def setUpClass(cls):
"""
Compile the network for this test
"""
def my_diagonal(pre, post, weight):
synapses = CSR()
for post_rk in post.ranks:
pre_ranks = []
delays = []
if post_rk - 1 in pre.ranks:
pre_ranks.append(post_rk - 1)
if post_rk in pre.ranks:
pre_ranks.append(post_rk)
if post_rk + 1 in pre.ranks:
pre_ranks.append(post_rk + 1)
synapses.add(post_rk, pre_ranks, [weight] * len(pre_ranks),
[0] * len(pre_ranks))
return synapses
def my_diagonal_with_uniform_delay(pre, post, weight, delay):
synapses = CSR()
for post_rk in post.ranks:
pre_ranks = []
delays = []
if post_rk - 1 in pre.ranks:
pre_ranks.append(post_rk - 1)
if post_rk in pre.ranks:
pre_ranks.append(post_rk)
if post_rk + 1 in pre.ranks:
pre_ranks.append(post_rk + 1)
synapses.add(post_rk, pre_ranks, [weight] * len(pre_ranks),
[delay] * len(pre_ranks))
return synapses
def my_diagonal_with_non_uniform_delay(pre, post, weight, delay):
synapses = CSR()
for post_rk in post.ranks:
pre_ranks = []
delays = []
if post_rk - 1 in pre.ranks:
pre_ranks.append(post_rk - 1)
if post_rk in pre.ranks:
pre_ranks.append(post_rk)
if post_rk + 1 in pre.ranks:
pre_ranks.append(post_rk + 1)
synapses.add(post_rk, pre_ranks, [weight] * len(pre_ranks),
delay.get_values(len(pre_ranks)))
return synapses
neuron = Neuron(equations="r = 1")
neuron2 = Neuron(equations="r = sum(exc)")
pop1 = Population(5, neuron)
pop2 = Population(5, neuron2)
proj1 = Projection(pre=pop1, post=pop2, target="exc")
proj1.connect_with_func(method=my_diagonal, weight=0.1)
proj2 = Projection(pre=pop1, post=pop2, target="exc2")
proj2.connect_with_func(method=my_diagonal_with_uniform_delay,
weight=0.1,
delay=2)
proj3 = Projection(pre=pop1, post=pop2, target="exc3")
proj3.connect_with_func(method=my_diagonal_with_non_uniform_delay,
weight=0.1,
delay=DiscreteUniform(1, 5))
cls.test_net = Network()
cls.test_net.add([pop1, pop2, proj1, proj2, proj3])
cls.test_net.compile(silent=True)
cls.test_proj1 = cls.test_net.get(proj1)
cls.test_proj2 = cls.test_net.get(proj2)
cls.test_proj3 = cls.test_net.get(proj3)
# A difference of Gaussians is used as weight wPos = Gaussian2D(1.0, [13, 13], [3, 3]) wNeg = Gaussian2D(2.0, [13, 13], [1.0, 1.0]) wDoG = (positive(wPos - wNeg) / np.sum(positive(wPos - wNeg)))[:, :, None] #V4L23_V4L4SUR = Convolution(V4L23, V4L4, target='S_SUR')####################################NEW #V4L23_V4L4SUR.connect_filter(weights=wDoG, subsampling=ssList24, keep_last_dimension=True)##NEW ## Connection from V4 L2/3 to FEF visual (excitatory) # The auxiliary population is used to pool the down-sampled V4L23 Population. # Afterwards it could be up-sampled again. The combination of the two is # currently not possible in ANNarchy ssList2v = ssList24[9::params['V4L4_shape'][-1]] V4L23_AuxE = Pooling(V4L23, AuxE, target='exc', operation='max') V4L23_AuxE.connect_pooling(extent=(1, 1) + params['PFC_shape']) AuxE_FEFv = Projection(AuxE, FEFv, target='exc', synapse=StandardSynapse) AuxE_FEFv.connect_with_func(con_scale, factor=2, delays=params['FEFv_delay']) ## Connections from FEF visual to FEF visuo-motoric(excitatory and suppressive) # A lowered Gaussian is used to simulate the combined responses G = Gaussian2D(1.0, changed['RFsizev_vm'], changed['RFsigmav_vm']) v_vm_shape = (params['FEFvm_shape'][-1], 1, 1) wvvm = np.tile((G - params['vSv2'])[None, :, :], v_vm_shape) wvvm *= params['dogScalingFactor_FEFvm']**np.arange(6)[:, None, None] # The plus sign(+) is needed, so that wvvm will not be overwritten FEFv_FEFvmE = Convolution(FEFv, FEFvm, target='E_v') FEFv_FEFvmE.connect_filters(weights=positive(+wvvm)) FEFv_FEFvmS = Convolution(FEFv, FEFvm, target='S_v') FEFv_FEFvmS.connect_filters(weights=positive(-wvvm)) ## Connection from FEF visuo-motoric to V4 L4 (amplification) # The auxiliary population is used to pool FEFvm activities over different