#!/usr/bin/env python3
"""
Simulate upstate conditions for Patch Samples 4-5 and Matrix Samples 2-3 models.
Modify local channel conductances at site of clustered input for each neuron
to achieve upstate duration and amplitude consistent with experimental averages.
Do current injection with modified conductances to confirm modifying them does
not greatly alter the fit to current injection data.
Simulate without blocking sodium channels.
Simulate with additional dispersed inputs.
Simulation steps:
For each neuron:
- Randomly select parameters from within a range to vary
- parameters to vary:
- Random seed necessary for selecting parameters?
For each set of parameters:
- Use same random seeds to control synapse selection
- [done] simulate upstate only: Need an upstate seed (same every sim/param set)
- [done] Simulate dispersed only: Need a Dispersed seed (same every sim/param set for now)
- [done] Simulate upstate and dispersed together: Use same upstate seed and same dispersed seed
- [ ] range over dispersion frequency params
- [later] Should we simulate "spatially disperse" the clustered inputs but at the same time as a control? Not for now
- [done] Simulate single EPSP (EPSP seed?-same every sim)
- [done] Simulate IV traces to compare to original model IV to see how much the optimization fit is messed up
- [ ] simulate upstate plus current injection at increasing steps...
TO DO:
Save voltage at soma from each simulation
File name scheme:
param_set_X_sim_name_sim_variable_name_value_neuron-name_vm.txt
e.g.
param_set_1__upstate_plus_dispersed__dispersed_freq_375__D1PatchSample5_vm.txt
Make plotting optional argument
Save parameter variation values (and any necessary random seeds?) for each simulation
param_set_list.csv
set_ID (corresponds to param_set_X in filenames), var1name, var2name...
e.g.
0, 2, 3,...
"""
global_test = False
import importlib
import numpy as np
def moose_main(corticalinput,LTP_amp_thresh_mod=1.158, LTD_amp_thresh_mod=1.656, LTP_dur_thresh_mod=1.653, LTD_dur_thresh_mod=0.867,LTP_gain_mod=0.704, LTD_gain_mod=1.671,nmda_mod=1,seed=42):
import logging
import numpy as np
np.random.seed(seed)
import matplotlib.pyplot as plt
from pprint import pprint
import moose
from moose_nerp.prototypes import (
create_model_sim,
cell_proto,
clocks,
inject_func,
create_network,
tables,
net_output,
logutil,
util,
standard_options,
ttables,
)
#from moose_nerp import d1opt as model
#from moose_nerp import D1MatrixSample2 as model
from moose_nerp import D1PatchSample5 as model
from moose_nerp import str_net as net
from moose_nerp.graph import net_graph, neuron_graph, spine_graph
# additional, optional parameter overrides specified from with python terminal
# model.Condset.D1.NaF[model.param_cond.prox] /= 3
# model.Condset.D1.KaS[model.param_cond.prox] *= 3
model.param_syn._SynAMPA.Gbar = .25e-9
model.param_syn._SynNMDA.Gbar = .25e-9
model.param_syn._SynGaba.Gbar = 1e-9
net.connect_dict["D1"]["ampa"]["extern1"].dend_loc.postsyn_fraction = 0.8
net.param_net.tt_Ctx_SPN.filename = corticalinput
print(
"cortical_fraction = {}".format(
net.connect_dict["D1"]["ampa"]["extern1"].dend_loc.postsyn_fraction
)
)
model.synYN = True
model.plasYN = True
model.calYN = True
model.spineYN = True
# vary plasticity params
model.CaPlasticityParams.Plas2_syn.LTP_amp_thresh*=LTP_amp_thresh_mod
model.CaPlasticityParams.Plas2_syn.LTD_amp_thresh*=LTD_amp_thresh_mod
model.CaPlasticityParams.Plas2_syn.LTP_dur_thresh*=LTP_dur_thresh_mod
model.CaPlasticityParams.Plas2_syn.LTD_dur_thresh*=LTD_dur_thresh_mod
model.CaPlasticityParams.Plas2_syn.LTP_gain*=LTP_gain_mod
model.CaPlasticityParams.Plas2_syn.LTD_gain*=LTD_gain_mod
# new changes: Universal function to modify variables. Needs to extract any attribute based (e.g. model.CaParams...) or dict/list based (e.g. model.dict['key']) parameter name as a string, parse the variable, and apply the modification. Additionally, need to specify a multiplicative change, additive change, or direct replacement.
