class PruningIllustration(object):
def __init__(self):
if os.path.dirname(__file__):
os.chdir(os.path.dirname(__file__))
self.network_path = "pruning_illustration_network"
self.config_file = os.path.join(self.network_path, "network-config.json")
self.position_file = os.path.join(self.network_path, "network-neuron-positions.hdf5")
self.save_file = os.path.join(self.network_path, "voxels", "network-synapses.hdf5")
create_cube_mesh(file_name=os.path.join(self.network_path, "mesh", "simple_mesh.obj"),
centre_point=(0, 0, 0),
side_len=500e-6)
sp = SnuddaPlace(config_file=self.config_file, d_view=None)
print("Calling read_config")
sp.parse_config()
print("Read done")
sp.write_data(self.position_file)
# We want to load in the ball and stick neuron that has 20 micrometer soma diameter, and axon (along y-axis),
# and dendrite along (x-axis) out to 200 micrometer distance from centre of soma.
self.sd = SnuddaDetect(config_file=self.config_file, position_file=self.position_file,
save_file=self.save_file, rc=None,
hyper_voxel_size=150)
# Reposition the neurons so we know how many synapses and where they will be located before pruning
neuron_positions = np.array([[0, 59, 0], # Postsynaptiska
[0, 89, 0],
[0, 119, 0],
[0, 149, 0],
[0, 179, 0],
[0, 209, 0],
[0, 239, 0],
[0, 269, 0],
[0, 299, 0],
[0, 329, 0],
[59, 0, 0], # Presynaptiska
[89, 0, 0],
[119, 0, 0],
[149, 0, 0],
[179, 0, 0],
[209, 0, 0],
[239, 0, 0],
[269, 0, 0],
[299, 0, 0],
[329, 0, 0],
]) * 1e-6
# TODO: Add potential for gap junctions also by having 5 + 5 neurons in other grid
for idx, pos in enumerate(neuron_positions):
self.sd.neurons[idx]["position"] = pos
ang = -np.pi / 2
R_x = np.array([[1, 0, 0],
[0, np.cos(ang), -np.sin(ang)],
[0, np.sin(ang), np.cos(ang)]])
ang = np.pi / 2
R_y = np.array([[np.cos(ang), 0, np.sin(ang)],
[0, 1, 0],
[-np.sin(ang), 0, np.cos(ang)]])
for idx in range(0, 10): # Post synaptic neurons
self.sd.neurons[idx]["rotation"] = R_x
for idx in range(10, 20): # Presynaptic neurons
self.sd.neurons[idx]["rotation"] = R_y
self.sd.detect(restart_detection_flag=True)
# Also update so that the new positions are saved in the place file
rn = RepositionNeurons(self.position_file)
for neuron_info in self.sd.neurons:
rn.place(neuron_info["neuronID"], position=neuron_info["position"], rotation=neuron_info["rotation"],
verbose=False)
rn.close()
if False:
self.sd.process_hyper_voxel(1)
plt, ax = self.sd.plot_hyper_voxel(plot_neurons=True, elev_azim=(90, 0),
draw_axon_voxels=False, draw_dendrite_voxels=False,
draw_axons=True, draw_dendrites=True,
show_axis=False, title="No pruning",
fig_file_name="Pruning-fig-1-no-pruning")
import pdb
pdb.set_trace()
def prune_network(self, pruning_config=None, fig_name=None, title=None):
work_log = os.path.join(self.network_path, "log", "network-detect-worklog.hdf5")
pruned_output = os.path.join(self.network_path, "network-synapses.hdf5")
if pruning_config is not None and not os.path.exists(pruning_config):
pruning_config = os.path.join(self.network_path, pruning_config)
sp = SnuddaPrune(network_path=self.network_path, config_file=pruning_config) # Use default config file
sp.prune(pre_merge_only=False)
sp = []
plot_axon = True
