# -*- coding: utf-8 -*-
"""
Created on Thu Jul 2 10:32:29 2015
@author: rennocosta
This script is part of the publication Renno-Costa & Tort, 2017, JNeurosci
This script relates to the data presented in the Figure 9a
Will run a single experiment with specific parameters determined below
Output data will be saved in file direction defined at support_filename.py script
"""
import sys, argparse
import numpy as np
from numpy import *
import gzip
import pickle
import support_filename as rfn
import copy
def normalize_weight(www,www_mean):
www /= np.tile(np.mean(www,axis=0),(www.shape[0],1))
return www
def learn_weight(www,activity_pre,activity_pos,lrate):
#www += lrate*(np.tile(activity_pre,(activity_pos.shape[0],1)).transpose()-www) * (np.tile(activity_pos,(activity_pre.shape[0],1)))
www += lrate*(np.tile(activity_pre,(activity_pos.shape[0],1)).transpose()) * (np.tile(activity_pos,(activity_pre.shape[0],1)))
return www
def lec_whichone(lectype,change,ccc,sss):
saida = np.zeros(lectype.shape)
saida[np.logical_and(lectype==1,change<ccc)] = 2
saida[np.logical_and(lectype==1,change>=ccc)] = 1
saida[np.logical_and(lectype==0,change<sss)] = 2
saida[np.logical_and(lectype==0,change>=sss)] = 1
saida[lectype==2] = 1
return saida
def main(argv):
parser = argparse.ArgumentParser(description='Will run a simulation instance.')
# arguments with seeds for the random number generator
parser.add_argument('seed_input', metavar='seed_input', type=int, nargs=1,
help='seed_input number')
parser.add_argument('seed_www', metavar='seed_www', type=int, nargs=1,
help='seed_www')
parser.add_argument('seed_path', metavar='seed_path', type=int, nargs=1,
help='seed_path')
# number of theta cycles, number of full runs for each session and the number of runs before expeirment
parser.add_argument('theta_cycles', metavar='theta_cycles', type=int, nargs=1,
help='theta_cycles')
parser.add_argument('arena_runs', metavar='arena_runs', type=int, nargs=1,
help='arena_runs')
parser.add_argument('pre_runs', metavar='pre_runs', type=int, nargs=1,
help='pre_runs')
# relative number of grid cells vs lec cells... relative number of place cells vs recurrent grid cells ...
# sensibility of pattern completion algorithm
parser.add_argument('lrate_hpc_mec', metavar='lrate_hpc_mec', type=int, nargs=1,
help='lrate_hpc_mec')
parser.add_argument('lrate_mec_hpc', metavar='lrate_mec_hpc', type=int, nargs=1,
help='lrate_mec_hpc')
parser.add_argument('lrate_lec_hpc', metavar='lrate_lec_hpc', type=int, nargs=1,
help='lrate_lec_hpc')
# percentage of correlation after morphing (from 0 to 100)
parser.add_argument('mec_ratio', metavar='mec_ratio', type=int, nargs=1,
help='MEC ratio (x100)')
parser.add_argument('hpc_ratio', metavar='hpc_ratio', type=int, nargs=1,
help='HPC ratio (x100)')
parser.add_argument('hpc_pcompl_th', metavar='hpc_pcompl_th', type=int, nargs=1,
help='HPC pattern completion th (x100)')
