def setUpClass(cls):
"""
Compile the network for this test
"""
# neuron defintions common used for test cases
local_eq = Neuron(
equations="""
noise = Uniform(0,1)
r = t
"""
)
global_eq = Neuron(
equations="""
noise = Uniform(0,1) : population
glob_r = t : population
r = t
"""
)
mixed_eq = Neuron(
parameters="glob_par = 1.0: population",
equations="""
r = t + glob_par
"""
)
bound_eq = Neuron(
parameters="""
min_r=1.0: population
max_r=3.0: population
""",
equations="""
r = t : min=min_r, max=max_r
"""
)
tc_loc_up_pop = Population(3, local_eq)
tc_glob_up_pop = Population(3, global_eq)
tc_mixed_up_pop = Population(3, mixed_eq)
tc_bound_up_pop = Population(3, bound_eq)
m = Monitor(tc_bound_up_pop, 'r')
cls.test_net = Network()
cls.test_net.add([tc_loc_up_pop, tc_glob_up_pop,
tc_mixed_up_pop, tc_bound_up_pop, m])
cls.test_net.compile(silent=True)
cls.net_loc_pop = cls.test_net.get(tc_loc_up_pop)
cls.net_glob_pop = cls.test_net.get(tc_glob_up_pop)
cls.net_mix_pop = cls.test_net.get(tc_mixed_up_pop)
cls.net_bound_pop = cls.test_net.get(tc_bound_up_pop)
cls.net_m = cls.test_net.get(m)
def setUpClass(self):
"""
Compile the network for this test. Adapted the example
from documentation.
"""
SimpleSpike = Neuron(equations="mp=g_exc", spike="mp >= 1.0", reset="")
inp = Population(1, neuron=Neuron(equations="r=sin(t)"))
out = Population(1, neuron=SimpleSpike)
m = Monitor(out, "mp")
proj = CurrentInjection(inp, out, 'exc')
proj.connect_current()
self.test_net = Network()
self.test_net.add([inp, out, proj, m])
self.test_net.compile(silent=True)
self.output = self.test_net.get(out)
self.m = self.test_net.get(m)
def setUpClass(cls):
"""
Build up the network
"""
simple_emit = Neuron(spike="t==1", )
simple_recv = Neuron(equations="""
g_exc1 = 0
g_exc2 = 0
g_exc3 = 0
""",
spike="g_exc1>30")
# simple in/out populations
in_pop = Population(5, neuron=simple_emit)
out_pop = Population(2, neuron=simple_recv)
# create the projections for the test cases (TC)
# TC: no delay
proj = Projection(pre=in_pop, post=out_pop, target="exc1")
proj.connect_all_to_all(weights=1.0, storage_format="csr")
# TC: uniform delay
proj_u = Projection(pre=in_pop, post=out_pop, target="exc2")
proj_u.connect_all_to_all(weights=1.0,
delays=2.0,
storage_format="csr")
# TC: non-uniform delay
proj_nu = Projection(pre=in_pop, post=out_pop, target="exc3")
proj_nu.connect_all_to_all(weights=1.0, delays=Uniform(2, 10))
# Monitor to record the currents
m = Monitor(out_pop, ["g_exc1", "g_exc2", "g_exc3"])
# build network and store required object
# instances
net = Network()
net.add([in_pop, out_pop, proj, proj_u, proj_nu, m])
cls.test_net = net
cls.test_net.compile(silent=True)
cls.test_g_exc_m = net.get(m)
cls.test_proj = net.get(proj_nu)
def setUpClass(self):
"""
Compile the network for this test
"""
BuiltinFuncs = Neuron(parameters="""
base = 2.0
""",
equations="""
r = modulo(t,3)
pr = power(base,3)
clip_below = clip(-2, -1, 1)
clip_within = clip(0, -1, 1)
clip_above = clip(2, -1, 1)
""")
pop1 = Population(1, BuiltinFuncs)
mon = Monitor(pop1,
['r', 'pr', 'clip_below', 'clip_within', 'clip_above'])
self.test_net = Network()
self.test_net.add([pop1, mon])
self.test_net.compile(silent=True)
self.test_mon = self.test_net.get(mon)
v_tmp = v/(2*pi) : int
v_new = (v/(2*pi) - v_tmp)*2*pi
""",
spike="(v_new > pi)*(v_old < pi)")
# population setup
pop1 = Population(N, neuron=Theta, name="ThetaPop1")
pop1.eta = eta + D * np.tan(
(np.pi / 2) * (2 * np.arange(1, N + 1) - N - 1) / (N + 1))
# projection setup
proj = Projection(pre=pop1, post=pop1, target='exc', name='fb')
proj.connect_all_to_all(100.0 / N, allow_self_connections=False)
# monitoring
obs = Monitor(pop1, variables=['spike', 'v_new'], start=True, period=0.01)
# simulation
compile()
simulate(duration=T)
# conversion to pyrates
theta = pyrates_from_annarchy(monitors=[obs],
vars=['v_new'],
pop_average=False)
rate = pyrates_from_annarchy(monitors=[obs], vars=['spike'], pop_average=True)
