"""
Classes used to specify the type of a function, variable or common
sub-expression.
"""
import collections
import functools
import numbers
from collections.abc import Mapping
import numpy as np
from brian2.units.fundamentalunits import (
DIMENSIONLESS,
Dimension,
Quantity,
fail_for_dimension_mismatch,
get_unit,
get_unit_for_display,
)
from brian2.utils.caching import CacheKey
from brian2.utils.logger import get_logger
from brian2.utils.stringtools import get_identifiers, word_substitute
from .base import device_override, weakproxy_with_fallback
from .preferences import prefs
__all__ = [
"Variable",
"Constant",
"ArrayVariable",
"DynamicArrayVariable",
"Subexpression",
"AuxiliaryVariable",
"VariableView",
"Variables",
"LinkedVariable",
"linked_var",
]
logger = get_logger(__name__)
[docs]
def get_dtype(obj):
"""
Helper function to return the `numpy.dtype` of an arbitrary object.
Parameters
----------
obj : object
Any object (but typically some kind of number or array).
Returns
-------
dtype : `numpy.dtype`
The type of the given object.
"""
if hasattr(obj, "dtype"):
return obj.dtype
else:
return np.dtype(type(obj))
[docs]
def get_dtype_str(val):
"""
Returns canonical string representation of the dtype of a value or dtype
Returns
-------
dtype_str : str
The numpy dtype name
"""
if isinstance(val, np.dtype):
return val.name
if isinstance(val, type):
return get_dtype_str(val())
is_bool = val is True or val is False or val is np.True_ or val is np.False_
if is_bool:
return "bool"
if hasattr(val, "dtype"):
return get_dtype_str(val.dtype)
if isinstance(val, numbers.Number):
return get_dtype_str(np.array(val).dtype)
return f"unknown[{str(val)}, {val.__class__.__name__}]"
[docs]
def variables_by_owner(variables, owner):
owner_name = getattr(owner, "name", None)
return {
varname: var
for varname, var in variables.items()
if getattr(var.owner, "name", None) is owner_name
}
[docs]
class Variable(CacheKey):
r"""
An object providing information about model variables (including implicit
variables such as ``t`` or ``xi``). This class should never be
instantiated outside of testing code, use one of its subclasses instead.
Parameters
----------
name : 'str'
The name of the variable. Note that this refers to the *original*
name in the owning group. The same variable may be known under other
names in other groups (e.g. the variable ``v`` of a `NeuronGroup` is
known as ``v_post`` in a `Synapse` connecting to the group).
dimensions : `Dimension`, optional
The physical dimensions of the variable.
owner : `Nameable`, optional
The object that "owns" this variable, e.g. the `NeuronGroup` or
`Synapses` object that declares the variable in its model equations.
Defaults to ``None`` (the value used for `Variable` objects without an
owner, e.g. external `Constant`\ s).
dtype : `dtype`, optional
The dtype used for storing the variable. Defaults to the preference
`core.default_scalar.dtype`.
scalar : bool, optional
Whether the variable is a scalar value (``True``) or vector-valued, e.g.
defined for every neuron (``False``). Defaults to ``False``.
constant: bool, optional
Whether the value of this variable can change during a run. Defaults
to ``False``.
read_only : bool, optional
Whether this is a read-only variable, i.e. a variable that is set
internally and cannot be changed by the user (this is used for example
for the variable ``N``, the number of neurons in a group). Defaults
to ``False``.
array : bool, optional
Whether this variable is an array. Allows for simpler check than testing
``isinstance(var, ArrayVariable)``. Defaults to ``False``.
"""
_cache_irrelevant_attributes = {"owner"}
def __init__(
self,
name,
dimensions=DIMENSIONLESS,
owner=None,
dtype=None,
scalar=False,
constant=False,
read_only=False,
dynamic=False,
array=False,
):
assert isinstance(dimensions, Dimension)
#: The variable's dimensions.
self.dim = dimensions
#: The variable's name.
self.name = name
#: The `Group` to which this variable belongs.
self.owner = weakproxy_with_fallback(owner) if owner is not None else None
#: The dtype used for storing the variable.
self.dtype = dtype
if dtype is None:
self.dtype = prefs.core.default_float_dtype
if self.is_boolean:
if dimensions is not DIMENSIONLESS:
raise ValueError("Boolean variables can only be dimensionless")
#: Whether the variable is a scalar
self.scalar = scalar
#: Whether the variable is constant during a run
self.constant = constant
#: Whether the variable is read-only
self.read_only = read_only
#: Whether the variable is dynamically sized (only for non-scalars)
self.dynamic = dynamic
#: Whether the variable is an array
self.array = array
def __getstate__(self):
state = self.__dict__.copy()
state["owner"] = state["owner"].__repr__.__self__ # replace proxy
return state
def __setstate__(self, state):
state["owner"] = weakproxy_with_fallback(state["owner"])
self.__dict__ = state
@property
def is_boolean(self):
return np.issubdtype(self.dtype, np.bool_)
@property
def is_integer(self):
return np.issubdtype(self.dtype, np.signedinteger)
@property
def dtype_str(self):
"""
String representation of the numpy dtype
"""
return get_dtype_str(self)
@property
def unit(self):
"""
The `Unit` of this variable
"""
return get_unit(self.dim)
[docs]
def get_value(self):
"""
Return the value associated with the variable (without units). This
is the way variables are accessed in generated code.
"""
raise TypeError(f"Cannot get value for variable {self}")
[docs]
def set_value(self, value):
"""
Set the value associated with the variable.
"""
raise TypeError(f"Cannot set value for variable {self}")
[docs]
def get_value_with_unit(self):
"""
Return the value associated with the variable (with units).
"""
return Quantity(self.get_value(), self.dim)
[docs]
def get_addressable_value(self, name, group):
"""
Get the value (without units) of this variable in a form that can be
indexed in the context of a group. For example, if a
postsynaptic variable ``x`` is accessed in a synapse ``S`` as
``S.x_post``, the synaptic indexing scheme can be used.
Parameters
----------
name : str
The name of the variable
group : `Group`
The group providing the context for the indexing. Note that this
`group` is not necessarily the same as `Variable.owner`: a variable
owned by a `NeuronGroup` can be indexed in a different way if
accessed via a `Synapses` object.
Returns
-------
variable : object
The variable in an indexable form (without units).
"""
return self.get_value()
[docs]
def get_addressable_value_with_unit(self, name, group):
"""
Get the value (with units) of this variable in a form that can be
indexed in the context of a group. For example, if a postsynaptic
variable ``x`` is accessed in a synapse ``S`` as ``S.x_post``, the
synaptic indexing scheme can be used.
Parameters
----------
name : str
The name of the variable
group : `Group`
The group providing the context for the indexing. Note that this
`group` is not necessarily the same as `Variable.owner`: a variable
owned by a `NeuronGroup` can be indexed in a different way if
accessed via a `Synapses` object.
Returns
-------
variable : object
The variable in an indexable form (with units).
"""
return self.get_value_with_unit()
[docs]
def get_len(self):
"""
Get the length of the value associated with the variable or ``0`` for
a scalar variable.