plot_dendrite = True
#plot_axon = np.ones((20,), dtype=bool)
#plot_dendrite = np.ones((20,), dtype=bool)
#plot_axon[:10] = False
#plot_dendrite[10:] = False
pn = PlotNetwork(pruned_output)
plt, ax = pn.plot(fig_name=fig_name, show_axis=False,
plot_axon=plot_axon, plot_dendrite=plot_dendrite,
title=title, title_pad=-14,
elev_azim=(90, 0))
class TouchDetectionHypervoxelIllustration(object):
def __init__(self):
if os.path.dirname(__file__):
os.chdir(os.path.dirname(__file__))
self.network_path = "touch_detection_hypervoxel_illustration_network"
self.config_file = os.path.join(self.network_path, "network-config.json")
self.position_file = os.path.join(self.network_path, "network-neuron-positions.hdf5")
self.save_file = os.path.join(self.network_path, "voxels", "network-putative-synapses.hdf5")
create_cube_mesh(file_name=os.path.join(self.network_path, "mesh", "simple_mesh.obj"),
centre_point=(0, 0, 0),
side_len=500e-6)
sp = SnuddaPlace(config_file=self.config_file, d_view=None)
sp.parse_config()
sp.write_data(self.position_file)
self.sd = SnuddaDetect(config_file=self.config_file, position_file=self.position_file,
save_file=self.save_file, rc=None,
hyper_voxel_size=60)
neuron_positions = np.array([[10, 30, 70], # Postsynaptiska
[50, 60, 70], # Presynaptiska
]) * 1e-6
for idx, pos in enumerate(neuron_positions):
self.sd.neurons[idx]["position"] = pos
ang = -np.pi / 2
R_x = np.array([[1, 0, 0],
[0, np.cos(ang), -np.sin(ang)],
[0, np.sin(ang), np.cos(ang)]])
ang = np.pi * 0.2
R_y = np.array([[np.cos(ang), 0, np.sin(ang)],
[0, 1, 0],
[-np.sin(ang), 0, np.cos(ang)]])
ang = np.pi * (0.5 + 0.2)
R_z0 = np.array([[np.cos(ang), -np.sin(ang), 0],
[np.sin(ang), np.cos(ang), 0],
[0, 0, 1]])
ang = np.pi*0.4
R_z1 = np.array([[np.cos(ang), -np.sin(ang), 0],
[np.sin(ang), np.cos(ang), 0],
[0, 0, 1]])
# Post synaptic
self.sd.neurons[0]["rotation"] = R_z0
# Presynaptic neuron
self.sd.neurons[1]["rotation"] = np.matmul(R_z1, R_y)
self.sd.detect(restart_detection_flag=True)
self.sd.process_hyper_voxel(0)
plt, ax = self.sd.plot_hyper_voxel(plot_neurons=False,
draw_axon_voxels=True,
draw_dendrite_voxels=True,
elev_azim=(50, -22),
title="",
fig_file_name="touch_detection_illustration-voxels.pdf",
dpi=300)
plt, ax = self.sd.plot_hyper_voxel(plot_neurons=True,
draw_axon_voxels=False,
draw_dendrite_voxels=False,
elev_azim=(50, -22),
title="",
fig_file_name="touch_detection_illustration-morph.pdf",
dpi=300)
print(f"\n--> Figures written to {self.sd.network_path}/figures")
class TestPrune(unittest.TestCase):
def setUp(self):
if os.path.dirname(__file__):
os.chdir(os.path.dirname(__file__))
self.network_path = os.path.join(os.path.dirname(__file__), "networks",
"network_testing_prune3")
create_cube_mesh(file_name=os.path.join(self.network_path, "mesh",
"simple_mesh.obj"),
centre_point=(0, 0, 0),
side_len=500e-6)
config_file = os.path.join(self.network_path, "network-config.json")
position_file = os.path.join(self.network_path,
"network-neuron-positions.hdf5")
save_file = os.path.join(self.network_path, "voxels",
"network-putative-synapses.hdf5")
sp = SnuddaPlace(config_file=config_file, d_view=None, verbose=True)
sp.parse_config()
sp.write_data(position_file)
