"""
Parameter set handling
:copyright: Copyright 2006-2020 by the PyNN team, see AUTHORS.
:license: CeCILL, see LICENSE for details.
"""
try: # Python 2
basestring
long
except NameError: # Python 3
basestring = str
long = int
import numpy
import collections
from pyNN.core import is_listlike
from pyNN import errors
from pyNN.random import RandomDistribution, NativeRNG
from lazyarray import larray, partial_shape
import numpy as np
[docs]class LazyArray(larray):
"""
Optimises storage of arrays in various ways:
- stores only a single value if all the values in the array are the same
- if the array is created from a :class:`~pyNN.random.RandomDistribution`
or a function `f(i,j)`, then elements are only evaluated when they are
accessed. Any operations performed on the array are also queued up to
be executed on access.
The main intention of the latter is to save memory for very large arrays by
accessing them one row or column at a time: the entire array need never be
in memory.
Arguments:
`value`:
may be an int, long, float, bool, NumPy array, iterator, generator
or a function, `f(i)` or `f(i,j)`, depending on the dimensions of
the array. `f(i,j)` should return a single number when `i` and `j`
are integers, and a 1D array when either `i` or `j` or both is a
NumPy array (in the latter case the two arrays must have equal
lengths).
`shape`:
a tuple giving the shape of the array, or `None`
`dtype`:
the NumPy `dtype`.
"""
# most of the implementation moved to external lazyarray package
# the plan is ultimately to move everything to lazyarray
def __init__(self, value, shape=None, dtype=None):
if isinstance(value, basestring):
errmsg = "Value should be a string expressing a function of d. "
try:
value = eval("lambda d: %s" % value)
except SyntaxError:
raise errors.InvalidParameterValueError(errmsg + "Incorrect syntax.")
try:
value(0.0)
except NameError as err:
raise errors.InvalidParameterValueError(errmsg + str(err))
super(LazyArray, self).__init__(value, shape, dtype)
def __setitem__(self, addr, new_value):
self.check_bounds(addr)
if (self.is_homogeneous
and isinstance(new_value, (int, long, float, bool))
and self.evaluate(simplify=True) == new_value):
pass
else:
self.base_value = self.evaluate()
self.base_value[addr] = new_value
self.operations = []
[docs] def by_column(self, mask=None):
"""
Iterate over the columns of the array. Columns will be yielded either
as a 1D array or as a single value (for a flat array).
`mask`: either `None` or a boolean array indicating which columns should be included.
"""
column_indices = numpy.arange(self.ncols)
if mask is not None:
assert len(mask) == self.ncols
column_indices = column_indices[mask]
if isinstance(self.base_value, RandomDistribution) and self.base_value.rng.parallel_safe:
if mask is None:
for j in column_indices:
yield self._partially_evaluate((slice(None), j), simplify=True)
else:
column_indices = numpy.arange(self.ncols)
for j, local in zip(column_indices, mask):
col = self._partially_evaluate((slice(None), j), simplify=True)
if local:
yield col
else:
for j in column_indices:
yield self._partially_evaluate((slice(None), j), simplify=True)
class ArrayParameter(object):
"""
Represents a parameter whose value consists of multiple values, e.g. a tuple or array.
The reason for defining this class rather than just using a NumPy array is
to avoid the ambiguity of "is a given array a single parameter value (e.g.
a spike train for one cell) or an array of parameter values (e.g. one number
per cell)?".
Arguments:
`value`:
anything which can be converted to a NumPy array, or another
:class:`ArrayParameter` object.
"""
def __init__(self, value):
if isinstance(value, ArrayParameter):
self.value = value.value
elif isinstance(value, numpy.ndarray):
# dont overwrite dtype of int arrays
self.value = value
else:
self.value = numpy.array(value, float)
# def __len__(self):
# This must not be defined, otherwise ArrayParameter is insufficiently different from NumPy array
def max(self):
"""Return the maximum value."""
return self.value.max()
def __add__(self, val):
"""
Return a new :class:`ArrayParameter` in which all values in the original
:class:`ArrayParameter` have `val` added to them.
If `val` is itself an array, return an array of :class:`ArrayParameter`
objects, where ArrayParameter `i` is the original ArrayParameter added to
element `i` of val.
"""
if hasattr(val, '__len__'):
return numpy.array([self.__class__(self.value + x) for x in val], dtype=self.__class__) # reshape if necessary?
else:
return self.__class__(self.value + val)
def __sub__(self, val):
"""
Return a new :class:`ArrayParameter` in which all values in the original
:class:`ArrayParameter` have `val` subtracted from them.
If `val` is itself an array, return an array of :class:`ArrayParameter`
objects, where ArrayParameter `i` is the original ArrayParameter with
element `i` of val subtracted from it.
"""
if hasattr(val, '__len__'):
return numpy.array([self.__class__(self.value - x) for x in val], dtype=self.__class__) # reshape if necessary?
else:
return self.__class__(self.value - val)
def __mul__(self, val):
"""
Return a new :class:`ArrayParameter` in which all values in the original
:class:`ArrayParameter` have been multiplied by `val`.
