Sequence.__setitem__() - Code Metrics - Inspection of "Make the Sequence brick implement the Sequence ADT..." - mila-udem/blocks - Measure and Improve Code Quality continuously with Scrutinizer

Completed

Pull Request — master (#1163)

by David

created 2016-11-11 07:15 UTC

Sequence.setitem() A

↳ Parent: Project

Complexity

Conditions

Size

Total Lines

Duplication

Lines	0
Ratio	0 %

Importance

Changes

Metric	Value
cc	1
dl	0
loc	3
rs	10
c	0
b	0
f	0

"""Bricks that compose together other bricks in linear sequences."""
from collections import MutableSequence
import copy
from toolz import interleave
from picklable_itertools.extras import equizip

from ..utils import pack
from .base import Brick, application, lazy
from .interfaces import Feedforward, Initializable
from .simple import Linear


class Sequence(Brick, MutableSequence):
    """A sequence of bricks.

    This brick applies a sequence of bricks, assuming that their in- and
    outputs are compatible.

    Parameters
    ----------
    application_methods : list
        List of :class:`.BoundApplication` to apply

    """
    def __init__(self, application_methods, **kwargs):
        self.application_methods = list(application_methods)

        seen = set()
        children = [app.brick for app in application_methods
                    if not (app.brick in seen or seen.add(app.brick))]
        kwargs.setdefault('children', []).extend(children)
        super(Sequence, self).__init__(**kwargs)

    @application
    def apply(self, *args):
        child_input = args
        for application_method in self.application_methods:
            output = application_method(*pack(child_input))
            child_input = output
        return output

    @apply.property('inputs')
    def apply_inputs(self):
        return self.application_methods[0].inputs

    @apply.property('outputs')
    def apply_outputs(self):
        return self.application_methods[-1].outputs

    def __delitem__(self, index):
        application_method = self.application_methods[index]
        brick = application_method.brick
        try:
            del self.application_methods[index]
        finally:
            apps_with_brick = [a for a in self.application_methods
                               if a.brick is brick]
            if len(apps_with_brick) == 0:
                self.children.remove(brick)

    def __getitem__(self, index):
        return self.application_methods[index]

    def __setitem__(self, index, new_application):
        del self[index]
        self.insert(index, new_application)

    def insert(self, index, new_application):
        if new_application.brick not in self.children:
            self.children.append(new_application.brick)
        self.application_methods.insert(index, new_application)

    def __len__(self):
        return len(self.application_methods)


class FeedforwardSequence(Sequence, Feedforward):
    """A sequence where the first and last bricks are feedforward.

    Parameters
    ----------
    application_methods : list
        List of :class:`.BoundApplication` to apply. The first and last
        application method should belong to a :class:`Feedforward` brick.

    """
    @property
    def input_dim(self):
        return self.children[0].input_dim

    @input_dim.setter
    def input_dim(self, value):
        self.children[0].input_dim = value

    @property
    def output_dim(self):
        return self.children[-1].output_dim

    @output_dim.setter
    def output_dim(self, value):
        self.children[-1].output_dim = value


class MLP(FeedforwardSequence, Initializable):
    """A simple multi-layer perceptron.

    Parameters
    ----------
    activations : list of :class:`.Brick`, :class:`.BoundApplication`,
                  or ``None``
        A list of activations to apply after each linear transformation.
        Give ``None`` to not apply any activation. It is assumed that the
        application method to use is ``apply``. Required for
        :meth:`__init__`.
    dims : list of ints
        A list of input dimensions, as well as the output dimension of the
        last layer. Required for :meth:`~.Brick.allocate`.
    prototype : :class:`.Brick`, optional
        The transformation prototype. A copy will be created for every
        activation. If not provided, an instance of :class:`~simple.Linear`
        will be used.

    Notes
    -----
    See :class:`Initializable` for initialization parameters.

    Note that the ``weights_init``, ``biases_init`` (as well as
    ``use_bias`` if set to a value other than the default of ``None``)
    configurations will overwrite those of the layers each time the
    :class:`MLP` is re-initialized. For more fine-grained control, push the
    configuration to the child layers manually before initialization.