#model.CaPlasticityParams.BufferCapacityDensity*=buffer_capacity_density_gain_mod
#
net.single = True
model.ConcOut = model.param_cond.ConcOut = 1.2
create_model_sim.setupOptions(model)
param_sim = model.param_sim
param_sim.useStreamer = True
param_sim.plotdt = 0.1e-3
param_sim.stim_loc = model.NAME_SOMA
param_sim.stim_paradigm = "inject"
param_sim.injection_current = [0] # [-0.2e-9, 0.26e-9]
param_sim.injection_delay = 0.2
param_sim.injection_width = 0.4
param_sim.simtime = 2 if not global_test else .01#21.5#3.5 # 21
net.num_inject = 0
model.name = model.__name__.split('.')[-1]
net.confile = "str_connect_plas_sim{}_{}_corticalfraction_{}".format(
model.name, net.param_net.tt_Ctx_SPN.filename, 0.8
)
if net.num_inject == 0:
param_sim.injection_current = [0]
#################################-----------create the model: neurons, and synaptic inputs
model = create_model_sim.setupNeurons(model, network=not net.single)
all_neur_types = model.neurons
# FSIsyn,neuron = cell_proto.neuronclasses(FSI)
# all_neur_types.update(neuron)
## Limit input trains to 2 seconds and shuffle ISIs:
ttables.TableSet.create_all()
randomize_input_trains(net.param_net.tt_Ctx_SPN)
population, connections, plas = create_network.create_network(
model, net, all_neur_types,create_all=False
)
###### Set up stimulation - could be current injection or plasticity protocol
# set num_inject=0 to avoid current injection
if net.num_inject < np.inf:
inject_pop = inject_func.inject_pop(population["pop"], net.num_inject)
else:
inject_pop = population["pop"]
# Does setupStim work for network?
# create_model_sim.setupStim(model)
pg = inject_func.setupinj(
model, param_sim.injection_delay, param_sim.injection_width, inject_pop
)
moose.showmsg(pg)
##############--------------output elements
if net.single:
# fname=model.param_stim.Stimulation.Paradigm.name+'_'+model.param_stim.location.stim_dendrites[0]+'.npz'
# simpath used to set-up simulation dt and hsolver
simpath = ["/" + neurotype for neurotype in all_neur_types]
create_model_sim.setupOutput(model)
else: # population of neurons
spiketab, vmtab, plastab, catab = net_output.SpikeTables(
model, population["pop"], net.plot_netvm, plas, net.plots_per_neur
)
# simpath used to set-up simulation dt and hsolver
simpath = [net.netname]
clocks.assign_clocks(
simpath,
param_sim.simdt,
param_sim.plotdt,
param_sim.hsolve,
model.param_cond.NAME_SOMA,
)
if model.synYN and (param_sim.plot_synapse or net.single):
# overwrite plastab above, since it is empty
syntab, plastab, stp_tab = tables.syn_plastabs(connections, model)
nonstim_plastab = tables.nonstimplastabs(plas)
# Streamer to prevent Tables filling up memory on disk
# This is a hack, should be better implemented
if param_sim.useStreamer == True:
allTables = moose.wildcardFind("/##[ISA=Table]")
streamer = moose.Streamer("/streamer")
streamer.outfile = "testdata/plas_sim{}_{}_seed_{}.npy".format(model.name,
net.param_net.tt_Ctx_SPN.filename, seed)
moose.setClock(streamer.tick, 0.1)
for t in allTables:
if any(s in t.path for s in ["plas", "VmD1_0", "extern", "Shell_0"]):
streamer.addTable(t)
else:
t.tick = -2
################### Actually run the simulation
def run_simulation(injection_current, simtime):
print(u"◢◤◢◤◢◤◢◤ injection_current = {} ◢◤◢◤◢◤◢◤".format(injection_current))
pg.firstLevel = injection_current
moose.reinit()
moose.start(simtime, True)
import os
print('does outfile {} exist before sim: {}'.format(streamer.outfile,os.path.exists(streamer.outfile)))
traces, names = [], []
for inj in param_sim.injection_current:
run_simulation(injection_current=inj, simtime=param_sim.simtime)
print('does outfile {} exist AFTER sim: {}'.format(streamer.outfile,os.path.exists(streamer.outfile)))
#weights = [w.value for w in moose.wildcardFind("/##/plas##[TYPE=Function]")]
#plt.figure()
#plt.hist(weights, bins=100)
#plt.title("plas_sim_{}".format(net.param_net.tt_Ctx_SPN.filename))
#plt.savefig("plas_simd1opt_{}.png".format(net.param_net.tt_Ctx_SPN.filename))
if param_sim.useStreamer == True:
import atexit
atexit.register(moose.quit)
#return weights
def randomize_input_trains(timetable):
for tt in timetable.stimtab:
v = tt[0].vector # copies time table vector to v (doesn't reference it)
