import collections
import numpy as np
from nengo.builder import Builder, Signal
from nengo.builder.ensemble import gen_eval_points, get_activities
from nengo.builder.node import SimPyFunc
from nengo.builder.operator import Copy, ElementwiseInc, Reset
from nengo.connection import Connection
from nengo.transforms import Dense
from nengo.ensemble import Ensemble, Neurons
from nengo.exceptions import BuildError
from nengo.neurons import Direct
from nengo.node import Node
from nengo.rc import rc
from nengo.solvers import NoSolver, Solver
from nengo.utils.numpy import is_integer, is_iterable
built_attrs = ["eval_points", "solver_info", "weights", "transform"]
[docs]class BuiltConnection(collections.namedtuple("BuiltConnection", built_attrs)):
"""Collects the parameters generated in `.build_connection`.
These are stored here because in the majority of cases the equivalent
attribute in the original connection is a `.Distribution`. The attributes
of a BuiltConnection are the full NumPy arrays used in the simulation.
See the `.Connection` documentation for more details on each parameter.
Parameters
----------
eval_points : ndarray
Evaluation points.
solver_info : dict
Information dictionary returned by the `.Solver`.
weights : ndarray
Connection weights. May be synaptic connection weights defined in
the connection's transform, or a combination of the decoders
automatically solved for and the specified transform.
transform : ndarray
The transform matrix.
"""
__slots__ = ()
def __new__(cls, eval_points, solver_info, weights, transform):
# Overridden to suppress the default __new__ docstring
return tuple.__new__(cls, (eval_points, solver_info, weights, transform))
[docs]def get_eval_points(model, conn, rng):
"""Get evaluation points for connection."""
if conn.eval_points is None:
view = model.params[conn.pre_obj].eval_points.view()
view.setflags(write=False)
assert view.dtype == rc.float_dtype
return view
else:
return gen_eval_points(
conn.pre_obj,
conn.eval_points,
rng,
conn.scale_eval_points,
dtype=rc.float_dtype,
)
[docs]def get_targets(conn, eval_points, dtype=None):
"""Get target points for connection with given evaluation points."""
dtype = rc.float_dtype if dtype is None else dtype
if conn.function is None:
targets = eval_points[:, conn.pre_slice].astype(dtype)
elif isinstance(conn.function, np.ndarray):
targets = conn.function
else:
targets = np.zeros((len(eval_points), conn.size_mid), dtype=dtype)
for i, ep in enumerate(eval_points[:, conn.pre_slice]):
out = conn.function(ep)
if out is None:
raise BuildError(
"Building %s: Connection function returned "
"None. Cannot solve for decoders." % (conn,)
)
targets[i] = out
return targets
[docs]def build_linear_system(model, conn, rng):
"""Get all arrays needed to compute decoders."""
eval_points = get_eval_points(model, conn, rng)
ens = conn.pre_obj
activities = get_activities(model.params[ens], ens, eval_points)
if np.count_nonzero(activities) == 0:
raise BuildError(
"Building %s: 'activites' matrix is all zero for %s. "
"This is because no evaluation points fall in the firing "
"ranges of any neurons." % (conn, conn.pre_obj)
)
targets = get_targets(conn, eval_points, dtype=rc.float_dtype)
return eval_points, activities, targets
[docs]def build_decoders(model, conn, rng):
"""Compute decoders for connection."""
encoders = model.params[conn.pre_obj].encoders
gain = model.params[conn.pre_obj].gain
bias = model.params[conn.pre_obj].bias
eval_points = get_eval_points(model, conn, rng)
targets = get_targets(conn, eval_points, dtype=rc.float_dtype)
if conn.solver.weights and not conn.solver.compositional:
# solver is solving for the whole weight matrix, so apply
# transform/encoders to targets
if not isinstance(conn.transform, Dense):
raise BuildError(
"Non-compositional solvers only work with Dense transforms"
)
transform = conn.transform.sample(rng=rng)
targets = np.dot(targets, transform.T)
# weight solvers only allowed on ensemble->ensemble connections
assert isinstance(conn.post_obj, Ensemble)
post_enc = model.params[conn.post_obj].scaled_encoders
targets = np.dot(targets, post_enc.T[conn.post_slice])
x = np.dot(eval_points, encoders.T / conn.pre_obj.radius)
wrapped_solver = (
model.decoder_cache.wrap_solver(solve_for_decoders)
if model.seeded[conn]
else solve_for_decoders
)
decoders, solver_info = wrapped_solver(conn, gain, bias, x, targets, rng=rng)
return eval_points, decoders.T, solver_info
[docs]def solve_for_decoders(conn, gain, bias, x, targets, rng):
"""Solver for decoders.