"""
if self.scalar:
return 0
else:
return len(self.get_value())
def __len__(self):
return self.get_len()
def __repr__(self):
description = (
"<{classname}(dimensions={dimensions}, "
" dtype={dtype}, scalar={scalar}, constant={constant},"
" read_only={read_only})>"
)
return description.format(
classname=self.__class__.__name__,
dimensions=repr(self.dim),
dtype=getattr(self.dtype, "__name__", repr(self.dtype)),
scalar=repr(self.scalar),
constant=repr(self.constant),
read_only=repr(self.read_only),
)
# ------------------------------------------------------------------------------
# Concrete classes derived from `Variable` -- these are the only ones ever
# instantiated.
# ------------------------------------------------------------------------------
[docs]
class Constant(Variable):
"""
A scalar constant (e.g. the number of neurons ``N``). Information such as
the dtype or whether this variable is a boolean are directly derived from
the `value`. Most of the time `Variables.add_constant` should be used
instead of instantiating this class directly.
Parameters
----------
name : str
The name of the variable
dimensions : `Dimension`, optional
The physical dimensions of the variable. Note that the variable itself
(as referenced by value) should never have units attached.
value: reference to the variable value
The value of the constant.
owner : `Nameable`, optional
The object that "owns" this variable, for constants that belong to a
specific group, e.g. the ``N`` constant for a `NeuronGroup`. External
constants will have ``None`` (the default value).
"""
def __init__(self, name, value, dimensions=DIMENSIONLESS, owner=None):
# Determine the type of the value
is_bool = (
value is True or value is False or value is np.True_ or value is np.False_
)
if is_bool:
dtype = bool
else:
dtype = get_dtype(value)
# Use standard Python types if possible for numpy scalars
if getattr(value, "shape", None) == () and hasattr(value, "dtype"):
numpy_type = value.dtype
if np.can_cast(numpy_type, int):
value = int(value)
elif np.can_cast(numpy_type, float):
value = float(value)
elif np.can_cast(numpy_type, complex):
value = complex(value)
elif value is np.True_:
value = True
elif value is np.False_:
value = False
#: The constant's value
self.value = value
super().__init__(
dimensions=dimensions,
name=name,
owner=owner,
dtype=dtype,
scalar=True,
constant=True,
read_only=True,
)
[docs]
def get_value(self):
return self.value
[docs]
def item(self):
return self.value
[docs]
class AuxiliaryVariable(Variable):
"""
Variable description for an auxiliary variable (most likely one that is
added automatically to abstract code, e.g. ``_cond`` for a threshold
condition), specifying its type and unit for code generation. Most of the
time `Variables.add_auxiliary_variable` should be used instead of
instantiating this class directly.
Parameters
----------
name : str
The name of the variable
dimensions : `Dimension`, optional
The physical dimensions of the variable.
dtype : `dtype`, optional
The dtype used for storing the variable. If none is given, defaults
to `core.default_float_dtype`.
scalar : bool, optional
Whether the variable is a scalar value (``True``) or vector-valued, e.g.
defined for every neuron (``False``). Defaults to ``False``.
"""
def __init__(self, name, dimensions=DIMENSIONLESS, dtype=None, scalar=False):
super().__init__(dimensions=dimensions, name=name, dtype=dtype, scalar=scalar)
[docs]
def get_value(self):
raise TypeError(
f"Cannot get the value for an auxiliary variable ({self.name})."
)
[docs]
class ArrayVariable(Variable):
"""
An object providing information about a model variable stored in an array
(for example, all state variables). Most of the time `Variables.add_array`
should be used instead of instantiating this class directly.
Parameters
----------
name : 'str'
The name of the variable. Note that this refers to the *original*
name in the owning group. The same variable may be known under other
names in other groups (e.g. the variable ``v`` of a `NeuronGroup` is
known as ``v_post`` in a `Synapse` connecting to the group).
dimensions : `Dimension`, optional
The physical dimensions of the variable
owner : `Nameable`
The object that "owns" this variable, e.g. the `NeuronGroup` or
`Synapses` object that declares the variable in its model equations.
size : int
The size of the array
device : `Device`
The device responsible for the memory access.
dtype : `dtype`, optional
The dtype used for storing the variable. If none is given, defaults
to `core.default_float_dtype`.
constant : bool, optional
Whether the variable's value is constant during a run.
Defaults to ``False``.
scalar : bool, optional
Whether this array is a 1-element array that should be treated like a
scalar (e.g. for a single delay value across synapses). Defaults to
``False``.
read_only : bool, optional
Whether this is a read-only variable, i.e. a variable that is set
internally and cannot be changed by the user. Defaults
to ``False``.
unique : bool, optional
Whether the values in this array are all unique. This information is
only important for variables used as indices and does not have to
reflect the actual contents of the array but only the possibility of
non-uniqueness (e.g. synaptic indices are always unique but the
corresponding pre- and post-synaptic indices are not). Defaults to
``False``.
"""
def __init__(
self,
name,
owner,
size,
device,
dimensions=DIMENSIONLESS,
dtype=None,
constant=False,
scalar=False,
read_only=False,
dynamic=False,
unique=False,
):
super().__init__(
dimensions=dimensions,
name=name,
owner=owner,
dtype=dtype,
scalar=scalar,
constant=constant,
read_only=read_only,
dynamic=dynamic,
array=True,
)
#: Wether all values in this arrays are necessarily unique (only
#: relevant for index variables).
self.unique = unique
#: The `Device` responsible for memory access.
self.device = device
#: The size of this variable.
self.size = size
if scalar and size != 1:
raise ValueError(f"Scalar variables need to have size 1, not size {size}.")
#: Another variable, on which the write is conditioned (e.g. a variable
#: denoting the absence of refractoriness)
self.conditional_write = None
[docs]
def set_conditional_write(self, var):
if not var.is_boolean:
raise TypeError(
"A variable can only be conditionally writeable "
f"depending on a boolean variable, '{var.name}' is not "
"boolean."
)
self.conditional_write = var
[docs]
def get_value(self):
return self.device.get_value(self)
[docs]
def item(self):
if self.size == 1:
return self.get_value().item()
else:
raise ValueError("can only convert an array of size 1 to a Python scalar")
[docs]
def set_value(self, value):
self.device.fill_with_array(self, value)
[docs]
def get_len(self):
return self.size
[docs]
def get_addressable_value(self, name, group):
return VariableView(name=name, variable=self, group=group, dimensions=None)
[docs]
def get_addressable_value_with_unit(self, name, group):
return VariableView(name=name, variable=self, group=group, dimensions=self.dim)
[docs]
class DynamicArrayVariable(ArrayVariable):
"""
An object providing information about a model variable stored in a dynamic
array (used in `Synapses`). Most of the time `Variables.add_dynamic_array`
should be used instead of instantiating this class directly.
Parameters
----------
name : 'str'
The name of the variable. Note that this refers to the *original*
name in the owning group. The same variable may be known under other
names in other groups (e.g. the variable ``v`` of a `NeuronGroup` is
known as ``v_post`` in a `Synapse` connecting to the group).
dimensions : `Dimension`, optional
The physical dimensions of the variable.
owner : `Nameable`
The object that "owns" this variable, e.g. the `NeuronGroup` or
`Synapses` object that declares the variable in its model equations.
size : int or tuple of int
The (initial) size of the variable.
device : `Device`
The device responsible for the memory access.
dtype : `dtype`, optional
The dtype used for storing the variable. If none is given, defaults
to `core.default_float_dtype`.
constant : bool, optional
Whether the variable's value is constant during a run.