# percentage of correlation after morphing (from 0 to 100)
parser.add_argument('morph_per', metavar='morph_per', type=int, nargs=1,
help='morph_per')
# parser.add_argument('input_noise', metavar='input_noise', type=int, nargs=1,
# help='input_noise')
# parser.add_argument('-w', '--windows',dest='envir',action='store_const',default="default",const="windows")
# parser.add_argument('-u', '--ufrgs',dest='envir',action='store_const',default="default",const="UFRGS")
parser.add_argument('-s', '--cluster',dest='envir',action='store_const',default="default",const="cluster")
parser.add_argument('-a', '--activity',dest='actsave',action='store_const',default="no",const="yes")
parser.add_argument('-k', '--KILL',dest='tokill',action='store_const',default="no",const="yes")
args = parser.parse_args()
envir = args.envir
# conntype = args.conntype
actsave = args.actsave
tokill = args.tokill;
ct = 0
conna = False
# if(conntype=="yes"):
# conna = True
# ct = 1
# else:
# ct = 0
# conna = False
if(actsave=="yes"):
acts = True
else:
acts = False
seed_input = args.seed_input[0]
seed_www = args.seed_www[0]
seed_path = args.seed_path[0]
mec_ratio = float(args.mec_ratio[0])/100
hpc_ratio = float(args.hpc_ratio[0])/100
hpc_pcompl_th = float(args.hpc_pcompl_th[0])/100
morphing_per = float(args.morph_per[0])/100
#hpc_noise = float(args.hpc_noise[0])/100
#mec_noise = float(args.mec_noise[0])/100
#input_noise = float(args.input_noise[0])/100
lrate_hpc_mec = float(args.lrate_hpc_mec[0])/1000
lrate_mec_hpc = float(args.lrate_mec_hpc[0])/1000
lrate_lec_hpc = float(args.lrate_lec_hpc[0])/1000
theta_cycles = args.theta_cycles[0]
arena_runs = args.arena_runs[0]
pre_runs = args.pre_runs[0]
simulation_num = 66;
listofvalues = [ct,args.seed_input[0],args.seed_www[0],args.seed_path[0],args.theta_cycles[0],args.arena_runs[0],args.pre_runs[0],args.lrate_hpc_mec[0],args.lrate_mec_hpc[0],args.lrate_lec_hpc[0],args.mec_ratio[0],args.hpc_ratio[0],args.hpc_pcompl_th[0],args.morph_per[0]]
filenames = rfn.remappingFileNames(envir)
filenames.prepareSimulation(listofvalues,simulation_num)
if (tokill == "no"):
try:
tosee = 0;
with gzip.open(filenames.fileRunPickle(listofvalues,simulation_num,0)+'z', 'rb') as ff:
tosee = tosee + 1
with gzip.open(filenames.fileRunPickle(listofvalues,simulation_num,1)+'z', 'rb') as ff:
tosee = tosee + 1
with gzip.open(filenames.fileRunPickle(listofvalues,simulation_num,2)+'z', 'rb') as ff:
tosee = tosee + 1
with gzip.open(filenames.fileRunPickle(listofvalues,simulation_num,3)+'z', 'rb') as ff:
tosee = tosee + 1
with gzip.open(filenames.fileRunPickle(listofvalues,simulation_num,4)+'z', 'rb') as ff:
tosee = tosee + 1
with gzip.open(filenames.fileRunPickle(listofvalues,simulation_num,5)+'z', 'rb') as ff:
tosee = tosee + 1
print("File exist. Will exit!")
torun = 0;
except:
print("File does not existe. Will do!")
print("... %s" % (filenames.fileRunPickle(listofvalues,simulation_num,0)))
torun = 1;
else:
print("Will do anyway!")