# We want to load in the ball and stick neuron that has 20 micrometer soma diameter, and axon (along y-axis),
# and dendrite along (x-axis) out to 100 micrometer distance from centre of soma.
self.sd = SnuddaDetect(config_file=config_file,
position_file=position_file,
save_file=save_file,
rc=None,
hyper_voxel_size=120,
verbose=True)
# Reposition the neurons so we know how many synapses and where they will be located before pruning
neuron_positions = np.array([
[0, 20, 0], # Postsynaptiska
[0, 40, 0],
[0, 60, 0],
[0, 80, 0],
[0, 100, 0],
[0, 120, 0],
[0, 140, 0],
[0, 160, 0],
[0, 180, 0],
[0, 200, 0],
[20, 0, 0], # Presynaptiska
[40, 0, 0],
[60, 0, 0],
[80, 0, 0],
[100, 0, 0],
[120, 0, 0],
[140, 0, 0],
[160, 0, 0],
[180, 0, 0],
[200, 0, 0],
[70, 0, 500], # For gap junction check
[110, 0, 500],
[150, 0, 500],
[190, 0, 500],
[0, 70, 500],
[0, 110, 500],
[0, 150, 500],
[0, 190, 500],
]) * 1e-6
# TODO: Add potential for gap junctions also by having 5 + 5 neurons in other grid
for idx, pos in enumerate(neuron_positions):
self.sd.neurons[idx]["position"] = pos
ang = -np.pi / 2
R_x = np.array([[1, 0, 0], [0, np.cos(ang), -np.sin(ang)],
[0, np.sin(ang), np.cos(ang)]])
ang = np.pi / 2
R_y = np.array([[np.cos(ang), 0, np.sin(ang)], [0, 1, 0],
[-np.sin(ang), 0, np.cos(ang)]])
for idx in range(0, 10): # Post synaptic neurons
self.sd.neurons[idx]["rotation"] = R_x
for idx in range(10, 20): # Presynaptic neurons
self.sd.neurons[idx]["rotation"] = R_y
for idx in range(24, 28): # GJ neurons
self.sd.neurons[idx]["rotation"] = R_x
ang = np.pi / 2
R_z = np.array([[np.cos(ang), -np.sin(ang), 0],
[np.sin(ang), np.cos(ang), 0], [0, 0, 1]])
for idx in range(20, 24): # GJ neurons
self.sd.neurons[idx]["rotation"] = np.matmul(R_z, R_x)
self.sd.detect(restart_detection_flag=True)
if False:
self.sd.process_hyper_voxel(1)
self.sd.plot_hyper_voxel(plot_neurons=True)
def test_prune(self):
pruned_output = os.path.join(self.network_path,
"network-synapses.hdf5")
with self.subTest(stage="No-pruning"):
sp = SnuddaPrune(network_path=self.network_path,
config_file=None,
verbose=True,
keep_files=True) # Use default config file
sp.prune()
sp = []
# Load the pruned data and check it
sl = SnuddaLoad(pruned_output)