plt.plot(theta)
plt.figure()
plt.plot(rate)
plt.show()
# input
#steps = int(T/dt)
#I_e_tmp = 5.0 + np.random.randn(steps, Ne) * 50.0 * np.sqrt(dt) # input current for excitatory neurons
#I_i_tmp = 4.0 + np.random.randn(steps, Ni) * 44.0 * np.sqrt(dt) # input current for inhibitory neurons
#I_e = TimedArray(rates=I_e_tmp, name="E_inp")
#I_i = TimedArray(rates=I_i_tmp, name="I_inp")
#inp_e = Projection(pre=I_e, post=E, target='exc')
#inp_i = Projection(pre=I_i, post=I, target='exc')
#inp_e.connect_one_to_one(1.0)
#inp_i.connect_one_to_one(1.0)
E.i_offset = 5.0
I.i_offset = 2.0
# monitoring
obs_e = Monitor(E, variables=['spike', 'v'], start=True)
obs_i = Monitor(I, variables=['spike', 'v'], start=True)
# simulation
############
# annarchy simulation
net = Network(everything=True)
net.compile()
net.simulate(duration=T)
# conversion to pyrates
rate_e = pyrates_from_annarchy(monitors=[net.get(obs_e)],
vars=['spike'],
pop_average=True)
rate_i = pyrates_from_annarchy(monitors=[net.get(obs_i)],
def grid_search_annarchy(param_grid: dict, param_map: dict, dt: float, simulation_time: float,
inputs: dict, outputs: dict, sampling_step_size: Optional[float] = None,
permute_grid: bool = False, circuit=None, **kwargs) -> DataFrame:
"""Function that runs multiple parametrizations of the same circuit in parallel and returns a combined output.
Parameters
----------
param_grid
Key-value pairs for each circuit parameter that should be altered over different circuit parametrizations.
param_map
Key-value pairs that map the keys of param_grid to concrete circuit variables.
dt
Simulation step-size in s.
simulation_time
Simulation time in s.
inputs
Inputs as provided to the `run` method of `:class:ComputeGraph`.
outputs
Outputs as provided to the `run` method of `:class:ComputeGraph`.
sampling_step_size
Sampling step-size as provided to the `run` method of `:class:ComputeGraph`.
permute_grid
If true, all combinations of the provided param_grid values will be realized. If false, the param_grid values
will be traversed pairwise.
circuit
Instance of ANNarchy network.
kwargs
Additional keyword arguments passed to the `:class:ComputeGraph` initialization.
Returns
-------
DataFrame
Simulation results stored in a multi-index data frame where each index lvl refers to one of the parameters of
param_grid.
"""
from ANNarchy import Population, Projection, Network, TimedArray, Monitor, ANNarchyException
# linearize parameter grid if necessary
if type(param_grid) is dict:
param_grid = linearize_grid(param_grid, permute_grid)
# create annarchy net if necessary
if circuit is None:
circuit = Network(everything=True)
# assign parameter updates to each circuit and combine them to unconnected network
circuit_names = []
param_info = []
param_split = "__"
val_split = "--"
comb = "_"
populations, projections = {}, {}
for n in range(param_grid.shape[0]):
# copy and re-parametrize populations
try:
for p in circuit.get_populations():
name = f'net{n}/{p.name}'
p_new = Population(geometry=p.geometry, neuron=p.neuron_type, name=name,
stop_condition=p.stop_condition, storage_order=p._storage_order,
copied=False)
p_new = adapt_pop(p_new, param_grid.iloc[n, :], param_map)
populations[name] = p_new
# add input to population
for node, inp in inputs.items():
if node in name:
inp_name = f'{name}_inp'
inp = TimedArray(rates=inp, name=inp_name)
proj = Projection(pre=inp, post=p_new, target='exc')
proj.connect_one_to_one(1.0)
populations[inp_name] = inp
projections[inp_name] = proj
except ANNarchyException:
pass
# copy and re-parametrize projections
try:
for c in circuit.get_projections():
source = c.pre if type(c.pre) is str else c.pre.name
target = c.post if type(c.post) is str else c.post.name
source = f'net{n}/{source}'
target = f'net{n}/{target}'
name = f'{source}/{target}/{c.name}'
c_new = Projection(pre=source, post=target, target=c.target, synapse=c.synapse_type, name=name,