torun = 0;
if(torun == 0):
sys.exit();
# %%
arena_binsize = [4,4]
context_per = 0
#morphing_per = 0
lec_numcells = 500
#mec_blocks = 50
hpc_numcells = 5000
np.random.seed(seed_input)
lec_numcells = 500
lec_activity = []
lec_activity.append(pow(np.random.uniform(0,1,(lec_numcells,arena_binsize[0],arena_binsize[1])),2))
lec_activity.append(pow(np.random.uniform(0,1,(lec_numcells,arena_binsize[0],arena_binsize[1])),2))
lec_type = np.random.uniform(0,1,(lec_numcells,arena_binsize[0],arena_binsize[1]))
lec_type[lec_type>(1-context_per)] = 1
lec_type[lec_type<morphing_per] = 0
lec_type[np.logical_and(lec_type<=(1-context_per),lec_type>=morphing_per)]=2
lec_change = np.random.uniform(0,1,(lec_numcells,arena_binsize[0],arena_binsize[1]))
hpc_memories = []
mec_blocksize = [2,4,6,8,10,12,14,16]
mec_blocks = len(mec_blocksize)
mec_numcells = np.sum(np.power(mec_blocksize,2))
mec_indexlist = []
init_val = 0
for ii in arange(mec_blocks):
mec_indexlist.append((init_val+arange(pow(mec_blocksize[ii],2))).reshape((mec_blocksize[ii],mec_blocksize[ii])))
init_val = np.max(mec_indexlist[ii])+1
del(init_val)
#lec_noise_activity = pow(np.random.uniform(0,1,(lec_numcells,arena_binsize[0],arena_binsize[1])),2)
#lec_noise_change = np.random.uniform(0,1,(lec_numcells,arena_binsize[0],arena_binsize[1]))
np.random.seed(seed_path)
xxx,yyy = np.meshgrid(arange(arena_binsize[0]),arange(arena_binsize[1]))
xxx = xxx.ravel()
popo = []
for ii in arange(100):
popo.append(np.random.permutation(len(xxx)))
np.random.seed(seed_www)
lec_hpc_weights_mean = 1
#lec_hpc_weights = np.random.uniform(0,1,(lec_numcells,hpc_numcells))
lec_hpc_weights = np.random.lognormal(1.0,1.0,(lec_numcells,hpc_numcells))
lec_hpc_weights[lec_hpc_weights<0] = 0
lec_hpc_weights = normalize_weight(lec_hpc_weights,lec_hpc_weights_mean)
mec_hpc_weights_mean = 1
#mec_hpc_weights = np.random.uniform(0,1,(mec_numcells,hpc_numcells))
mec_hpc_weights = np.random.lognormal(1.0,1.0,(mec_numcells,hpc_numcells))
mec_hpc_weights[mec_hpc_weights<0] = 0
mec_hpc_weights = normalize_weight(mec_hpc_weights,mec_hpc_weights_mean)
hpc_mec_weights_mean = 1
hpc_mec_weights = np.random.lognormal(1.0,1.0,(hpc_numcells,mec_numcells))
#hpc_mec_weights = np.random.uniform(0,1,(hpc_numcells,mec_numcells))
hpc_mec_weights[hpc_mec_weights<0] = 0
hpc_mec_weights = normalize_weight(hpc_mec_weights,hpc_mec_weights_mean)
current_emax = 0.90
current_emax_plast = 0
current_lrate_hpc_mec = lrate_hpc_mec
current_lrate_mec_hpc = lrate_mec_hpc
current_lrate_lec_hpc = lrate_lec_hpc
lec_hpc_weights_mean = 1
mec_hpc_weights_mean = 1
hpc_mec_weights_mean = 1
lec_noise = 0
mec_noise = 0
hpc_noise = 0
# %% PREPARE-PRE-LEARN
lllf = [1.0,1.0]
mooo = [mec_ratio,mec_ratio]
shape_vec = [0.0,1.0]
context_vec = [0.0,0.0]
pzzz = [0,0]
# %% PRE-LEARN
for sessions in arange(pre_runs):
print("session %d of %d" % (sessions,pre_runs))
hhhr = hpc_ratio * (sessions/(pre_runs-1))