# TODO: Call a plot function to plot entire network with synapses and all
self.assertEqual(sl.data["nSynapses"], (20 * 8 + 10 * 2) *
2) # Update, now AMPA+GABA, hence *2 at end
# This checks that all synapses are in order
# The synapse sort order is destID, sourceID, synapsetype (channel model id).
syn = sl.data["synapses"][:sl.data["nSynapses"], :]
syn_order = (syn[:, 1] * len(self.sd.neurons) + syn[:, 0]
) * 12 + syn[:, 6] # The 12 is maxChannelModelID
self.assertTrue((np.diff(syn_order) >= 0).all())
# Note that channel model id is dynamically allocated, starting from 10 (GJ have ID 3)
# Check that correct number of each type
self.assertEqual(np.sum(sl.data["synapses"][:, 6] == 10),
20 * 8 + 10 * 2)
self.assertEqual(np.sum(sl.data["synapses"][:, 6] == 11),
20 * 8 + 10 * 2)
self.assertEqual(sl.data["nGapJunctions"], 4 * 4 * 4)
gj = sl.data["gapJunctions"][:sl.data["nGapJunctions"], :2]
gj_order = gj[:, 1] * len(self.sd.neurons) + gj[:, 0]
self.assertTrue((np.diff(gj_order) >= 0).all())
with self.subTest(stage="load-testing"):
sl = SnuddaLoad(pruned_output, verbose=True)
# Try and load a neuron
n = sl.load_neuron(0)
self.assertTrue(type(n) == NeuronMorphology)
syn_ctr = 0
for s in sl.synapse_iterator(chunk_size=50):
syn_ctr += s.shape[0]
self.assertEqual(syn_ctr, sl.data["nSynapses"])
gj_ctr = 0
for gj in sl.gap_junction_iterator(chunk_size=50):
gj_ctr += gj.shape[0]
self.assertEqual(gj_ctr, sl.data["nGapJunctions"])
syn, syn_coords = sl.find_synapses(pre_id=14)
self.assertTrue((syn[:, 0] == 14).all())
self.assertEqual(syn.shape[0], 40)
syn, syn_coords = sl.find_synapses(post_id=3)
self.assertTrue((syn[:, 1] == 3).all())
self.assertEqual(syn.shape[0], 36)
cell_id_perm = sl.get_cell_id_of_type("ballanddoublestick",
random_permute=True,
num_neurons=28)
cell_id = sl.get_cell_id_of_type("ballanddoublestick",
random_permute=False)
self.assertEqual(len(cell_id_perm), 28)
self.assertEqual(len(cell_id), 28)
for cid in cell_id_perm:
self.assertTrue(cid in cell_id)
# It is important merge file has synapses sorted with dest_id, source_id as sort order since during pruning
# we assume this to be able to quickly find all synapses on post synaptic cell.
# TODO: Also include the ChannelModelID in sorting check
with self.subTest("Checking-merge-file-sorted"):
for mf in [
"temp/synapses-for-neurons-0-to-28-MERGE-ME.hdf5",
"temp/gapJunctions-for-neurons-0-to-28-MERGE-ME.hdf5",
"network-synapses.hdf5"
]:
merge_file = os.path.join(self.network_path, mf)
sl = SnuddaLoad(merge_file, verbose=True)
if "synapses" in sl.data:
syn = sl.data["synapses"][:sl.data["nSynapses"], :2]
syn_order = syn[:, 1] * len(self.sd.neurons) + syn[:, 0]
self.assertTrue((np.diff(syn_order) >= 0).all())
if "gapJunctions" in sl.data:
gj = sl.data["gapJunctions"][:sl.data["nGapJunctions"], :2]
gj_order = gj[:, 1] * len(self.sd.neurons) + gj[:, 0]
self.assertTrue((np.diff(gj_order) >= 0).all())
with self.subTest("synapse-f1"):
# Test of f1
testing_config_file = os.path.join(self.network_path,
"network-config-test-1.json")
sp = SnuddaPrune(network_path=self.network_path,
config_file=testing_config_file,
verbose=True,
keep_files=True) # Use default config file
sp.prune()
# Load the pruned data and check it
sl = SnuddaLoad(pruned_output, verbose=True)