Defaults to ``False``.
needs_reference_update : bool, optional
Whether the code objects need a new reference to the underlying data at
every time step. This should be set if the size of the array can be
changed by other code objects. Defaults to ``False``.
scalar : bool, optional
Whether this array is a 1-element array that should be treated like a
scalar (e.g. for a single delay value across synapses). Defaults to
``False``.
read_only : bool, optional
Whether this is a read-only variable, i.e. a variable that is set
internally and cannot be changed by the user. Defaults
to ``False``.
unique : bool, optional
Whether the values in this array are all unique. This information is
only important for variables used as indices and does not have to
reflect the actual contents of the array but only the possibility of
non-uniqueness (e.g. synaptic indices are always unique but the
corresponding pre- and post-synaptic indices are not). Defaults to
``False``.
"""
# The size of a dynamic variable can of course change and changes in
# size should not invalidate the cache
_cache_irrelevant_attributes = ArrayVariable._cache_irrelevant_attributes | {"size"}
def __init__(
self,
name,
owner,
size,
device,
dimensions=DIMENSIONLESS,
dtype=None,
constant=False,
needs_reference_update=False,
resize_along_first=False,
scalar=False,
read_only=False,
unique=False,
):
if isinstance(size, int):
ndim = 1
else:
ndim = len(size)
#: The number of dimensions
self.ndim = ndim
if constant and needs_reference_update:
raise ValueError("A variable cannot be constant and need reference updates")
#: Whether this variable needs an update of the reference to the
#: underlying data whenever it is passed to a code object
self.needs_reference_update = needs_reference_update
#: Whether this array will be only resized along the first dimension
self.resize_along_first = resize_along_first
super().__init__(
dimensions=dimensions,
owner=owner,
name=name,
size=size,
device=device,
constant=constant,
dtype=dtype,
scalar=scalar,
dynamic=True,
read_only=read_only,
unique=unique,
)
@property
def dimensions(self):
logger.warn(
"The DynamicArrayVariable.dimensions attribute is "
"deprecated, use .ndim instead",
"deprecated_dimensions",
once=True,
)
return self.ndim
[docs]
def resize(self, new_size):
"""
Resize the dynamic array. Calls `self.device.resize` to do the
actual resizing.
Parameters
----------
new_size : int or tuple of int
The new size.
"""
if self.resize_along_first:
self.device.resize_along_first(self, new_size)
else:
self.device.resize(self, new_size)
self.size = new_size
[docs]
class Subexpression(Variable):
"""
An object providing information about a named subexpression in a model.
Most of the time `Variables.add_subexpression` should be used instead of
instantiating this class directly.
Parameters
----------
name : str
The name of the subexpression.
dimensions : `Dimension`, optional
The physical dimensions of the subexpression.
owner : `Group`
The group to which the expression refers.
expr : str
The subexpression itself.
device : `Device`
The device responsible for the memory access.
dtype : `dtype`, optional
The dtype used for the expression. Defaults to
`core.default_float_dtype`.
scalar: bool, optional
Whether this is an expression only referring to scalar variables.
Defaults to ``False``
"""
def __init__(
self,
name,
owner,
expr,
device,
dimensions=DIMENSIONLESS,
dtype=None,
scalar=False,
):
super().__init__(
dimensions=dimensions,
owner=owner,
name=name,
dtype=dtype,
scalar=scalar,
constant=False,
read_only=True,
)
#: The `Device` responsible for memory access
self.device = device
#: The expression defining the subexpression
self.expr = expr.strip()
#: The identifiers used in the expression
self.identifiers = get_identifiers(expr)
[docs]
def get_addressable_value(self, name, group):
return VariableView(
name=name, variable=self, group=group, dimensions=DIMENSIONLESS
)
[docs]
def get_addressable_value_with_unit(self, name, group):
return VariableView(name=name, variable=self, group=group, dimensions=self.dim)
def __contains__(self, var):
return var in self.identifiers
def __repr__(self):
description = (
"<{classname}(name={name}, dimensions={dimensions}, dtype={dtype}, "
"expr={expr}, owner=<{owner}>)>"
)
return description.format(
classname=self.__class__.__name__,
name=repr(self.name),
dimensions=repr(self.dim),
dtype=repr(self.dtype),
expr=repr(self.expr),
owner=self.owner.name,
)
# ------------------------------------------------------------------------------
# Classes providing views on variables and storing variables information
# ------------------------------------------------------------------------------
[docs]
class LinkedVariable:
"""
A simple helper class to make linking variables explicit. Users should use
`linked_var` instead.
Parameters
----------
group : `Group`
The group through which the `variable` is accessed (not necessarily the
same as ``variable.owner``.
name : str
The name of `variable` in `group` (not necessarily the same as
``variable.name``).
variable : `Variable`
The variable that should be linked.
index : str or `ndarray`, optional
An indexing array (or the name of a state variable), providing a mapping
from the entries in the link source to the link target.
"""
def __init__(self, group, name, variable, index=None):
self.group = group
self.name = name
self.variable = variable
self.index = index
[docs]
def linked_var(group_or_variable, name=None, index=None):
"""
Represents a link target for setting a linked variable.
Parameters
----------
group_or_variable : `NeuronGroup` or `VariableView`
Either a reference to the target `NeuronGroup` (e.g. ``G``) or a direct
reference to a `VariableView` object (e.g. ``G.v``). In case only the
group is specified, `name` has to be specified as well.
name : str, optional
The name of the target variable, necessary if `group_or_variable` is a
`NeuronGroup`.
index : str or `ndarray`, optional
An indexing array (or the name of a state variable), providing a mapping
from the entries in the link source to the link target.
Examples
--------
>>> from brian2 import *
>>> G1 = NeuronGroup(10, 'dv/dt = -v / (10*ms) : volt')
>>> G2 = NeuronGroup(10, 'v : volt (linked)')
>>> G2.v = linked_var(G1, 'v')
>>> G2.v = linked_var(G1.v) # equivalent
"""
if isinstance(group_or_variable, VariableView):
if name is not None:
raise ValueError(
"Cannot give a variable and a variable name at the same time."
)
return LinkedVariable(
group_or_variable.group,
group_or_variable.name,
group_or_variable.variable,
index=index,
)
elif name is None:
raise ValueError("Need to provide a variable name")
else:
return LinkedVariable(
group_or_variable, name, group_or_variable.variables[name], index=index
)
[docs]
class VariableView:
"""
A view on a variable that allows to treat it as an numpy array while
allowing special indexing (e.g. with strings) in the context of a `Group`.
Parameters
----------
name : str
The name of the variable (not necessarily the same as ``variable.name``).
variable : `Variable`
The variable description.
group : `Group`
The group through which the variable is accessed (not necessarily the
same as `variable.owner`).
dimensions : `Dimension`, optional
The physical dimensions to be used for the variable, should be `None`
when a variable is accessed without units (e.g. when accessing
``G.var_``).