#if(sessions==0):
# hhhr = 0.0
lllf[0] = 1.0
lllf[1] = 1.0
pzzz[0] = sessions + 16
pzzz[1] = sessions + 16 + pre_runs
lec_act_vect = []
mec_act_vect = []
hpc_act_vect = []
mec_inact_vect = np.zeros((len(shape_vec),mec_numcells,arena_binsize[0],arena_binsize[1]))
hpc_inact_vect = np.zeros((len(shape_vec),hpc_numcells,arena_binsize[0],arena_binsize[1]))
lec_inact_vect = np.zeros((len(shape_vec),lec_numcells,arena_binsize[0],arena_binsize[1]))
for ii in arange(len(shape_vec)):
print("shape %d of %d" % (ii,len(shape_vec)))
mec_ratio = mooo[ii]
lec_act = zeros((lec_numcells,arena_binsize[0],arena_binsize[1]))
hpc_act = zeros((hpc_numcells,arena_binsize[0],arena_binsize[1]))
mec_act = zeros((mec_numcells,arena_binsize[0],arena_binsize[1]))
xxx,yyy = np.int0(np.meshgrid(arange(arena_binsize[0]),arange(arena_binsize[0])))
xxx = xxx.ravel()
yyy = yyy.ravel()
#ppp = np.random.permutation(len(xxx))
ppp = popo[pzzz[ii]]
xxx = xxx[ppp]
yyy = yyy[ppp]
#
# xxx = xxx + 4
if (lllf[ii]>0):
xxxr = []
yyyr = []
for arena_runss in arange(arena_runs):
xxxr = concatenate([xxx,xxxr])
yyyr = concatenate([yyy,yyyr])
xxx = xxxr
yyy = yyyr
if((conna == True) and (ii > 0)):
current_pos = current_pos - array((4,0))
else:
current_pos = array((xxx[0],yyy[0]))
current_pos = np.int0(current_pos)
current_hpc_activity = np.zeros(hpc_numcells)
if ((ii<1) or (conna == False)):
current_mec_activity = np.zeros(mec_numcells)
current_context = context_vec[ii]
current_shape = shape_vec[ii]
current_vector = lec_whichone(lec_type,lec_change,current_context,current_shape)
base_lec = np.zeros(current_vector.shape)
base_lec = lec_activity[0].copy()
base_lec[current_vector==2] = lec_activity[1][current_vector==2]
for pp in arange(len(xxx)):
print("aaa %d of %d" % (pp,len(xxx)))
current_pos_old = current_pos
current_pos = np.int0(array((xxx[pp],yyy[pp])))
current_speed = current_pos - current_pos_old
current_lec_activity = base_lec[:,current_pos[0],current_pos[1]]
current_lec_noise = np.random.uniform(0.0,lec_noise,current_lec_activity.shape)
current_mec_noise = np.random.uniform(0.0,mec_noise,current_mec_activity.shape)
current_hpc_noise = np.random.uniform(0.0,hpc_noise,current_hpc_activity.shape)
lec_inact_vect[ii,:,np.int0(xxx[pp]),np.int0(yyy[pp])] = current_lec_activity
for kk in arange(theta_cycles):
if (kk>0): current_speed = array((0,0))
current_mec_input = (current_mec_activity+current_mec_noise)
if (mec_ratio>0.0):
for jj in arange(mec_blocks):
gxx,gyy = meshgrid(arange(mec_blocksize[jj])+(-1)*current_speed[0],arange(mec_blocksize[jj])+(-1)*current_speed[1])
gyy[mod(divide(gxx-mod(gxx,mec_blocksize[jj]),mec_blocksize[jj]),2)>0] = gyy[mod(divide(gxx-mod(gxx,mec_blocksize[jj]),mec_blocksize[jj]),2)>0] + floor(mec_blocksize[jj]/2)
gxx = int0(mod(gxx,mec_blocksize[jj]))
gyy = int0(mod(gyy,mec_blocksize[jj]))
current_mec_input[mec_indexlist[jj]] = current_mec_input[mec_indexlist[jj]][gyy,gxx]