# Setting f1=0.5 in config should remove 50% of GABA synapses, but does so randomly, for AMPA we used f1=0.9
gaba_id = sl.data["connectivityDistributions"][
"ballanddoublestick",
"ballanddoublestick"]["GABA"]["channelModelID"]
ampa_id = sl.data["connectivityDistributions"][
"ballanddoublestick",
"ballanddoublestick"]["AMPA"]["channelModelID"]
n_gaba = np.sum(sl.data["synapses"][:, 6] == gaba_id)
n_ampa = np.sum(sl.data["synapses"][:, 6] == ampa_id)
self.assertTrue((20 * 8 + 10 * 2) * 0.5 -
10 < n_gaba < (20 * 8 + 10 * 2) * 0.5 + 10)
self.assertTrue((20 * 8 + 10 * 2) * 0.9 -
10 < n_ampa < (20 * 8 + 10 * 2) * 0.9 + 10)
with self.subTest("synapse-softmax"):
# Test of softmax
testing_config_file = os.path.join(
self.network_path, "network-config-test-2.json"
) # Only GABA synapses in this config
sp = SnuddaPrune(network_path=self.network_path,
config_file=testing_config_file,
verbose=True,
keep_files=True) # Use default config file
sp.prune()
# Load the pruned data and check it
sl = SnuddaLoad(pruned_output)
# Softmax reduces number of synapses
self.assertTrue(sl.data["nSynapses"] < 20 * 8 + 10 * 2)
with self.subTest("synapse-mu2"):
# Test of mu2
testing_config_file = os.path.join(self.network_path,
"network-config-test-3.json")
sp = SnuddaPrune(network_path=self.network_path,
config_file=testing_config_file,
verbose=True,
keep_files=True) # Use default config file
sp.prune()
# Load the pruned data and check it
sl = SnuddaLoad(pruned_output)
# With mu2 having 2 synapses means 50% chance to keep them, having 1 will be likely to have it removed
self.assertTrue(
20 * 8 * 0.5 - 10 < sl.data["nSynapses"] < 20 * 8 * 0.5 + 10)
with self.subTest("synapse-a3"):
# Test of a3
testing_config_file = os.path.join(self.network_path,
"network-config-test-4.json")
sp = SnuddaPrune(network_path=self.network_path,
config_file=testing_config_file,
verbose=True,
keep_files=True) # Use default config file
sp.prune()
# Load the pruned data and check it
sl = SnuddaLoad(pruned_output)
# a3=0.6 means 40% chance to remove all synapses between a pair
self.assertTrue(
(20 * 8 + 10 * 2) * 0.6 -
14 < sl.data["nSynapses"] < (20 * 8 + 10 * 2) * 0.6 + 14)
with self.subTest("synapse-distance-dependent-pruning"):
# Testing distance dependent pruning
testing_config_file = os.path.join(self.network_path,
"network-config-test-5.json")
sp = SnuddaPrune(network_path=self.network_path,
config_file=testing_config_file,
verbose=True,
keep_files=True) # Use default config file
sp.prune()
# Load the pruned data and check it
sl = SnuddaLoad(pruned_output)
# "1*(d >= 100e-6)" means we remove all synapses closer than 100 micrometers
self.assertEqual(sl.data["nSynapses"], 20 * 6)
self.assertTrue(
(sl.data["synapses"][:, 8] >=
100).all()) # Column 8 -- distance to soma in micrometers
# TODO: Need to do same test for Gap Junctions also -- but should be same results, since same codebase
with self.subTest("gap-junction-f1"):
# Test of f1
testing_config_file = os.path.join(self.network_path,
"network-config-test-6.json")
sp = SnuddaPrune(network_path=self.network_path,
config_file=testing_config_file,
verbose=True,
keep_files=True) # Use default config file
sp.prune()
# Load the pruned data and check it
sl = SnuddaLoad(pruned_output)