If `val` is itself an array, return an array of :class:`ArrayParameter`
objects, where ArrayParameter `i` is the original ArrayParameter multiplied by
element `i` of `val`.
"""
if hasattr(val, '__len__'):
return numpy.array([self.__class__(self.value * x) for x in val], dtype=self.__class__) # reshape if necessary?
else:
return self.__class__(self.value * val)
__rmul__ = __mul__
def __div__(self, val):
"""
Return a new :class:`ArrayParameter` in which all values in the original
:class:`ArrayParameter` have been divided by `val`.
If `val` is itself an array, return an array of :class:`ArrayParameter`
objects, where ArrayParameter `i` is the original ArrayParameter divided by
element `i` of `val`.
"""
if hasattr(val, '__len__'):
return numpy.array([self.__class__(self.value / x) for x in val], dtype=self.__class__) # reshape if necessary?
else:
return self.__class__(self.value / val)
__truediv__ = __div__ # Python 3
def __eq__(self, other):
if isinstance(other, ArrayParameter):
return self.value.size == other.value.size and (self.value == other.value).all()
elif isinstance(other, numpy.ndarray) and other.size > 0 and isinstance(other[0], ArrayParameter):
return numpy.array([(self == seq).all() for seq in other])
else:
return False
def __repr__(self):
return "%s(%s)" % (self.__class__.__name__, self.value)
[docs]class Sequence(ArrayParameter):
"""
Represents a sequence of numerical values.
Arguments:
`value`:
anything which can be converted to a NumPy array, or another
:class:`Sequence` object.
"""
# should perhaps use neo.SpikeTrain instead of this class, or at least allow a neo SpikeTrain
pass
[docs]class ParameterSpace(object):
"""
Representation of one or more points in a parameter space.
i.e. represents one or more parameter sets, where each parameter set has
the same parameter names and types but the parameters may have different
values.
Arguments:
`parameters`:
a dict containing values of any type that may be used to construct a
`lazy array`_, i.e. `int`, `float`, NumPy array,
:class:`~pyNN.random.RandomDistribution`, function that accepts a
single argument.
`schema`:
a dict whose keys are the expected parameter names and whose values
are the expected parameter types
`component`:
optional - class for which the parameters are destined. Used in
error messages.
`shape`:
the shape of the lazy arrays that will be constructed.
.. _`lazy array`: https://lazyarray.readthedocs.org/
"""
def __init__(self, parameters, schema=None, shape=None, component=None):
"""
"""
self._parameters = {}
self.schema = schema
self._shape = shape
self.component = component
self.update(**parameters)
self._evaluated = False
def _set_shape(self, shape):
for value in self._parameters.values():
value.shape = shape
self._shape = shape
shape = property(fget=lambda self: self._shape, fset=_set_shape,
doc="Size of the lazy arrays contained within the parameter space")
[docs] def keys(self):
"""
PS.keys() -> list of PS's keys.
"""
return self._parameters.keys()
[docs] def items(self):
"""
PS.items() -> an iterator over the (key, value) items of PS.
Note that the values will all be :class:`LazyArray` objects.
"""
if hasattr(self._parameters, "iteritems"):
return self._parameters.iteritems()
else:
return self._parameters.items()
def __repr__(self):
return "<ParameterSpace %s, shape=%s>" % (", ".join(self.keys()), self.shape)
[docs] def update(self, **parameters):
"""
Update the contents of the parameter space according to the
`(key, value)` pairs in ``**parameters``. All values will be turned into
lazy arrays.
If the :class:`ParameterSpace` has a schema, the keys and the data types
of the values will be checked against the schema.
"""
if self.schema:
for name, value in parameters.items():
try:
expected_dtype = self.schema[name]
except KeyError:
if self.component:
model_name = self.component.__name__
else:
model_name = 'unknown'
raise errors.NonExistentParameterError(name,
model_name,
valid_parameter_names=self.schema.keys())
if issubclass(expected_dtype, ArrayParameter) and isinstance(value, collections.Sized):
if len(value) == 0:
value = ArrayParameter([])
elif not isinstance(value[0], ArrayParameter): # may be a more generic way to do it, but for now this special-casing seems like the most robust approach
if isinstance(value[0], collections.Sized): # e.g. list of tuples
value = type(value)([ArrayParameter(x) for x in value])
else:
value = ArrayParameter(value)
try:
self._parameters[name] = LazyArray(value, shape=self._shape,
dtype=expected_dtype)
except (TypeError, errors.InvalidParameterValueError):
raise errors.InvalidParameterValueError("For parameter %s expected %s, got %s" % (name, expected_dtype, type(value)))
except ValueError as err:
raise errors.InvalidDimensionsError(err) # maybe put the more specific error classes into lazyarray
else:
for name, value in parameters.items():
self._parameters[name] = LazyArray(value, shape=self._shape)
[docs] def __getitem__(self, name):
"""x.__getitem__(y) <==> x[y]"""
return self._parameters[name]
def __setitem__(self, name, value):
# need to add check against schema
self._parameters[name] = value
[docs] def pop(self, name, d=None):
"""
Remove the given parameter from the parameter set and from its schema,
and return its value.