    >>> from blocks.bricks import Tanh
    >>> from blocks.initialization import IsotropicGaussian, Constant
    >>> mlp = MLP(activations=[Tanh(), None], dims=[30, 20, 10],
    ...           weights_init=IsotropicGaussian(),
    ...           biases_init=Constant(1))
    >>> mlp.push_initialization_config()  # Configure children
    >>> mlp.children[0].weights_init = IsotropicGaussian(0.1)
    >>> mlp.initialize()

    """
    @lazy(allocation=['dims'])
    def __init__(self, activations, dims, prototype=None, **kwargs):
        self.activations = activations
        self.prototype = Linear() if prototype is None else prototype
        self.linear_transformations = []
        for i in range(len(activations)):
            linear = copy.deepcopy(self.prototype)
            name = self.prototype.__class__.__name__.lower()
            linear.name = '{}_{}'.format(name, i)
            self.linear_transformations.append(linear)
        # Interleave the transformations and activations
        application_methods = []
        for entity in interleave([self.linear_transformations, activations]):
            if entity is None:
                continue
            if isinstance(entity, Brick):
                application_methods.append(entity.apply)
            else:
                application_methods.append(entity)
        if not dims:
            dims = [None] * (len(activations) + 1)
        self.dims = dims
        super(MLP, self).__init__(application_methods, **kwargs)

    @property
    def input_dim(self):
        return self.dims[0]

    @input_dim.setter
    def input_dim(self, value):

        self.dims[0] = value

    @property
    def output_dim(self):
        return self.dims[-1]

    @output_dim.setter
    def output_dim(self, value):

        self.dims[-1] = value

    def _push_allocation_config(self):
        if not len(self.dims) - 1 == len(self.linear_transformations):
            raise ValueError
        for input_dim, output_dim, layer in \
                equizip(self.dims[:-1], self.dims[1:],
                        self.linear_transformations):
            layer.input_dim = input_dim
            layer.output_dim = output_dim
            if getattr(self, 'use_bias', None) is not None:
                layer.use_bias = self.use_bias


