#!/usr/bin/env python
# -*- coding: utf-8 -*-
import os
import sys
import math
import datetime
import pickle
## from node000:
## mpiexec -machinefile ~/hostfile -n <numtrains*numtrials+1> ~/Python-2.6.4/bin/python2.6 odor_whitenoise.py <simtype>
## nohup mpiexec -machinefile ~/hostfile -n 251 ~/Python-2.6.4/bin/python2.6 odor_whitenoise.py -1 < /dev/null &
## typical value for numtrains = 250
## typical value for num of trials = 1
## (depends on number of available processing nodes and number of odorfiles generated)
## Need to set the type of simulation i.e. whether for mainmitral kernel (simtype = -1)
## or simtype = 0..6 index of the varied_mainrate, etc. variable in stimuliConstants.py
## See varied_mainrate, varied_distance, varied_gran_baseline in stimuliConstants.py after
## setting the option of 'varied' to one of ('mainrate', 'distance', 'gran_baseline') )
## Set various option like NO_PGs or ONLY_TWO_MITS in simset_odor.py
## single simulation USAGE:
## python2.6 odor_whitenoise.py -1
sys.path.extend(["..","../networks","../generators","../simulations"])
from moose_utils import * # imports moose
from data_utils import * # has mpi import and variables also
from OBNetwork import *
from sim_utils import *
from stimuliConstants import * # has SETTLETIME, varied...
from simset_odor import *
from pylab import * # part of matplotlib that depends on numpy but not scipy
#-----------------------------------------------------------
class odorResponse:
def __init__(self, simtype):
self.mpirank = mpirank
self.context = moose.PyMooseBase.getContext()
self.simtype = int(simtype)
def setupStim(self,network,trainnum,trialnum):
self.setupOdor(network, trainnum, trialnum)
print "Setup trainnum =",trainnum,"at",self.mpirank
def setupOdor(self, network, trainnum, trailnum):
## first figure out which PG belongs to which glom
## PG_glom_map[pgname] returns the glom num of the PG:
## needed for ORN to PG connections.
PG_glom_map = {}
for projname in network.projectionDict.keys():
if 'PG_mitral' in projname:
for i,proj in enumerate(network.projectionDict[projname][2]):
## get the glomnum from the post path proj[2]
## name of the mitral cell from '/mitrals_2/...'
mitname = string.split(proj[2],'/')[1]
## glomerulus number from 'mitrals_2' by integer division i.e. 2/2 = 1
glomnum = int(string.split(mitname,'_')[1]) / 2
## name of the PG cell from '/PGs_2/...'
pgname = string.split(proj[1],'/')[1]
PG_glom_map[pgname] = glomnum
## Now connect the ORNs
for projname in network.projectionDict.keys():
#### Calling attach_spikes() for each projection,
#### would reconnect files to the same segment multiple times.
#### But attach_files_uniquely() checks whether timetable.tableSize is zero or not
#### i.e. files already attached or not.
## connect ORNs to mitrals
if 'ORN_mitral' in projname:
print "connecting ORN files to mitrals"
for i,proj in enumerate(network.projectionDict[projname][2]):
## get the glomnum from the post path proj[2]
## name of the mitral cell from '/mitrals_2/...'
mitname = string.split(proj[2],'/')[1]
## glomerulus number from 'mitrals_2' by integer division i.e. 2/2 = 1
glomnum = int(string.split(mitname,'_')[1]) / 2
self.attach_appropriate_files_to_glom(proj[0],proj[2],trainnum,trialnum,glomnum)
## connect ORNs to PG
if 'ORN_PG' in projname:
print "connecting ORN files to PGs"
for i,proj in enumerate(network.projectionDict[projname][2]):
pgname = string.split(proj[2],'/')[1] # name of the PG cell from '/PGs_2/...'
glomnum = PG_glom_map[pgname]
self.attach_appropriate_files_to_glom(proj[0],proj[2],trainnum,trialnum,glomnum)
def attach_appropriate_files_to_glom(self,proj1,proj2,trainnum,trialnum,glomnum):
## for the excitatory kernel
if self.simtype == -1:
## only connect glom0 to noise train to get exc kernel
if glomnum == central_glom:
filebase = ORNpathINHstr+'firetimes_whitenoise_glom'+str(glomnum)
self.attach_files_uniquely(filebase,proj1,proj2,trainnum,trialnum)
## for inhibitory kernel (simtype != -1)
else:
## connect glom0 to a constant firing rate
if glomnum == central_glom:
if varied == 'mainrate':
firingrate = varied_mainrate[self.simtype]
filebase = ORNpathINHstr+'firetimes_constrate'\
+str(firingrate)+'_avg'+str(trainnum)
self.attach_files_uniquely(filebase,proj1,proj2)
## connect other glom to noise train to get inh kernel
else:
filebase = ORNpathINHstr+'firetimes_whitenoise_glom'+str(glomnum)
self.attach_files_uniquely(filebase,proj1,proj2,trainnum,trialnum)
def attach_files_uniquely(self,filebase,synname,postsegpath,trainnum=None,trialnum=None):
ttpath = postsegpath+'/'+synname+'_tt'
if self.context.exists(ttpath):
# timetable already created by networkML reader - just wrap it below.
tt = moose.TimeTable(ttpath) # post_segment_path+'/'+syn_name+'_tt'
else:
## if timetable was not already created by networkML reader,
## it means that the synaptic weights must be zero!
## (no extra inhibition - only main inhibition)
## hence do not attach spikefiles
return
if tt.tableSize != 0: return # if files are already attached, do nothing!
if trainnum is not None: filebase += '_train'+str(trainnum)
if trialnum is not None: filebase += '_trial'+str(trialnum)
## attach_spikes() accesses filenumbers to this segment
## from 'fileNumbers' field (of the timetable object in MOOSE)
## which is created while reading in networkML.
attach_spikes(filebase, tt, self.mpirank)
def run(self,network, binned):
print "Resetting MOOSE."