#mec_input_vect[ii,kk,mec_indexlist[jj],xxx[pp],yyy[pp]] = current_mec_input[mec_indexlist[jj]]
h_h = np.dot(current_hpc_activity+current_hpc_noise,hpc_mec_weights)
if(np.max(h_h)>0.0):
h_h = h_h/np.max(h_h)
h_h[isnan(h_h)] = 0.0
current_mec_input = (1-mec_ratio)*h_h + mec_ratio*current_mec_input
current_lec_noise = np.random.uniform(0.0,lec_noise,current_lec_activity.shape)
current_mec_noise = np.random.uniform(0.0,mec_noise,current_mec_activity.shape)
current_hpc_noise = np.random.uniform(0.0,hpc_noise,current_hpc_activity.shape)
for jj in arange(mec_blocks):
current_mec_activity[mec_indexlist[jj]] = (current_mec_input[mec_indexlist[jj]] - current_emax*np.max(current_mec_input[mec_indexlist[jj]]))
current_mec_activity[current_mec_activity<0] = 0.0
current_mec_activity[mec_indexlist[jj]] /= np.max(current_mec_activity[mec_indexlist[jj]])
current_mec_activity[isnan(current_mec_activity)] = 0.0
mec_inact_vect[ii,mec_indexlist[jj],np.int0(xxx[pp]),np.int0(yyy[pp])] = current_mec_activity[mec_indexlist[jj]]
h_l = np.dot(current_lec_activity+current_lec_noise,lec_hpc_weights)
h_l = h_l/np.max(h_l)
h_l[isnan(h_l)] = 0.0
if(hhhr>0):
h_m = np.dot(current_mec_activity+current_mec_noise,mec_hpc_weights)
if(np.max(h_m)>0.0):
h_m = h_m/np.max(h_m)
h_m[isnan(h_m)] = 0.0
current_hpc_input = (1-hhhr)*h_l + hhhr*h_m
else:
current_hpc_input = h_l;
#hpc_input_vect[ii,kk,:,xxx[pp],yyy[pp]] = current_hpc_input
#hpc_input_vect[ii,kk,:,xxx[pp],yyy[pp]] = current_hpc_input
if (kk>0):
ddd = current_hpc_activity * 0
for mm in arange(len(hpc_memories)):
ccc = corrcoef(hpc_memories[mm],current_hpc_activity+current_hpc_noise)[0][1]
if ccc<hpc_pcompl_th:
ccc=0
else:
ddd += hpc_memories[mm] #ccc=1
# - current_hpc_activity
#ddd[ddd<0] = 0
#current_hpc_activity += ddd * ccc
if (np.max(ddd) > 0):
ddd = ddd/np.max(ddd)
ddd[isnan(ddd)] = 0.0
current_hpc_input = (1-mec_ratio)*current_hpc_input + mec_ratio*ddd
current_hpc_activity = (current_hpc_input - current_emax*np.max(current_hpc_input))
current_hpc_activity[current_hpc_activity<0] = 0.0
current_hpc_activity /= np.max(current_hpc_activity)
current_hpc_activity[current_hpc_activity<current_emax_plast] = 0
# for mm in arange(len(hpc_memories)):
# ccc = corrcoef(hpc_memories[mm],current_hpc_activity+current_hpc_noise)[0][1]
# if ccc<hpc_pcompl_th:
# ccc=0
# else:
# ccc=1
# ddd = hpc_memories[mm] - current_hpc_activity
# ddd[ddd<0] = 0
# current_hpc_activity += ddd * ccc
hpc_inact_vect[ii,:,np.int0(xxx[pp]),np.int0(yyy[pp])] = current_hpc_activity
#hpc_inact_vect[ii,kk,:,xxx[pp],yyy[pp]] = current_hpc_activity
if (lllf[ii]>0):
#hpc_mec_weights_t = copy.copy(hpc_mec_weights)
#mec_hpc_weights_t = copy.copy(mec_hpc_weights)
#lec_hpc_weights_t = copy.copy(lec_hpc_weights)
lec_hpc_weights = normalize_weight(learn_weight(lec_hpc_weights,current_lec_activity+current_lec_noise,current_hpc_activity+current_hpc_noise,current_lrate_lec_hpc),lec_hpc_weights_mean)