1			"""Bricks that compose together other bricks in linear sequences."""
2			from collections import MutableSequence
3			import copy
4			from toolz import interleave
5			from picklable_itertools.extras import equizip
6
7			from ..utils import pack
8			from .base import Brick, application, lazy
9			from .interfaces import Feedforward, Initializable
10			from .simple import Linear
11
12
13			class Sequence(Brick, MutableSequence):
14			"""A sequence of bricks.
15
16			This brick applies a sequence of bricks, assuming that their in- and
17			outputs are compatible.
18
19			Parameters
20			----------
21			application_methods : list
22			List of :class:`.BoundApplication` to apply
23
24			"""
25			def __init__(self, application_methods, **kwargs):
26			self.application_methods = list(application_methods)
27
28			seen = set()
29			children = [app.brick for app in application_methods
30			if not (app.brick in seen or seen.add(app.brick))]
31			kwargs.setdefault('children', []).extend(children)
32			super(Sequence, self).__init__(**kwargs)
33
34			@application
35			def apply(self, *args):
36			child_input = args
37			for application_method in self.application_methods:
38			output = application_method(*pack(child_input))
39			child_input = output
40			return output
41
42			@apply.property('inputs')
43			def apply_inputs(self):
44			return self.application_methods[0].inputs
45
46			@apply.property('outputs')
47			def apply_outputs(self):
48			return self.application_methods[-1].outputs
49
50			def __delitem__(self, index):
51			application_method = self.application_methods[index]
52			brick = application_method.brick
53			try:
54			del self.application_methods[index]
55			finally:
56			apps_with_brick = [a for a in self.application_methods
57			if a.brick is brick]
58			if len(apps_with_brick) == 0:
59			self.children.remove(brick)
60
61			def __getitem__(self, index):
62			return self.application_methods[index]
63
64			def __setitem__(self, index, new_application):
65			del self[index]
66			self.insert(index, new_application)
67
68			def insert(self, index, new_application):
69			if new_application.brick not in self.children:
70			self.children.append(new_application.brick)
71			self.application_methods.insert(index, new_application)
72
73			def __len__(self):
74			return len(self.application_methods)
75
76
77			class FeedforwardSequence(Sequence, Feedforward):
78			"""A sequence where the first and last bricks are feedforward.
79
80			Parameters
81			----------
82			application_methods : list
83			List of :class:`.BoundApplication` to apply. The first and last
84			application method should belong to a :class:`Feedforward` brick.
85
86			"""
87			@property
88			def input_dim(self):
89			return self.children[0].input_dim
90
91			@input_dim.setter
92			def input_dim(self, value):
93			self.children[0].input_dim = value
94
95			@property
96			def output_dim(self):
97			return self.children[-1].output_dim
98
99			@output_dim.setter
100			def output_dim(self, value):
101			self.children[-1].output_dim = value
102
103
104			class MLP(FeedforwardSequence, Initializable):
105			"""A simple multi-layer perceptron.
106
107			Parameters
108			----------
109			activations : list of :class:`.Brick`, :class:`.BoundApplication`,
110			or ``None``
111			A list of activations to apply after each linear transformation.
112			Give ``None`` to not apply any activation. It is assumed that the
113			application method to use is ``apply``. Required for
114			:meth:`__init__`.
115			dims : list of ints
116			A list of input dimensions, as well as the output dimension of the
117			last layer. Required for :meth:`~.Brick.allocate`.
118			prototype : :class:`.Brick`, optional
119			The transformation prototype. A copy will be created for every
120			activation. If not provided, an instance of :class:`~simple.Linear`
121			will be used.
122
123			Notes
124			-----
125			See :class:`Initializable` for initialization parameters.
126
127			Note that the ``weights_init``, ``biases_init`` (as well as
128			``use_bias`` if set to a value other than the default of ``None``)
129			configurations will overwrite those of the layers each time the
130			:class:`MLP` is re-initialized. For more fine-grained control, push the
131			configuration to the child layers manually before initialization.
132
133			>>> from blocks.bricks import Tanh
134			>>> from blocks.initialization import IsotropicGaussian, Constant
135			>>> mlp = MLP(activations=[Tanh(), None], dims=[30, 20, 10],
136			... weights_init=IsotropicGaussian(),
137			... biases_init=Constant(1))
138			>>> mlp.push_initialization_config() # Configure children
139			>>> mlp.children[0].weights_init = IsotropicGaussian(0.1)
140			>>> mlp.initialize()
141
142			"""
143			@lazy(allocation=['dims'])
144			def __init__(self, activations, dims, prototype=None, **kwargs):
145			self.activations = activations
146			self.prototype = Linear() if prototype is None else prototype
147			self.linear_transformations = []
148			for i in range(len(activations)):
149			linear = copy.deepcopy(self.prototype)
150			name = self.prototype.__class__.__name__.lower()
151			linear.name = '{}_{}'.format(name, i)
152			self.linear_transformations.append(linear)
153			# Interleave the transformations and activations
154			application_methods = []
155			for entity in interleave([self.linear_transformations, activations]):
156			if entity is None:
157			continue
158			if isinstance(entity, Brick):
159			application_methods.append(entity.apply)
160			else:
161			application_methods.append(entity)
162			if not dims:
163			dims = [None] * (len(activations) + 1)
164			self.dims = dims
165			super(MLP, self).__init__(application_methods, **kwargs)
166
167			@property
168			def input_dim(self):
169			return self.dims[0]
170
171			@input_dim.setter
172			def input_dim(self, value):
			0 ignored issues – show Bug introduced 2016-03-10 17:46 UTC by Report Bug Copy Issue Report Arguments number differs from overridden 'input_dim' method Loading history...
173			self.dims[0] = value
174
175			@property
176			def output_dim(self):
177			return self.dims[-1]
178
179			@output_dim.setter
180			def output_dim(self, value):
			0 ignored issues – show Bug introduced 2016-03-10 17:46 UTC by Report Bug Copy Issue Report Arguments number differs from overridden 'output_dim' method Loading history...
181			self.dims[-1] = value
182
183			def _push_allocation_config(self):
184			if not len(self.dims) - 1 == len(self.linear_transformations):
185			raise ValueError
186			for input_dim, output_dim, layer in \
187			equizip(self.dims[:-1], self.dims[1:],
188			self.linear_transformations):
189			layer.input_dim = input_dim
190			layer.output_dim = output_dim
191			if getattr(self, 'use_bias', None) is not None:
192			layer.use_bias = self.use_bias
193

mila-udem / blocks

Pull Request — master (#1163)

Sequence.__setitem__() A

Complexity

Size

Duplication

Importance

Duplication Side-by-Side

Filter issues like

Sequence.setitem() A