"""
__array_priority__ = 10
def __init__(self, name, variable, group, dimensions=None):
self.name = name
self.variable = variable
self.index_var_name = group.variables.indices[name]
if self.index_var_name in ("_idx", "0"):
self.index_var = self.index_var_name
else:
self.index_var = group.variables[self.index_var_name]
if isinstance(variable, Subexpression):
# For subexpressions, we *always* have to go via codegen to get
# their value -- since we cannot do this without the group, we
# hold a strong reference
self.group = group
else:
# For state variable arrays, we can do most access without the full
# group, using the indexing reference below. We therefore only keep
# a weak reference to the group.
self.group = weakproxy_with_fallback(group)
self.group_name = group.name
# We keep a strong reference to the `Indexing` object so that basic
# indexing is still possible, even if the group no longer exists
self.indexing = self.group._indices
self.dim = dimensions
@property
def unit(self):
"""
The `Unit` of this variable
"""
return get_unit(self.dim)
[docs]
def get_item(self, item, level=0, namespace=None):
"""
Get the value of this variable. Called by `__getitem__`.
Parameters
----------
item : slice, `ndarray` or string
The index for the setting operation
level : int, optional
How much farther to go up in the stack to find the implicit
namespace (if used, see `run_namespace`).
namespace : dict-like, optional
An additional namespace that is used for variable lookup (if not
defined, the implicit namespace of local variables is used).
"""
from brian2.core.namespace import get_local_namespace # avoids circular import
if isinstance(item, str):
# Check whether the group still exists to give a more meaningful
# error message if it does not
try:
self.group.name
except ReferenceError:
raise ReferenceError(
"Cannot use string expressions, the "
f"group '{self.group_name}', providing the "
"context for the expression, no longer exists. "
"Consider holding an explicit reference "
"to it to keep it alive."
)
if namespace is None:
namespace = get_local_namespace(level=level + 1)
values = self.get_with_expression(item, run_namespace=namespace)
else:
if isinstance(self.variable, Subexpression):
if namespace is None:
namespace = get_local_namespace(level=level + 1)
values = self.get_subexpression_with_index_array(
item, run_namespace=namespace
)
else:
values = self.get_with_index_array(item)
if self.dim is DIMENSIONLESS or self.dim is None:
return values
else:
return Quantity(values, self.dim)
def __getitem__(self, item):
return self.get_item(item, level=1)
[docs]
def set_item(self, item, value, level=0, namespace=None):
"""
Set this variable. This function is called by `__setitem__` but there
is also a situation where it should be called directly: if the context
for string-based expressions is higher up in the stack, this function
allows to set the `level` argument accordingly.
Parameters
----------
item : slice, `ndarray` or string
The index for the setting operation
value : `Quantity`, `ndarray` or number
The value for the setting operation
level : int, optional
How much farther to go up in the stack to find the implicit
namespace (if used, see `run_namespace`).
namespace : dict-like, optional
An additional namespace that is used for variable lookup (if not
defined, the implicit namespace of local variables is used).
"""
from brian2.core.namespace import get_local_namespace # avoids circular import
variable = self.variable
if variable.read_only:
raise TypeError(f"Variable {self.name} is read-only.")
# Check whether the group allows writing to the variable (e.g. for
# synaptic variables, writing is only allowed after a connect)
try:
self.group.check_variable_write(variable)
except ReferenceError:
# Ignore problems with weakly referenced groups that don't exist
# anymore at this time (e.g. when doing neuron.axon.var = ...)
pass
# The second part is equivalent to item == slice(None) but formulating
# it this way prevents a FutureWarning if one of the elements is a
# numpy array
if isinstance(item, slice) and (
item.start is None and item.stop is None and item.step is None
):
item = "True"
check_units = self.dim is not None
if namespace is None:
namespace = get_local_namespace(level=level + 1)
# Both index and values are strings, use a single code object do deal
# with this situation
if isinstance(value, str) and isinstance(item, str):
self.set_with_expression_conditional(
item, value, check_units=check_units, run_namespace=namespace
)
elif isinstance(item, str):
try:
if isinstance(value, str):
raise TypeError # Will be dealt with below
value = np.asanyarray(value).item()
except (TypeError, ValueError):
if item != "True":
raise TypeError(
"When setting a variable based on a string "
"index, the value has to be a string or a "
"scalar."
)
if item == "True":
# We do not want to go through code generation for runtime
self.set_with_index_array(slice(None), value, check_units=check_units)
else:
self.set_with_expression_conditional(
item, repr(value), check_units=check_units, run_namespace=namespace
)
elif isinstance(value, str):
self.set_with_expression(
item, value, check_units=check_units, run_namespace=namespace
)
else: # No string expressions involved
self.set_with_index_array(item, value, check_units=check_units)
def __setitem__(self, item, value):
self.set_item(item, value, level=1)
[docs]
@device_override("variableview_set_with_expression")
def set_with_expression(self, item, code, run_namespace, check_units=True):
"""
Sets a variable using a string expression. Is called by
`VariableView.set_item` for statements such as
``S.var[:, :] = 'exp(-abs(i-j)/space_constant)*nS'``
Parameters
----------
item : `ndarray`
The indices for the variable (in the context of this `group`).
code : str
The code that should be executed to set the variable values.
Can contain references to indices, such as `i` or `j`
run_namespace : dict-like, optional
An additional namespace that is used for variable lookup (if not
defined, the implicit namespace of local variables is used).
check_units : bool, optional
Whether to check the units of the expression.
run_namespace : dict-like, optional
An additional namespace that is used for variable lookup (if not
defined, the implicit namespace of local variables is used).
"""
# Some fairly complicated code to raise a warning in ambiguous
# situations, when indexing with a group. For example, in:
# group.v[subgroup] = 'i'
# the index 'i' is the index of 'group' ("absolute index") and not of
# subgroup ("relative index")
if hasattr(item, "variables") or (
isinstance(item, tuple)
and any(hasattr(one_item, "variables") for one_item in item)
):
# Determine the variables that are used in the expression
from brian2.codegen.translation import get_identifiers_recursively
identifiers = get_identifiers_recursively([code], self.group.variables)
variables = self.group.resolve_all(
identifiers, run_namespace, user_identifiers=set()
)
if not isinstance(item, tuple):
index_groups = [item]
else:
index_groups = item
for varname, var in variables.items():
for index_group in index_groups:
if not hasattr(index_group, "variables"):
continue
if (
varname in index_group.variables
or var.name in index_group.variables
):
indexed_var = index_group.variables.get(
varname, index_group.variables.get(var.name)
)
if indexed_var is not var:
logger.warn(
"The string expression used for setting "
f"'{self.name}' refers to '{varname}' which "
"might be ambiguous. It will be "
"interpreted as referring to "
f"'{varname}' in '{self.group.name}', not as "
"a variable of a group used for "
"indexing.",
"ambiguous_string_expression",
)
break # no need to warn more than once for a variable
indices = np.atleast_1d(self.indexing(item))
abstract_code = f"{self.name} = {code}"
variables = Variables(self.group)
variables.add_array(
"_group_idx", size=len(indices), dtype=np.int32, values=indices
)
# TODO: Have an additional argument to avoid going through the index
# array for situations where iterate_all could be used
from brian2.codegen.codeobject import create_runner_codeobj
from brian2.devices.device import get_device
device = get_device()
codeobj = create_runner_codeobj(
self.group,
abstract_code,
"group_variable_set",
additional_variables=variables,
check_units=check_units,
run_namespace=run_namespace,
codeobj_class=device.code_object_class(
fallback_pref="codegen.string_expression_target"
),
)
codeobj()
[docs]
@device_override("variableview_set_with_expression_conditional")
def set_with_expression_conditional(
self, cond, code, run_namespace, check_units=True
):
"""
Sets a variable using a string expression and string condition. Is
called by `VariableView.set_item` for statements such as
``S.var['i!=j'] = 'exp(-abs(i-j)/space_constant)*nS'``
Parameters
----------
cond : str
The string condition for which the variables should be set.
code : str
The code that should be executed to set the variable values.
run_namespace : dict-like, optional
An additional namespace that is used for variable lookup (if not
defined, the implicit namespace of local variables is used).
check_units : bool, optional
Whether to check the units of the expression.