mec_hpc_weights = normalize_weight(learn_weight(mec_hpc_weights,current_mec_activity+current_mec_noise,current_hpc_activity+current_hpc_noise,current_lrate_mec_hpc),mec_hpc_weights_mean)
hpc_mec_weights = normalize_weight(learn_weight(hpc_mec_weights,current_hpc_activity+current_hpc_noise,current_mec_activity+current_mec_noise,current_lrate_hpc_mec),hpc_mec_weights_mean)
#lec_hpc_weights = normalize_weight(learn_weight(lec_hpc_weights,current_lec_activity+current_lec_noise,current_hpc_activity+current_hpc_noise,current_lrate_lec_hpc),lec_hpc_weights_mean)
#paaa = 0
#pbbb = 0
#pccc = 0
#for pll in arange(400):
# paaa += np.corrcoef(mec_hpc_weights_t[:,pll],mec_hpc_weights[:,pll])[0,1]
# pbbb += np.corrcoef(hpc_mec_weights_t[:,pll],hpc_mec_weights[:,pll])[0,1]
# pccc += np.corrcoef(lec_hpc_weights_t[:,pll],lec_hpc_weights[:,pll])[0,1]
#mec_hpc_www_vect[ii,kk,pp] = paaa/400 #np.corrcoef(mec_hpc_weights_t.ravel(),mec_hpc_weights.ravel())[0,1] #np.mean(np.abs(mec_hpc_weights_t - mec_hpc_weights))
#hpc_mec_www_vect[ii,kk,pp] = pbbb/400#np.corrcoef(hpc_mec_weights_t.ravel(),hpc_mec_weights.ravel())[0,1] #np.mean(np.abs(hpc_mec_weights_t - hpc_mec_weights))
#lec_hpc_www_vect[ii,kk,pp] = pccc/400#np.corrcoef(lec_hpc_weights_t.ravel(),lec_hpc_weights.ravel())[0,1] #np.mean(np.abs(lec_hpc_weights_t - lec_hpc_weights))
if ((lllf[ii]>0) and (hpc_pcompl_th<1.0)):
hpc_memories.append(current_hpc_activity)
lec_act[:,np.int0(xxx[pp]),np.int0(yyy[pp])] = current_lec_activity
mec_act[:,np.int0(xxx[pp]),np.int0(yyy[pp])] = current_mec_activity
hpc_act[:,np.int0(xxx[pp]),np.int0(yyy[pp])] = current_hpc_activity
mec_act_vect.append(mec_act)
lec_act_vect.append(lec_act)
hpc_act_vect.append(hpc_act)
# %%
# 0 : learn #1
# 1 : learn #2
# 2-17 : try #1 grid cells
# 18-33 : try #1 regular
# 34-49 : try #2 grid cells
# 50-65 : try #1 regular
# 66-86 : morphing
# 87-107 : morphing-meclesion
# 108-128 : morphing input nois 1%
# 129-149 : morphing input nois 5%
# 150-170 : morphing input nois 10%
# 171-191 : morphing input nois 20%
# mooo = 0.9999 * np.ones((108)) #[0.9999,0.9999,0.9999,0.9999,0.9999,0.9999,0.9999,0.9999]
# hooo = hpc_ratio * np.ones((108)) #[0.9999,0.9999,0.9999,0.9999,0.9999,0.9999,0.9999,0.9999]
# shape_vec = 0.0 * np.ones((108)) # [0.0,0.0, 0.0,0.0,0.0,1.0,1.0,1.0 ]
# context_vec = 0.0 * np.ones((108)) #[0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.0,0.5]
# lllf = 0.0 * np.ones((108))#[1,1, 0,0,0,0,0,0,]
mooo = 0.9999 * np.ones((87)) #[0.9999,0.9999,0.9999,0.9999,0.9999,0.9999,0.9999,0.9999]
hooo = hpc_ratio * np.ones((87)) #[0.9999,0.9999,0.9999,0.9999,0.9999,0.9999,0.9999,0.9999]
shape_vec = 0.0 * np.ones((87)) # [0.0,0.0, 0.0,0.0,0.0,1.0,1.0,1.0 ]
context_vec = 0.0 * np.ones((87)) #[0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.0,0.5]
lllf = 0.0 * np.ones((87))#[1,1, 0,0,0,0,0,0,]