Factored out from `.build_decoders` for use with the cache system.
"""
activities = conn.pre_obj.neuron_type.rates(x, gain, bias)
if np.count_nonzero(activities) == 0:
raise BuildError(
"Building %s: 'activities' matrix is all zero for %s. "
"This is because no evaluation points fall in the firing "
"ranges of any neurons." % (conn, conn.pre_obj)
)
decoders, solver_info = conn.solver(activities, targets, rng=rng)
return decoders, solver_info
[docs]def slice_signal(model, signal, sl):
"""Apply a slice operation to given signal."""
assert signal.ndim == 1
if isinstance(sl, slice) and (sl.step is None or sl.step == 1):
return signal[sl]
else:
size = np.arange(signal.size, dtype=rc.float_dtype)[sl].size
sliced_signal = Signal(shape=size, name="%s.sliced" % signal.name)
model.add_op(Copy(signal, sliced_signal, src_slice=sl))
return sliced_signal
[docs]@Builder.register(Solver)
def build_solver(model, solver, conn, rng):
"""Apply decoder solver to connection."""
return build_decoders(model, conn, rng)
[docs]@Builder.register(NoSolver)
def build_no_solver(model, solver, conn, rng):
"""Special builder for NoSolver to skip unnecessary steps."""
activities = np.zeros((1, conn.pre_obj.n_neurons), dtype=rc.float_dtype)
targets = np.zeros((1, conn.size_mid), dtype=rc.float_dtype)
# No need to invoke the cache for NoSolver
decoders, solver_info = conn.solver(activities, targets, rng=rng)
weights = decoders.T
return None, weights, solver_info
[docs]@Builder.register(Connection) # noqa: C901
def build_connection(model, conn):
"""Builds a `.Connection` object into a model.
A brief summary of what happens in the connection build process,
in order:
1. Solve for decoders.
2. Combine transform matrix with decoders to get weights.
3. Add operators for computing the function
or multiplying neural activity by weights.
4. Call build function for the synapse.
5. Call build function for the learning rule.
6. Add operator for applying learning rule delta to weights.
Some of these steps may be altered or omitted depending on the parameters
of the connection, in particular the pre and post types.
Parameters
----------
model : Model
The model to build into.
conn : Connection
The connection to build.
Notes
-----
Sets ``model.params[conn]`` to a `.BuiltConnection` instance.
"""
# Create random number generator
rng = np.random.RandomState(model.seeds[conn])
# Get input and output connections from pre and post
def get_prepost_signal(is_pre):
target = conn.pre_obj if is_pre else conn.post_obj
key = "out" if is_pre else "in"
if target not in model.sig:
raise BuildError(
"Building %s: the %r object %s is not in the "
"model, or has a size of zero."
% (conn, "pre" if is_pre else "post", target)
)
if key not in model.sig[target]:
raise BuildError(
"Building %s: the %r object %s has a %r size of zero."