"""
variable = self.variable
if variable.scalar and cond != "True":
raise IndexError(
f"Cannot conditionally set the scalar variable '{self.name}'."
)
abstract_code_cond = f"_cond = {cond}"
abstract_code = f"{self.name} = {code}"
variables = Variables(None)
variables.add_auxiliary_variable("_cond", dtype=bool)
from brian2.codegen.codeobject import create_runner_codeobj
# TODO: Have an additional argument to avoid going through the index
# array for situations where iterate_all could be used
from brian2.devices.device import get_device
device = get_device()
codeobj = create_runner_codeobj(
self.group,
{"condition": abstract_code_cond, "statement": abstract_code},
"group_variable_set_conditional",
additional_variables=variables,
check_units=check_units,
run_namespace=run_namespace,
codeobj_class=device.code_object_class(
fallback_pref="codegen.string_expression_target"
),
)
codeobj()
[docs]
@device_override("variableview_get_with_expression")
def get_with_expression(self, code, run_namespace):
"""
Gets a variable using a string expression. Is called by
`VariableView.get_item` for statements such as
``print(G.v['g_syn > 0'])``.
Parameters
----------
code : str
An expression that states a condition for elements that should be
selected. Can contain references to indices, such as ``i`` or ``j``
and to state variables. For example: ``'i>3 and v>0*mV'``.
run_namespace : dict-like
An additional namespace that is used for variable lookup (either
an explicitly defined namespace or one taken from the local
context).
"""
variable = self.variable
if variable.scalar:
raise IndexError(
f"Cannot access the variable '{self.name}' with a "
"string expression, it is a scalar variable."
)
# Add the recorded variable under a known name to the variables
# dictionary. Important to deal correctly with
# the type of the variable in C++
variables = Variables(None)
variables.add_auxiliary_variable(
"_variable",
dimensions=variable.dim,
dtype=variable.dtype,
scalar=variable.scalar,
)
variables.add_auxiliary_variable("_cond", dtype=bool)
abstract_code = f"_variable = {self.name}\n"
abstract_code += f"_cond = {code}"
from brian2.codegen.codeobject import create_runner_codeobj
from brian2.devices.device import get_device
device = get_device()
codeobj = create_runner_codeobj(
self.group,
abstract_code,
"group_variable_get_conditional",
additional_variables=variables,
run_namespace=run_namespace,
codeobj_class=device.code_object_class(
fallback_pref="codegen.string_expression_target"
),
)
return codeobj()
[docs]
@device_override("variableview_get_with_index_array")
def get_with_index_array(self, item):
variable = self.variable
if variable.scalar:
if not (isinstance(item, slice) and item == slice(None)):
raise IndexError(
f"Illegal index for variable '{self.name}', it is a "
"scalar variable."
)
indices = 0
elif (
isinstance(item, slice) and item == slice(None) and self.index_var == "_idx"
):
indices = slice(None)
# Quick fix for matplotlib calling 1-d variables as var[:, np.newaxis]
# The test is a bit verbose, but we need to avoid comparisons that raise errors
# (e.g. comparing an array to slice(None))
elif (
isinstance(item, tuple)
and len(item) == 2
and isinstance(item[0], slice)
and item[0] == slice(None)
and item[1] is None
):
if self.index_var == "_idx":
return variable.get_value()[item]
else:
return variable.get_value()[self.index_var.get_value()][item]
else:
indices = self.indexing(item, self.index_var)
return variable.get_value()[indices]
[docs]
@device_override("variableview_get_subexpression_with_index_array")
def get_subexpression_with_index_array(self, item, run_namespace):
variable = self.variable
if variable.scalar:
if not (isinstance(item, slice) and item == slice(None)):
raise IndexError(
f"Illegal index for variable '{self.name}', it is a "
"scalar variable."
)
indices = np.array(0)
else:
indices = self.indexing(item, self.index_var)
# For "normal" variables, we can directly access the underlying data
# and use the usual slicing syntax. For subexpressions, however, we
# have to evaluate code for the given indices
variables = Variables(None, default_index="_group_index")
variables.add_auxiliary_variable(
"_variable",
dimensions=variable.dim,
dtype=variable.dtype,
scalar=variable.scalar,
)
if indices.shape == ():
single_index = True
indices = np.array([indices])
else:
single_index = False
variables.add_array("_group_idx", size=len(indices), dtype=np.int32)
variables["_group_idx"].set_value(indices)
# Force the use of this variable as a replacement for the original
# index variable
using_orig_index = [
varname
for varname, index in self.group.variables.indices.items()
if index == self.index_var_name and index != "0"
]
for varname in using_orig_index:
variables.indices[varname] = "_idx"
abstract_code = f"_variable = {self.name}\n"
from brian2.codegen.codeobject import create_runner_codeobj
from brian2.devices.device import get_device
device = get_device()
codeobj = create_runner_codeobj(
self.group,
abstract_code,
"group_variable_get",
# Setting the user code to an empty
# string suppresses warnings if the
# subexpression refers to variable
# names that are also present in the
# local namespace
user_code="",
needed_variables=["_group_idx"],
additional_variables=variables,
run_namespace=run_namespace,
codeobj_class=device.code_object_class(
fallback_pref="codegen.string_expression_target"
),
)
result = codeobj()
if single_index and not variable.scalar:
return result[0]
else:
return result
[docs]
@device_override("variableview_set_with_index_array")
def set_with_index_array(self, item, value, check_units):
variable = self.variable
if check_units:
fail_for_dimension_mismatch(
variable.dim, value, f"Incorrect unit for setting variable {self.name}"
)
if variable.scalar:
if not (isinstance(item, slice) and item == slice(None)):
raise IndexError(
"Illegal index for variable '{self.name}', it is a scalar variable."
)
indices = 0
elif (
isinstance(item, slice) and item == slice(None) and self.index_var == "_idx"
):
indices = slice(None)
else:
indices = self.indexing(item, self.index_var)
q = Quantity(value)
if len(q.shape):
if not len(q.shape) == 1 or len(q) != 1 and len(q) != len(indices):
raise ValueError(
"Provided values do not match the size "
"of the indices, "
f"{len(q)} != {len(indices)}."