# from moose_utils.py sets clocks and resets
resetSim(network.context, SIMDT, PLOTDT)
print "Running at",self.mpirank
network.context.step(PULSE_RUNTIME)
mitral_responses = []
if ONLY_TWO_MITS: mits = [mitralidx, mitralsidekickidx]
else: mits = range(NUM_GLOMS*MIT_SISTERS)
## network.mitralTable is a dictionary.
for mitnum in mits:
mitral = network.mitralTable[mitnum]
## need to convert to numpy's array(),
## else MOOSE table cannot be pickled for mpi4py send()
mitral_responses.append(array(mitral._vmTableSoma))
return mitral_responses
#----------------------------------------------------------------
if __name__ == "__main__":
## simtype == '-1': main mitral kernel
## other simtype gives the index of varied... variable
## to set the conditions for inh kernel due to nearby glomerulus.
simtype = sys.argv[1]
#### if only one process is called, plot one sim directly
if mpisize == 1:
#### run the slave processes
sim = odorResponse(simtype)
## includeProjections gets used only if ONLY_TWO_MITS is True:
## Keep below projections to 'second order cells'
## i.e. to cells (granules) connected to mits0&1.
## The connections between second order cell
## and mits0&1 are automatically retained of course.
# 'PG' includes 'ORN_PG', 'PG_mitral', 'mitral_PG' and 'SA_PG'
includeProjections = ['PG','granule_baseline']
tweaks = build_tweaks(CLUB_MITRALS, NO_SPINE_INH, NO_SINGLES, NO_JOINTS,\
NO_MULTIS, NO_PGS, ONLY_TWO_MITS, includeProjections, mitralsidekickidx)
network = OBNetwork(OBNet_file, synchan_activation_correction, tweaks,\
mpirank, granfilebase+'_noresp', spiketable=False)
#printNetTree() # from moose_utils.py
trialnum = 0
trainnum = 0
sim.setupStim(network, trainnum, trialnum)
mitral_responses = sim.run(network, binned=False)
timevec = arange(SETTLETIME,SETTLETIME+PULSE_RUNTIME+3*PLOTDT/2.0,PLOTDT)
plot(timevec, mitral_responses[0])
show()
#### multiple processes
else:
if mpirank == boss:
#### collate at boss process
## mitral_responses_list[avgnum][trainnum][mitnum][spikenum]
mitral_responses_list = []
numavgs = (mpisize-1)/NUMWHITETRAINS
for avgnum in range(numavgs):
response_set = []
for trainnum in range(NUMWHITETRAINS):
procnum = avgnum*NUMWHITETRAINS + trainnum + 1
print 'waiting for process '+str(procnum)+'.'
## below: you get a numpy array of
## rows=NUM_GLOMS*MIT_SISTERS and cols=spike times
## mitral responses has spike times
## we calculate STA to get kernel from spike times.
mitral_responses = mpicomm.recv(source=procnum, tag=0)
response_set.append( mitral_responses )
mitral_responses_list.append(response_set)
# write results to a file
today = datetime.date.today()
if NO_SINGLES: singles_str = '_NOSINGLES'
else: singles_str = '_SINGLES'
if NO_JOINTS: joints_str = '_NOJOINTS'
else: joints_str = '_JOINTS'
if NO_PGS: pgs_str = '_NOPGS'
else: pgs_str = '_PGS'
now = datetime.datetime.now().strftime("%Y_%m_%d_%H_%M")
outfilename = '../results/odor_whitenoise/'+now+'_whitenoise'+singles_str+\
joints_str+pgs_str+'_numgloms'+str(NUM_GLOMS)+'_simtype'+simtype+'.pickle'
f = open(outfilename,'w')
pickle.dump(mitral_responses_list, f)
f.close()
print "Wrote", outfilename
else:
#### run the slave processes
sim = odorResponse(simtype)
## includeProjections gets used only if ONLY_TWO_MITS is True:
## Keep below projections to 'second order cells'
## i.e. to cells (granules) connected to mits0&1.
## The connections between second order cell
## and mits0&1 are automatically retained of course.
# 'PG' includes 'ORN_PG', 'PG_mitral', 'mitral_PG' and 'SA_PG'
includeProjections = ['PG','granule_baseline']
tweaks = build_tweaks(CLUB_MITRALS, NO_SPINE_INH, NO_SINGLES, NO_JOINTS,\
NO_MULTIS, NO_PGS, ONLY_TWO_MITS, includeProjections, mitralsidekickidx)
network = OBNetwork(OBNet_file, synchan_activation_correction, tweaks,\
mpirank, granfilebase+'_noresp', spiketable=True)
#printNetTree() # from moose_utils.py
trialnum = (mpirank-1)/NUMWHITETRAINS
trainnum = (mpirank-1)%NUMWHITETRAINS
sim.setupStim(network, trainnum, trialnum)
mitral_responses = sim.run(network, binned=True)
mpicomm.send( mitral_responses, dest=boss, tag=0 )
print 'sent from process',mpirank