# Setting f1=0.7 in config should remove 30% of gap junctions, but does so randomly
self.assertTrue(
64 * 0.7 - 10 < sl.data["nGapJunctions"] < 64 * 0.7 + 10)
with self.subTest("gap-junction-softmax"):
# Test of softmax
testing_config_file = os.path.join(self.network_path,
"network-config-test-7.json")
sp = SnuddaPrune(network_path=self.network_path,
config_file=testing_config_file,
verbose=True,
keep_files=True) # Use default config file
sp.prune()
# Load the pruned data and check it
sl = SnuddaLoad(pruned_output)
# Softmax reduces number of synapses
self.assertTrue(sl.data["nGapJunctions"] < 16 * 2 + 10)
with self.subTest("gap-junction-mu2"):
# Test of mu2
testing_config_file = os.path.join(self.network_path,
"network-config-test-8.json")
sp = SnuddaPrune(network_path=self.network_path,
config_file=testing_config_file,
verbose=True,
keep_files=True) # Use default config file
sp.prune()
# Load the pruned data and check it
sl = SnuddaLoad(pruned_output)
# With mu2 having 4 synapses means 50% chance to keep them, having 1 will be likely to have it removed
self.assertTrue(
64 * 0.5 - 10 < sl.data["nGapJunctions"] < 64 * 0.5 + 10)
with self.subTest("gap-junction-a3"):
# Test of a3
testing_config_file = os.path.join(self.network_path,
"network-config-test-9.json")
sp = SnuddaPrune(network_path=self.network_path,
config_file=testing_config_file,
verbose=True,
keep_files=True) # Use default config file
sp.prune()
# Load the pruned data and check it
sl = SnuddaLoad(pruned_output, verbose=True)
# a3=0.7 means 30% chance to remove all synapses between a pair
self.assertTrue(
64 * 0.7 - 10 < sl.data["nGapJunctions"] < 64 * 0.7 + 10)
if False: # Distance dependent pruning currently not implemented for gap junctions
with self.subTest("gap-junction-distance-dependent-pruning"):
# Testing distance dependent pruning
testing_config_file = os.path.join(
self.network_path, "network-config-test-10.json")
sp = SnuddaPrune(network_path=self.network_path,
config_file=testing_config_file,
verbose=True,
keep_files=True) # Use default config file
sp.prune()
# Load the pruned data and check it
sl = SnuddaLoad(pruned_output, verbose=True)
# "1*(d <= 120e-6)" means we remove all synapses further away than 100 micrometers
self.assertEqual(sl.data["nGapJunctions"], 2 * 4 * 4)
self.assertTrue(
(sl.data["gapJunctions"][:, 8] <=
120).all()) # Column 8 -- distance to soma in micrometers
class PruningIllustration(object):
def __init__(self, verbose=False, n_repeats=1000):
self.n_repeats = n_repeats
if os.path.dirname(__file__):
os.chdir(os.path.dirname(__file__))
self.network_path = "pruning_illustration_network"
# self.config_file = os.path.join(self.network_path, "network-config.json")
# self.save_file = os.path.join(self.network_path, "voxels", "network-synapses.hdf5")
create_cube_mesh(file_name=os.path.join(self.network_path, "mesh", "simple_mesh.obj"),
centre_point=(0, 0, 0),
side_len=500e-6)
# Default uses network_config.json
sp = SnuddaPlace(network_path=self.network_path, d_view=None, verbose=verbose)
print("Calling read_config")
sp.parse_config()
print("Read done")
position_file = os.path.join(self.network_path, "network-neuron-positions.hdf5")
sp.write_data(position_file)