)
variable.get_value()[indices] = value
# Allow some basic calculations directly on the ArrayView object
def __array__(self, dtype=None):
try:
# This will fail for subexpressions that refer to external
# parameters
self[:]
except ValueError:
var = self.variable.name
raise ValueError(
f"Cannot get the values for variable {var}. If it "
"is a subexpression referring to external "
f"variables, use 'group.{var}[:]' instead of "
f"'group.{var}'"
)
return np.asanyarray(self[:], dtype=dtype)
def __array__ufunc__(self, ufunc, method, *inputs, **kwargs):
if method == "__call__":
return ufunc(self[:], *inputs, **kwargs)
else:
return NotImplemented
def __len__(self):
return len(self.get_item(slice(None), level=1))
def __neg__(self):
return -self.get_item(slice(None), level=1)
def __pos__(self):
return self.get_item(slice(None), level=1)
def __add__(self, other):
return self.get_item(slice(None), level=1) + np.asanyarray(other)
def __radd__(self, other):
return np.asanyarray(other) + self.get_item(slice(None), level=1)
def __sub__(self, other):
return self.get_item(slice(None), level=1) - np.asanyarray(other)
def __rsub__(self, other):
return np.asanyarray(other) - self.get_item(slice(None), level=1)
def __mul__(self, other):
return self.get_item(slice(None), level=1) * np.asanyarray(other)
def __rmul__(self, other):
return np.asanyarray(other) * self.get_item(slice(None), level=1)
def __div__(self, other):
return self.get_item(slice(None), level=1) / np.asanyarray(other)
def __truediv__(self, other):
return self.get_item(slice(None), level=1) / np.asanyarray(other)
def __floordiv__(self, other):
return self.get_item(slice(None), level=1) // np.asanyarray(other)
def __rdiv__(self, other):
return np.asanyarray(other) / self.get_item(slice(None), level=1)
def __rtruediv__(self, other):
return np.asanyarray(other) / self.get_item(slice(None), level=1)
def __rfloordiv__(self, other):
return np.asanyarray(other) // self.get_item(slice(None), level=1)
def __mod__(self, other):
return self.get_item(slice(None), level=1) % np.asanyarray(other)
def __pow__(self, power, modulo=None):
if modulo is not None:
return self.get_item(slice(None), level=1) ** power % modulo
else:
return self.get_item(slice(None), level=1) ** power
def __rpow__(self, other):
if self.dim is not DIMENSIONLESS:
raise TypeError(
f"Cannot use '{self.name}' as an exponent, it has "
f"dimensions {get_unit_for_display(self.unit)}."
)
return other ** self.get_item(slice(None), level=1)
def __iadd__(self, other):
if isinstance(other, str):
raise TypeError(
"In-place modification with strings not "
"supported. Use group.var = 'var + expression' "
"instead of group.var += 'expression'."
)
elif isinstance(self.variable, Subexpression):
raise TypeError("Cannot assign to a subexpression.")
else:
rhs = self[:] + np.asanyarray(other)
self[:] = rhs
return self
def __isub__(self, other):
if isinstance(other, str):
raise TypeError(
"In-place modification with strings not "
"supported. Use group.var = 'var - expression' "
"instead of group.var -= 'expression'."
)
elif isinstance(self.variable, Subexpression):
raise TypeError("Cannot assign to a subexpression.")
else:
rhs = self[:] - np.asanyarray(other)
self[:] = rhs
return self
def __imul__(self, other):
if isinstance(other, str):
raise TypeError(
"In-place modification with strings not "
"supported. Use group.var = 'var * expression' "
"instead of group.var *= 'expression'."
)
elif isinstance(self.variable, Subexpression):
raise TypeError("Cannot assign to a subexpression.")
else:
rhs = self[:] * np.asanyarray(other)
self[:] = rhs
return self
def __idiv__(self, other):
if isinstance(other, str):
raise TypeError(
"In-place modification with strings not "
"supported. Use group.var = 'var / expression' "
"instead of group.var /= 'expression'."
)
elif isinstance(self.variable, Subexpression):
raise TypeError("Cannot assign to a subexpression.")
else:
rhs = self[:] / np.asanyarray(other)
self[:] = rhs
return self
def __ifloordiv__(self, other):
if isinstance(other, str):
raise TypeError(
"In-place modification with strings not "
"supported. Use group.var = 'var // expression' "
"instead of group.var //= 'expression'."
)
elif isinstance(self.variable, Subexpression):
raise TypeError("Cannot assign to a subexpression.")
else:
rhs = self[:] // np.asanyarray(other)
self[:] = rhs
return self
def __imod__(self, other):
if isinstance(other, str):
raise TypeError(
"In-place modification with strings not "
"supported. Use group.var = 'var // expression' "
"instead of group.var //= 'expression'."
)
elif isinstance(self.variable, Subexpression):
raise TypeError("Cannot assign to a subexpression.")
else:
rhs = self[:] % np.asanyarray(other)
self[:] = rhs
return self
def __ipow__(self, other):
if isinstance(other, str):
raise TypeError(
"In-place modification with strings not "
"supported. Use group.var = 'var ** expression' "
"instead of group.var **= 'expression'."
)
elif isinstance(self.variable, Subexpression):
raise TypeError("Cannot assign to a subexpression.")
else:
rhs = self[:] ** np.asanyarray(other)
self[:] = rhs
return self
# Also allow logical comparisons
def __eq__(self, other):
return self.get_item(slice(None), level=1) == np.asanyarray(other)
def __ne__(self, other):
return self.get_item(slice(None), level=1) != np.asanyarray(other)
def __lt__(self, other):
return self.get_item(slice(None), level=1) < np.asanyarray(other)
def __le__(self, other):
return self.get_item(slice(None), level=1) <= np.asanyarray(other)
def __gt__(self, other):
return self.get_item(slice(None), level=1) > np.asanyarray(other)
def __ge__(self, other):
return self.get_item(slice(None), level=1) >= np.asanyarray(other)
def __repr__(self):
varname = self.name
if self.dim is None:
varname += "_"
if self.variable.scalar:
dim = self.dim if self.dim is not None else DIMENSIONLESS
values = repr(Quantity(self.variable.get_value().item(), dim=dim))
else:
try:
# This will fail for subexpressions that refer to external
# parameters
values = repr(self[:])
except KeyError:
values = (
"[Subexpression refers to external parameters. Use "
f"'group.{self.variable.name}[:]']"
)
return f"<{self.group_name}.{varname}: {values}>"
def __hash__(self):
return hash((self.group_name, self.name))
# Get access to some basic properties of the underlying array
@property
def shape(self):
if self.ndim == 1:
return (self.variable.size,)
else:
return self.variable.size
@property
def ndim(self):
return getattr(self.variable, "ndim", 1)
@property
def dtype(self):
return self.variable.dtype
[docs]
class Variables(Mapping):
"""
A container class for storing `Variable` objects. Instances of this class
are used as the `Group.variables` attribute and can be accessed as
(read-only) dictionaries.
Parameters
----------
owner : `Nameable`
The object (typically a `Group`) "owning" the variables.
default_index : str, optional
The index to use for the variables (only relevant for `ArrayVariable`
and `DynamicArrayVariable`). Defaults to ``'_idx'``.