"""
value = self._parameters.pop(name, d)
if self.schema:
self.schema.pop(name, d)
return value
@property
def is_homogeneous(self):
"""
True if all of the lazy arrays within are homogeneous.
"""
return all(value.is_homogeneous for value in self._parameters.values())
[docs] def evaluate(self, mask=None, simplify=False):
"""
Evaluate all lazy arrays contained in the parameter space, using the
given mask.
"""
if self._shape is None:
raise Exception("Must set shape of parameter space before evaluating")
if mask is None:
for name, value in self._parameters.items():
self._parameters[name] = value.evaluate(simplify=simplify)
self._evaluated_shape = self._shape
else:
for name, value in self._parameters.items():
if isinstance(value.base_value, RandomDistribution) and value.base_value.rng.parallel_safe:
value = value.evaluate() # can't partially evaluate if using parallel safe
self._parameters[name] = value[mask]
self._evaluated_shape = partial_shape(mask, self._shape)
self._evaluated = True
# should possibly update self.shape according to mask?
[docs] def as_dict(self):
"""
Return a plain dict containing the same keys and values as the
parameter space. The values must first have been evaluated.
"""
if not self._evaluated:
raise Exception("Must call evaluate() method before calling ParameterSpace.as_dict()")
D = {}
for name, value in self._parameters.items():
D[name] = value
assert not isinstance(D[name], LazyArray) # should all have been evaluated by now
return D
[docs] def __iter__(self):
r"""
Return an array-element-wise iterator over the parameter space.
Each item in the iterator is a dict, containing the same keys as the
:class:`ParameterSpace`. For the `i`\th dict returned by the iterator,
each value is the `i`\th element of the corresponding lazy array in the
parameter space.
Example:
>>> ps = ParameterSpace({'a': [2, 3, 5, 8], 'b': 7, 'c': lambda i: 3*i+2}, shape=(4,))
>>> ps.evaluate()
>>> for D in ps:
... print(D)
...
{'a': 2, 'c': 2, 'b': 7}
{'a': 3, 'c': 5, 'b': 7}
{'a': 5, 'c': 8, 'b': 7}
{'a': 8, 'c': 11, 'b': 7}
"""
if not self._evaluated:
raise Exception("Must call evaluate() method before iterating over a ParameterSpace")
for i in range(self._evaluated_shape[0]):
D = {}
for name, value in self._parameters.items():
if is_listlike(value):
D[name] = value[i]
else:
D[name] = value
assert not isinstance(D[name], LazyArray) # should all have been evaluated by now
yield D
[docs] def columns(self):
"""
For a 2D space, return a column-wise iterator over the parameter space.
"""
if not self._evaluated:
raise Exception("Must call evaluate() method before iterating over a ParameterSpace")
assert len(self.shape) == 2
if len(self._evaluated_shape) == 1: # values will be one-dimensional
yield self._parameters
else:
for j in range(self._evaluated_shape[1]):
D = {}
for name, value in self._parameters.items():
if is_listlike(value):
D[name] = value[:, j]
else:
D[name] = value
assert not isinstance(D[name], LazyArray) # should all have been evaluated by now
yield D
def __eq__(self, other):
return (all(a == b for a, b in zip(self._parameters.items(), other._parameters.items()))
and self.schema == other.schema
and self._shape == other._shape)
@property
def parallel_safe(self):
return any(isinstance(value.base_value, RandomDistribution) and value.base_value.rng.parallel_safe
for value in self._parameters.values())
@property
def has_native_rngs(self):
"""
Return True if the parameter set contains any NativeRNGs
"""
return any(isinstance(rd.base_value.rng, NativeRNG)
for rd in self._random_distributions())
def _random_distributions(self):
"""
An iterator over those values contained in the PS that are
derived from random distributions.
"""
return (value for value in self._parameters.values() if isinstance(value.base_value, RandomDistribution))
[docs] def expand(self, new_shape, mask):
"""
Increase the size of the ParameterSpace.
Existing array values are mapped to the indices given in mask.
New array values are set to NaN.
"""
for name, value in self._parameters.items():
if isinstance(value.base_value, numpy.ndarray):
new_base_value = numpy.ones(new_shape) * numpy.nan
new_base_value[mask] = value.base_value
self._parameters[name].base_value = new_base_value
self.shape = new_shape
def simplify(value):
"""
If `value` is a homogeneous array, return the single value that all elements
share. Otherwise, pass the value through.
"""
if isinstance(value, numpy.ndarray):
if (value == value[0]).all():
return value[0]
else:
return value
else:
return value
# alternative - need to benchmark
#if numpy.any(arr != arr[0]):
# return arr
#else:
# return arr[0]