# We want to load in the ball and stick neuron that has 20 micrometer soma diameter, and axon (along y-axis),
# and dendrite along (x-axis) out to 200 micrometer distance from centre of soma.
self.sd = SnuddaDetect(network_path=self.network_path, rc=None, hyper_voxel_size=150, verbose=verbose)
# Reposition the neurons so we know how many synapses and where they will be located before pruning
neuron_positions = np.array([[0, 59, 0], # Postsynaptiska
[0, 89, 0],
[0, 119, 0],
[0, 149, 0],
[0, 179, 0],
[0, 209, 0],
[0, 239, 0],
[0, 269, 0],
[0, 299, 0],
[0, 329, 0],
[59, 0, 0], # Presynaptiska
[89, 0, 0],
[119, 0, 0],
[149, 0, 0],
[179, 0, 0],
[209, 0, 0],
[239, 0, 0],
[269, 0, 0],
[299, 0, 0],
[329, 0, 0],
]) * 1e-6
# TODO: Add potential for gap junctions also by having 5 + 5 neurons in other grid
for idx, pos in enumerate(neuron_positions):
self.sd.neurons[idx]["position"] = pos
ang = -np.pi / 2
R_x = np.array([[1, 0, 0],
[0, np.cos(ang), -np.sin(ang)],
[0, np.sin(ang), np.cos(ang)]])
ang = np.pi / 2
R_y = np.array([[np.cos(ang), 0, np.sin(ang)],
[0, 1, 0],
[-np.sin(ang), 0, np.cos(ang)]])
for idx in range(0, 10): # Post synaptic neurons
self.sd.neurons[idx]["rotation"] = R_x
for idx in range(10, 20): # Presynaptic neurons
self.sd.neurons[idx]["rotation"] = R_y
self.sd.detect(restart_detection_flag=True)
# Also update so that the new positions are saved in the place file
rn = RepositionNeurons(position_file)
for neuron_info in self.sd.neurons:
rn.place(neuron_info["neuronID"], position=neuron_info["position"], rotation=neuron_info["rotation"],
verbose=False)
rn.close()
if False:
self.sd.process_hyper_voxel(1)
plt, ax = self.sd.plot_hyper_voxel(plot_neurons=True, elev_azim=(90, 0),
draw_axon_voxels=False, draw_dendrite_voxels=False,
draw_axons=True, draw_dendrites=True,
show_axis=False, title="No pruning",
fig_file_name="Pruning-fig-1-no-pruning")
import pdb
pdb.set_trace()
def prune_network(self, pruning_config=None, fig_name=None, title=None, verbose=False, plot_network=True,
random_seed=None, n_repeats=None):
if n_repeats is None:
n_repeats = self.n_repeats
work_log = os.path.join(self.network_path, "log", "network-detect-worklog.hdf5")
pruned_output = os.path.join(self.network_path, "network-synapses.hdf5")
if pruning_config is not None and not os.path.exists(pruning_config):
pruning_config = os.path.join(self.network_path, pruning_config)
# We keep temp files
sp = SnuddaPrune(network_path=self.network_path, config_file=pruning_config,
verbose=verbose, keep_files=True, random_seed=random_seed) # Use default config file
sp.prune()
n_synapses = sp.out_file["network/synapses"].shape[0]
n_gap_junctions = sp.out_file["network/gapJunctions"].shape[0]
sp = []
plot_axon = True
plot_dendrite = True
#plot_axon = np.ones((20,), dtype=bool)
#plot_dendrite = np.ones((20,), dtype=bool)
#plot_axon[:10] = False
#plot_dendrite[10:] = False
if plot_network:
pn = PlotNetwork(pruned_output)
plt, ax = pn.plot(fig_name=fig_name, show_axis=False,
plot_axon=plot_axon, plot_dendrite=plot_dendrite,
title=title, title_pad=-14,
elev_azim=(90, 0))
if n_repeats > 1:
n_syn_mean, n_syn_std, _, _ = self.gather_pruning_statistics(pruning_config=pruning_config, n_repeats=n_repeats)
plt.figtext(0.5, 0.15, f"(${n_syn_mean:.1f} \pm {n_syn_std:.1f}$)", ha="center", fontsize=16)
plt.savefig(fig_name, dpi=300, bbox_inches='tight')
# Load the pruned data and check it
# sl = SnuddaLoad(pruned_output)
return n_synapses, n_gap_junctions
def gather_pruning_statistics(self, pruning_config, n_repeats):
n_synapses = np.zeros((n_repeats,))
n_gap_junctions = np.zeros((n_repeats,))
ss = np.random.SeedSequence()
random_seeds = ss.generate_state(n_repeats)
for i, rand_seed in enumerate(random_seeds):
n_synapses[i], n_gap_junctions[i] = self.prune_network(pruning_config=pruning_config,
plot_network=False,
random_seed=rand_seed)
mean_syn, std_syn = np.mean(n_synapses), np.std(n_synapses)
mean_gj, std_gj = np.mean(n_gap_junctions), np.std(n_gap_junctions)
print(f"{pruning_config}\nsynapses : {mean_syn:.1f} +/- {std_syn:.1f}\ngap junctions: {mean_gj:.1f} +/- {std_gj:.1f}")
return mean_syn, std_syn, mean_gj, std_gj