"""
def __init__(self, owner, default_index="_idx"):
#: A reference to the `Group` owning these variables
self.owner = weakproxy_with_fallback(owner)
# The index that is used for arrays if no index is given explicitly
self.default_index = default_index
# We do the import here to avoid a circular dependency.
from brian2.devices.device import get_device
self.device = get_device()
self._variables = {}
#: A dictionary given the index name for every array name
self.indices = collections.defaultdict(functools.partial(str, default_index))
# Note that by using functools.partial (instead of e.g. a lambda
# function) above, this object remains pickable.
def __getstate__(self):
state = self.__dict__.copy()
state["owner"] = state["owner"].__repr__.__self__
return state
def __setstate__(self, state):
state["owner"] = weakproxy_with_fallback(state["owner"])
self.__dict__ = state
def __getitem__(self, item):
return self._variables[item]
def __len__(self):
return len(self._variables)
def __iter__(self):
return iter(self._variables)
def _add_variable(self, name, var, index=None):
if name in self._variables:
raise KeyError(
f"The name '{name}' is already present in the variables dictionary."
)
# TODO: do some check for the name, part of it has to be device-specific
self._variables[name] = var
if isinstance(var, ArrayVariable):
# Tell the device to actually create the array (or note it down for
# later code generation in standalone).
self.device.add_array(var)
if getattr(var, "scalar", False):
if index not in (None, "0"):
raise ValueError("Cannot set an index for a scalar variable")
self.indices[name] = "0"
if index is not None:
self.indices[name] = index
[docs]
def add_array(
self,
name,
size,
dimensions=DIMENSIONLESS,
values=None,
dtype=None,
constant=False,
read_only=False,
scalar=False,
unique=False,
index=None,
):
"""
Add an array (initialized with zeros).
Parameters
----------
name : str
The name of the variable.
dimensions : `Dimension`, optional
The physical dimensions of the variable.
size : int
The size of the array.
values : `ndarray`, optional
The values to initalize the array with. If not specified, the array
is initialized to zero.
dtype : `dtype`, optional
The dtype used for storing the variable. If none is given, defaults
to `core.default_float_dtype`.
constant : bool, optional
Whether the variable's value is constant during a run.
Defaults to ``False``.
scalar : bool, optional
Whether this is a scalar variable. Defaults to ``False``, if set to
``True``, also implies that `size` equals 1.
read_only : bool, optional
Whether this is a read-only variable, i.e. a variable that is set
internally and cannot be changed by the user. Defaults
to ``False``.
index : str, optional
The index to use for this variable. Defaults to
`Variables.default_index`.
unique : bool, optional
See `ArrayVariable`. Defaults to ``False``.
"""
if np.asanyarray(size).shape == ():
# We want a basic Python type for the size instead of something
# like numpy.int64
size = int(size)
var = ArrayVariable(
name=name,
dimensions=dimensions,
owner=self.owner,
device=self.device,
size=size,
dtype=dtype,
constant=constant,
scalar=scalar,
read_only=read_only,
unique=unique,
)
self._add_variable(name, var, index)
# This could be avoided, but we currently need it so that standalone
# allocates the memory
self.device.init_with_zeros(var, dtype)
if values is not None:
if scalar:
if np.asanyarray(values).shape != ():
raise ValueError("Need a scalar value.")
self.device.fill_with_array(var, values)
else:
if len(values) != size:
raise ValueError(
"Size of the provided values does not match "
f"size: {len(values)} != {size}"
)
self.device.fill_with_array(var, values)
[docs]
def add_arrays(
self,
names,
size,
dimensions=DIMENSIONLESS,
dtype=None,
constant=False,
read_only=False,
scalar=False,
unique=False,
index=None,
):
"""
Adds several arrays (initialized with zeros) with the same attributes
(size, units, etc.).
Parameters
----------
names : list of str
The names of the variable.
dimensions : `Dimension`, optional
The physical dimensions of the variable.
size : int
The sizes of the arrays.
dtype : `dtype`, optional
The dtype used for storing the variables. If none is given, defaults
to `core.default_float_dtype`.
constant : bool, optional
Whether the variables' values are constant during a run.
Defaults to ``False``.
scalar : bool, optional
Whether these are scalar variables. Defaults to ``False``, if set to
``True``, also implies that `size` equals 1.
read_only : bool, optional
Whether these are read-only variables, i.e. variables that are set
internally and cannot be changed by the user. Defaults
to ``False``.
index : str, optional
The index to use for these variables. Defaults to
`Variables.default_index`.
unique : bool, optional
See `ArrayVariable`. Defaults to ``False``.
"""
for name in names:
self.add_array(
name,
dimensions=dimensions,
size=size,
dtype=dtype,
constant=constant,
read_only=read_only,
scalar=scalar,
unique=unique,
index=index,
)
[docs]
def add_dynamic_array(
self,
name,
size,
dimensions=DIMENSIONLESS,
values=None,
dtype=None,
constant=False,
needs_reference_update=False,
resize_along_first=False,
read_only=False,
unique=False,
scalar=False,
index=None,
):
"""
Add a dynamic array.
Parameters
----------
name : str
The name of the variable.
dimensions : `Dimension`, optional
The physical dimensions of the variable.
size : int or tuple of int
The (initital) size of the array.
values : `ndarray`, optional
The values to initalize the array with. If not specified, the array
is initialized to zero.
dtype : `dtype`, optional
The dtype used for storing the variable. If none is given, defaults
to `core.default_float_dtype`.
constant : bool, optional
Whether the variable's value is constant during a run.
Defaults to ``False``.
needs_reference_update : bool, optional
Whether the code objects need a new reference to the underlying data at
every time step. This should be set if the size of the array can be
changed by other code objects. Defaults to ``False``.
scalar : bool, optional
Whether this is a scalar variable. Defaults to ``False``, if set to
``True``, also implies that `size` equals 1.
read_only : bool, optional
Whether this is a read-only variable, i.e. a variable that is set
internally and cannot be changed by the user. Defaults
to ``False``.
index : str, optional
The index to use for this variable. Defaults to
`Variables.default_index`.
unique : bool, optional
See `DynamicArrayVariable`. Defaults to ``False``.
"""
var = DynamicArrayVariable(
name=name,
dimensions=dimensions,
owner=self.owner,
device=self.device,
size=size,
dtype=dtype,
constant=constant,
needs_reference_update=needs_reference_update,
resize_along_first=resize_along_first,
scalar=scalar,
read_only=read_only,
unique=unique,
)
self._add_variable(name, var, index)
if np.prod(size) > 0:
self.device.resize(var, size)
if values is None and np.prod(size) > 0:
self.device.init_with_zeros(var, dtype)
elif values is not None:
if len(values) != size:
raise ValueError(
"Size of the provided values does not match "
f"size: {len(values)} != {size}"
)
if np.prod(size) > 0:
self.device.fill_with_array(var, values)
[docs]
def add_arange(
self,
name,
size,
start=0,
dtype=np.int32,
constant=True,
read_only=True,
unique=True,
index=None,
):
"""
Add an array, initialized with a range of integers.
Parameters
----------
name : str
The name of the variable.
size : int
The size of the array.
start : int
The start value of the range.
dtype : `dtype`, optional
The dtype used for storing the variable. If none is given, defaults
to `np.int32`.
constant : bool, optional
Whether the variable's value is constant during a run.
Defaults to ``True``.
read_only : bool, optional
Whether this is a read-only variable, i.e. a variable that is set
internally and cannot be changed by the user. Defaults
to ``True``.
index : str, optional
The index to use for this variable. Defaults to
`Variables.default_index`.
unique : bool, optional
See `ArrayVariable`. Defaults to ``True`` here.