copied=False)
c_new._store_connectivity(c._connection_method, c._connection_args, c._connection_delay, c._storage_format)
c_new = adapt_proj(c_new, param_grid.iloc[n, :], param_map)
projections[name] = c_new
except ANNarchyException:
pass
# collect parameter and circuit name infos
circuit_names.append(f'net{n}')
param_names = list(param_grid.columns.values)
param_info_tmp = [f"{param_names[i]}{val_split}{val}" for i, val in enumerate(param_grid.iloc[n, :])]
param_info.append(param_split.join(param_info_tmp))
net = Network()
for p in populations.values():
net.add(p)
for c in projections.values():
net.add(c)
# adjust output of simulation to combined network
nodes = [p.name for p in circuit.get_populations()]
out_names, var_names, out_lens, monitors, monitor_names = [], [], [], [], []
for out_key, out in outputs.copy().items():
out_names_tmp, out_lens_tmp = [], []
if out[0] in nodes:
for i, name in enumerate(param_info):
out_tmp = list(out)
out_tmp[0] = f'{circuit_names[i]}/{out_tmp[0]}'
p = net.get_population(out_tmp[0])
monitors.append(Monitor(p, variables=out_tmp[-1], period=sampling_step_size, start=True,
net_id=net.id))
monitor_names.append(f'{name}{param_split}out_var{val_split}{out_key}{comb}{out[0]}')
var_names.append(out_tmp[-1])
out_names_tmp.append(f'{out_key}{comb}{out[0]}')
out_lens_tmp.append(p.geometry[0])
elif out[0] == 'all':
for node in nodes:
for i, name in enumerate(param_info):
out_tmp = list(out)
out_tmp[0] = f'{circuit_names[i]}/{node}'
p = net.get_population(out_tmp[0])
monitors.append(Monitor(p, variables=out_tmp[-1], period=sampling_step_size, start=True,
net_id=net.id))
monitor_names.append(f'{name}{param_split}out_var{val_split}{out_key}{comb}{node}')
var_names.append(out_tmp[-1])
out_names_tmp.append(f'{out_key}{comb}{node}')
out_lens_tmp.append(p.geometry[0])
else:
node_found = False
for node in nodes:
if out[0] in node:
node_found = True
for i, name in enumerate(param_info):
out_tmp = list(out)
out_tmp[0] = f'{circuit_names[i]}/{node}'
p = net.get_population(out_tmp[0])
monitors.append(Monitor(p, variables=out_tmp[-1], period=sampling_step_size, start=True,
net_id=net.id))
monitor_names.append(f'{name}{param_split}out_var{val_split}{out_key}{comb}{node}')
var_names.append(out_tmp[-1])
out_names_tmp.append(f'{out_key}{comb}{node}')
out_lens_tmp.append(p.geometry[0])
if not node_found:
raise ValueError(f'Invalid output identifier in output: {out_key}. '
f'Node {out[0]} is not part of this network')
out_names += list(set(out_names_tmp))
out_lens += list(set(out_lens_tmp))
#net.add(monitors)
# simulate the circuits behavior
net.compile()
net.simulate(duration=simulation_time)
# transform output into pyrates-compatible data format
results = pyrates_from_annarchy(monitors, vars=list(set(var_names)),
monitor_names=monitor_names, **kwargs)
# transform results into long-form dataframe with changed parameters as columns
multi_idx = [param_grid[key].values for key in param_grid.keys()]
n_iters = len(multi_idx[0])
outs = []
for out_name, out_len in zip(out_names, out_lens):
outs += [f'{out_name}_n{i}' for i in range(out_len)] * n_iters
multi_idx_final = []
for idx in multi_idx:
for val in idx:
for out_len in out_lens:
multi_idx_final += [val]*len(out_names)*out_len
index = MultiIndex.from_arrays([multi_idx_final, outs], names=list(param_grid.keys()) + ["out_var"])
index = MultiIndex.from_tuples(list(set(index)), names=list(param_grid.keys()) + ["out_var"])
results_final = DataFrame(columns=index, data=np.zeros_like(results.values), index=results.index)
for col in results.keys():
params = col.split(param_split)
indices = [None] * len(results_final.columns.names)
for param in params:
var, val = param.split(val_split)[:2]
idx = list(results_final.columns.names).index(var)
try:
indices[idx] = float(val)
except ValueError:
indices[idx] = val
results_final.loc[:, tuple(indices)] = results[col].values
return results_final