spec=False)

This function converts the given dictionary into a list.

Parameters

• input_dict (dict) – The dictionary to be converted into a list. It is assumed that each key in the dictionary is a string and each value is either a scalar numerical value or an iterable.

• get_dict_spec (bool) –

  This is a flag that indicates whether one wants the dictionary specification tuple to be returned. It is disabled by default

  Dict-Specification:

  The Dict-Specification tuple is required to convert back from the list to dictionary.

  It is a tuple of tuples, one tuple for each key of the dictionary. Each tuple is of the following form:

  (key_name, value_type, value_length)

  key_type may be one of DictEntryType.Sequence or DictEntryType.Scalar. In case the object is a scalar, value_length is 1.

Returns

The list representing the contents of the dict. If get_dict_spec is True, the dict specification is also returned. Note that the keys are always sorted ascendingly by name

l2l.stdout_redirected(filename)

This context manager causes all writes to stdout (whether within python or its subprocesses) to be redirected to the filename specified. For usage, look at example below:

import os

with stdout_redirected(filename):
    print("from Python")
    os.system("echo non-Python applications are also supported")

inspired from the article http://eli.thegreenplace.net/2015/redirecting-all-kinds-of-stdout-in-python/

Parameters

filename – The filename (NOT file stream object, this is to ensure that the stream is always a valid file object) to which the stdout is to be redirected

l2l.stdout_discarded()

This context manager causes all writes to stdout (whether within python or its subprocesses) to be redirected to os.devnull, thereby effectively discarding the output. For usage look at example below:

import os

with stdout_discarded():
    print("from Python")
    os.system("echo non-Python applications are also supported")

l2l.convert_dict_to_numpy(input_dict)

This function takes a dictionary as input and converts the values to numpy data types. This is useful when importing parameters from config files.

NOTE: Only the following data types are converted:

1. Scalars integer or float. They are converted to np.int64 and numpy.float64 respectively

2. Iterables. All elements of the iterable are converted to a numpy array via np.array(...)

3. Dictionaries. Dictionaries are recursively converted

Parameters

input_dict – This is the dictionary to convert

Returns

The converted output dictionary