"""
self.add_array(
name=name,
dimensions=DIMENSIONLESS,
size=size,
dtype=dtype,
constant=constant,
read_only=read_only,
unique=unique,
index=index,
)
self.device.init_with_arange(self._variables[name], start, dtype=dtype)
[docs]
def add_constant(self, name, value, dimensions=DIMENSIONLESS):
"""
Add a scalar constant (e.g. the number of neurons `N`).
Parameters
----------
name : str
The name of the variable
value: reference to the variable value
The value of the constant.
dimensions : `Dimension`, optional
The physical dimensions of the variable. Note that the variable
itself (as referenced by value) should never have units attached.
"""
var = Constant(name=name, dimensions=dimensions, owner=self.owner, value=value)
self._add_variable(name, var)
[docs]
def add_subexpression(
self, name, expr, dimensions=DIMENSIONLESS, dtype=None, scalar=False, index=None
):
"""
Add a named subexpression.
Parameters
----------
name : str
The name of the subexpression.
dimensions : `Dimension`
The physical dimensions of the subexpression.
expr : str
The subexpression itself.
dtype : `dtype`, optional
The dtype used for the expression. Defaults to
`core.default_float_dtype`.
scalar : bool, optional
Whether this is an expression only referring to scalar variables.
Defaults to ``False``
index : str, optional
The index to use for this variable. Defaults to
`Variables.default_index`.
"""
var = Subexpression(
name=name,
dimensions=dimensions,
expr=expr,
owner=self.owner,
dtype=dtype,
device=self.device,
scalar=scalar,
)
self._add_variable(name, var, index=index)
[docs]
def add_auxiliary_variable(
self, name, dimensions=DIMENSIONLESS, dtype=None, scalar=False
):
"""
Add an auxiliary variable (most likely one that is added automatically
to abstract code, e.g. ``_cond`` for a threshold condition),
specifying its type and unit for code generation.
Parameters
----------
name : str
The name of the variable
dimensions : `Dimension`
The physical dimensions of the variable.
dtype : `dtype`, optional
The dtype used for storing the variable. If none is given, defaults
to `core.default_float_dtype`.
scalar : bool, optional
Whether the variable is a scalar value (``True``) or vector-valued,
e.g. defined for every neuron (``False``). Defaults to ``False``.
"""
var = AuxiliaryVariable(
name=name, dimensions=dimensions, dtype=dtype, scalar=scalar
)
self._add_variable(name, var)
[docs]
def add_referred_subexpression(self, name, group, subexpr, index):
identifiers = subexpr.identifiers
substitutions = {}
for identifier in identifiers:
if identifier not in subexpr.owner.variables:
# external variable --> nothing to do
continue
subexpr_var = subexpr.owner.variables[identifier]
if hasattr(subexpr_var, "owner"):
new_name = f"_{name}_{subexpr.owner.name}_{identifier}"
else:
new_name = f"_{name}_{identifier}"
substitutions[identifier] = new_name
subexpr_var_index = group.variables.indices[identifier]
if subexpr_var_index == group.variables.default_index:
subexpr_var_index = index
elif subexpr_var_index == "0":
pass # nothing to do for a shared variable
elif subexpr_var_index == index:
pass # The same index as the main subexpression
elif index != self.default_index:
index_var = self._variables.get(index, None)
if isinstance(index_var, DynamicArrayVariable):
raise TypeError(
f"Cannot link to subexpression '{name}': it refers "
f"to the variable '{identifier}' which is indexed "
f"with the dynamic index '{subexpr_var_index}'."
)
else:
self.add_reference(subexpr_var_index, group)
self.indices[new_name] = subexpr_var_index
if isinstance(subexpr_var, Subexpression):
self.add_referred_subexpression(
new_name, group, subexpr_var, subexpr_var_index
)
else:
self.add_reference(new_name, group, identifier, subexpr_var_index)
new_expr = word_substitute(subexpr.expr, substitutions)
new_subexpr = Subexpression(
name,
self.owner,
new_expr,
dimensions=subexpr.dim,
device=subexpr.device,
dtype=subexpr.dtype,
scalar=subexpr.scalar,
)
self._variables[name] = new_subexpr
[docs]
def add_reference(self, name, group, varname=None, index=None):
"""
Add a reference to a variable defined somewhere else (possibly under
a different name). This is for example used in `Subgroup` and
`Synapses` to refer to variables in the respective `NeuronGroup`.
Parameters
----------
name : str
The name of the variable (in this group, possibly a different name
from `var.name`).
group : `Group`
The group from which `var` is referenced
varname : str, optional
The variable to refer to. If not given, defaults to `name`.
index : str, optional
The index that should be used for this variable (defaults to
`Variables.default_index`).
"""
if varname is None:
varname = name
if varname not in group.variables:
raise KeyError(f"Group {group.name} does not have a variable {varname}.")
if index is None:
if group.variables[varname].scalar:
index = "0"
else:
index = self.default_index
if (
self.owner is not None
and self.owner.name != group.name
and index in self.owner.variables
):
if (
not self.owner.variables[index].read_only
or isinstance(self.owner.variables[index], DynamicArrayVariable)
) and group.variables.indices[varname] != group.variables.default_index:
raise TypeError(
f"Cannot link variable '{name}' to '{varname}' in "
f"group '{group.name}' -- need to precalculate "
f"direct indices but index {index} can change"
)
# We don't overwrite existing names with references
if name not in self._variables:
var = group.variables[varname]
if isinstance(var, Subexpression):
self.add_referred_subexpression(name, group, var, index)
else:
self._variables[name] = var
self.indices[name] = index
[docs]
def add_references(self, group, varnames, index=None):
"""
Add all `Variable` objects from a name to `Variable` mapping with the
same name as in the original mapping.
Parameters
----------
group : `Group`
The group from which the `variables` are referenced
varnames : iterable of str
The variables that should be referred to in the current group
index : str, optional
The index to use for all the variables (defaults to
`Variables.default_index`)
"""
for name in varnames:
self.add_reference(name, group, name, index)
[docs]
def add_object(self, name, obj):
"""
Add an arbitrary Python object. This is only meant for internal use
and therefore only names starting with an underscore are allowed.
Parameters
----------
name : str
The name used for this object (has to start with an underscore).
obj : object
An arbitrary Python object that needs to be accessed directly from
a `CodeObject`.
"""
if not name.startswith("_"):
raise ValueError(
"This method is only meant for internally used "
"objects, the name therefore has to start with "
"an underscore"
)
self._variables[name] = obj
[docs]
def create_clock_variables(self, clock, prefix=""):
"""
Convenience function to add the ``t`` and ``dt`` attributes of a
`clock`.
Parameters
----------
clock : `Clock`
The clock that should be used for ``t`` and ``dt``.
prefix : str, optional
A prefix for the variable names. Used for example in monitors to
not confuse the dynamic array of recorded times with the current
time in the recorded group.
"""
self.add_reference(f"{prefix}t", clock, "t")
self.add_reference(f"{prefix}dt", clock, "dt")
self.add_reference(f"{prefix}t_in_timesteps", clock, "timestep")