v = v[v<=2] # Use only first trial, i.e. spike times less than 2
diff = np.diff(v,prepend=0) # compute ISIs using diff
np.random.shuffle(diff) # Shuffle ISIs to randomize timing pattern
v = np.cumsum(diff) # create new spike train from shuffle differences
tt[0].vector = v # Replace timetable vector with new values
##import pdb;pdb.set_trace()
return timetable
def subprocess_main(function, corticalinput,kwds,time_limit):
print('enter subprocess_main')
from multiprocessing import Process, Queue
import time
# q = Queue()
p = Process(target=function, args=(corticalinput,), kwargs=kwds)
p.start()
# result = q.get()
# print(result)
remaining = time_limit - time.time()
if remaining <=0:
p.terminate()
return
p.join(timeout=remaining-10)
p.terminate()
# return result
def make_rand_mod_dict(n=100):
#plas_mod_values =[.5,2]
#plas_mod_keys = ['LTP_amp_thresh_mod', 'LTD_amp_thresh_mod','LTP_dur_thresh_mod', #'LTD_dur_thresh_mod','LTP_gain_mod', 'LTD_gain_mod']
#rand_mod_list = np.random.uniform(*plas_mod_values,size=(n,6))
#rand_mod_dicts = [{k:v for k,v in zip(plas_mod_keys,r)} for r in rand_mod_list]
np.random.seed(42)
rand_mod_dicts = [{'seed':v} for v in np.random.randint(100000,size=n)]
return rand_mod_dicts
def mpi_main():
if __name__ == "__main__":
from mpi4py import MPI
from mpi4py.futures import MPICommExecutor
import time
import pickle
with MPICommExecutor(MPI.COMM_WORLD, root=0) as executor:
time_limit = time.time() + 60 * 60 * 8 if not global_test else time.time() + 60*60*.1#3.75 # 3.75 hours
if executor is not None:
results = []
make_new_params = True
if make_new_params:
n = 1000 if not global_test else 52
param_set_list = make_rand_mod_dict(n=n)
with open("testparams.pickle", "wb") as f:
pickle.dump(param_set_list, f)
else:
with open("testparams.pickle",'rb') as f:
param_set_list = pickle.load(f)
# print(param_set_list)
for i, param_set in enumerate(param_set_list):#1020
print(i, param_set)
r = executor.submit(
subprocess_main, *(moose_main,"FullTrialLowVariabilitySimilarTrialsTruncatedNormal", param_set, time_limit)
)
results.append(r)
while True:
if all([res.done() for res in results]):
print('all results returned done; breaking')
#import pdb;pdb.set_trace()
break
if time.time() >= time_limit:
print("****************** TIME LIMIT EXCEEDED***********")
for res in results:
res.cancel()
#print('canceling', res)
#executor.shutdown(wait=False)
print('shutting down')
MPI.COMM_WORLD.Abort()
print('aborting')
break
print('done')
return
#while True:
# if time.time() >= time_limit:
# break
#MPI.COMM_WORLD.Abort()
### Notes
### What to randomize on each simulation:
### 1. Spatiotemporal input pattern. Select X+/-i in vivo spike trains, jitter spike times by Y, move spikes to different trains with prob Z, shuffle ISIs?
### 2. Neuron type: patch or matrix?
### 3. Soma NaF: blocked or not?
### 4. Other model parameters: varied +/- 10%? Which parameters?
if __name__ == "__main__":
import sys
args = sys.argv
# args.append('--single')
if "--test" in args:
global_test = True
print('global test = {}'.format(global_test))
if len(args) > 1 and args[1] == "--single":
# upstate_main(list(mod_dict.keys())[0],mod_dict)
moose_main("FullTrialLowVariabilitySimilarTrialsTruncatedNormal", seed=42)
elif len(args) > 1 and args[1] == "--iv":
# upstate_main(list(mod_dict.keys())[0],mod_dict)
iv_main("D1PatchSample5", mod_dict, filename="test")
elif len(args) > 1 and args[1] == "--mp":
results = []
from multiprocessing import Pool
with Pool(16, maxtasksperchild=1) as p:
param_set_list = [rand_mod_dict() for i in range(10000)]
import pickle
with open("params.pickle", "wb") as f:
pickle.dump(param_set_list, f)
#print(param_set_list)
for i, param_set in enumerate(param_set_list):
for key in mod_dict:
for sim in sims:
# param_set_1__upstate_plus_dispersed__dispersed_freq_375__D1PatchSample5_vm.txt
filename = (
"param_set_"
+ str(i)
+ "__"
+ sim["name"]
+ "__dispersed_freq_"
+ str(sim["kwds"].get("freq_dispersed"))
+ "__"
+ key
)
kwds = {k: v for k, v in sim["kwds"].items()}
kwds["filename"] = filename
# r = p.apply_async(upstate_main, args=(key, mod_dict),kwds={'num_dispersed':0})
r = p.apply_async(sim["f"], args=(key, param_set), kwds=kwds)
results.append(r)
for res in results:
res.wait()
else:
mpi_main()
print('done?')