% (conn, "pre" if is_pre else "post", target, key)
)
return model.sig[target][key]
model.sig[conn]["in"] = get_prepost_signal(is_pre=True)
model.sig[conn]["out"] = get_prepost_signal(is_pre=False)
decoders = None
encoders = None
eval_points = None
solver_info = None
post_slice = conn.post_slice
# Figure out the signal going across this connection
in_signal = model.sig[conn]["in"]
if isinstance(conn.pre_obj, Node) or (
isinstance(conn.pre_obj, Ensemble)
and isinstance(conn.pre_obj.neuron_type, Direct)
):
# Node or Decoded connection in directmode
sliced_in = slice_signal(model, in_signal, conn.pre_slice)
if conn.function is None:
in_signal = sliced_in
elif isinstance(conn.function, np.ndarray):
raise BuildError("Cannot use function points in direct connection")
else:
in_signal = Signal(shape=conn.size_mid, name="%s.func" % conn)
model.add_op(SimPyFunc(in_signal, conn.function, None, sliced_in))
elif isinstance(conn.pre_obj, Ensemble): # Normal decoded connection
eval_points, decoders, solver_info = model.build(conn.solver, conn, rng)
if conn.solver.weights:
model.sig[conn]["out"] = model.sig[conn.post_obj.neurons]["in"]
# weight solvers only allowed on ensemble->ensemble connections
assert isinstance(conn.post_obj, Ensemble)
encoders = model.params[conn.post_obj].scaled_encoders.T
encoders = encoders[conn.post_slice]
# post slice already applied to encoders (either here or in
# `build_decoders`), so don't apply later
post_slice = None
else:
in_signal = slice_signal(model, in_signal, conn.pre_slice)
# Build transform
if conn.solver.weights and not conn.solver.compositional:
# special case for non-compositional weight solvers, where
# the solver is solving for the full weight matrix. so we don't
# need to combine decoders/transform/encoders.
weighted, weights = model.build(
Dense(decoders.shape, init=decoders), in_signal, rng=rng
)
else:
weighted, weights = model.build(
conn.transform, in_signal, decoders=decoders, encoders=encoders, rng=rng
)
model.sig[conn]["weights"] = weights
# Build synapse
if conn.synapse is not None:
weighted = model.build(conn.synapse, weighted, mode="update")
# Store the weighted-filtered output in case we want to probe it
model.sig[conn]["weighted"] = weighted
if isinstance(conn.post_obj, Neurons):
# Apply neuron gains (we don't need to do this if we're connecting to
# an Ensemble, because the gains are rolled into the encoders)
gains = Signal(
model.params[conn.post_obj.ensemble].gain[post_slice],
name="%s.gains" % conn,
)
if is_integer(post_slice) or isinstance(post_slice, slice):
sliced_out = model.sig[conn]["out"][post_slice]
else:
# advanced indexing not supported on Signals, so we need to set up an
# intermediate signal and use a Copy op to perform the indexing
sliced_out = Signal(shape=gains.shape, name="%s.sliced_out" % conn)
model.add_op(Reset(sliced_out))
model.add_op(
Copy(sliced_out, model.sig[conn]["out"], dst_slice=post_slice, inc=True)
)
model.add_op(
ElementwiseInc(
gains, weighted, sliced_out, tag="%s.gains_elementwiseinc" % conn,
)
)
else:
# Copy to the proper slice
model.add_op(
Copy(
weighted,
model.sig[conn]["out"],
dst_slice=post_slice,
inc=True,
tag="%s" % conn,
)
)
# Build learning rules
if conn.learning_rule is not None:
# TODO: provide a general way for transforms to expose learnable params
if not isinstance(conn.transform, Dense):
raise NotImplementedError(
"Learning on connections with %s transforms is not supported"
% (type(conn.transform).__name__)
)
rule = conn.learning_rule
rule = [rule] if not is_iterable(rule) else rule
targets = []
for r in rule.values() if isinstance(rule, dict) else rule:
model.build(r)
targets.append(r.modifies)
if "encoders" in targets:
encoder_sig = model.sig[conn.post_obj]["encoders"]
encoder_sig.readonly = False
if "decoders" in targets or "weights" in targets:
if weights.ndim < 2:
raise BuildError(
"'transform' must be a 2-dimensional array for learning"
)
model.sig[conn]["weights"].readonly = False
model.params[conn] = BuiltConnection(
eval_points=eval_points,
solver_info=solver_info,
transform=conn.transform,
weights=weights.initial_value,
)