pzzz = np.concatenate((arange(1),arange(1),arange(0,16),arange(0,16),arange(0,16),arange(0,16),arange(0,21),arange(0,21),arange(0,21),arange(0,21),arange(0,21),arange(0,21)))
lllf[0] = 0.0
mooo[0] = mec_ratio
mooo[18:34] = mec_ratio
lllf[1] = 0.0
mooo[1] = mec_ratio
mooo[50:67] = mec_ratio
mooo[66:] = mec_ratio
#hooo[87:108] = 0.0
shape_vec[1] = 1.0
shape_vec[34:66] = 1.0
shape_vec[66:87]=np.linspace(0.0,1.0,21)
#shape_vec[87:108]=np.linspace(0.0,1.0,21)
#shape_vec[108:129]=np.linspace(0.0,1.0,21)
#shape_vec[129:150]=np.linspace(0.0,1.0,21)
#shape_vec[150:171]=np.linspace(0.0,1.0,21)
#shape_vec[171:192]=np.linspace(0.0,1.0,21)
nono = 0.0 * np.ones((87))
#nono[108:129] = 0.01
#nono[129:150] = 0.05
#nono[150:171] = 0.10
#nono[171:192] = 0.20
# %%
#input_noise_vect = [0.0,0.10,0.20,0.30,0.40,0.50,0.60,0.7,0.8,0.9,0.99]
input_noise_vect = [0,1,2,3,4,5,6,7,8,9,10,11,12]
num_runsss = len(input_noise_vect)
corrVectMEC1 = -1* ones(num_runsss)
corrVectHPC1 = -1* ones(num_runsss)
corrVectMECGRID1 = -1* ones(num_runsss)
corrVectHPCGRID1 = -1* ones(num_runsss)
corrVectMECvsGRID1 = -1* ones(num_runsss)
corrVectMEC2 = -1* ones(num_runsss)
corrVectHPC2 = -1* ones(num_runsss)
corrVectMECGRID2 = -1* ones(num_runsss)
corrVectHPCGRID2 = -1* ones(num_runsss)
corrVectMECvsGRID2 = -1* ones(num_runsss)
corrVectMECx = -1* ones(num_runsss)
corrVectHPCx = -1* ones(num_runsss)
corrVectMECGRIDx = -1* ones(num_runsss)
corrVectHPCGRIDx = -1* ones(num_runsss)
dist_pf1 = -1*ones((num_runsss,16))
dist_pf2 = -1*ones((num_runsss,16))
pvCorrelationCurveHPC1 = -1*ones((num_runsss,21))
pvCorrelationCurveMEC1 = -1*ones((num_runsss,21))
pvCorrelationCurveHPC2 = -1*ones((num_runsss,21))
pvCorrelationCurveMEC2 = -1*ones((num_runsss,21))
pvCorrelationCurveHPC = -1*ones((num_runsss,21))
pvCorrelationCurveMEC = -1*ones((num_runsss,21))
for sessions in arange(num_runsss):
np.random.seed(seed_path);
nono = input_noise_vect[sessions] * np.ones((87))
print("session %d of %d" % (sessions,num_runsss))
#if(sessions==0):
# lllf[0] = 0.0
# lllf[1] = 0.0
#else:
# lllf[0] = 1.0
# lllf[1] = 1.0
#pzzz[0] = sessions + 16 + pre_runs
#pzzz[1] = sessions + 16 + pre_runs + num_runsss
pzzz[0] = 16 + pre_runs
pzzz[1] = 16 + pre_runs + 1
lec_act_vect = []
mec_act_vect = []
hpc_act_vect = []
mec_inact_vect = np.zeros((len(shape_vec),mec_numcells,arena_binsize[0],arena_binsize[1]))
hpc_inact_vect = np.zeros((len(shape_vec),hpc_numcells,arena_binsize[0],arena_binsize[1]))
lec_inact_vect = np.zeros((len(shape_vec),lec_numcells,arena_binsize[0],arena_binsize[1]))
for ii in arange(len(shape_vec)):
print("shape %d of %d" % (ii,len(shape_vec)))
mec_ratio = mooo[ii]
hpc_ratio = hooo[ii]
lec_act = zeros((lec_numcells,arena_binsize[0],arena_binsize[1]))
mec_act = zeros((mec_numcells,arena_binsize[0],arena_binsize[1]))
hpc_act = zeros((hpc_numcells,arena_binsize[0],arena_binsize[1]))
xxx,yyy = np.meshgrid(arange(arena_binsize[0]),arange(arena_binsize[0]))
xxx = xxx.ravel()
yyy = yyy.ravel()
#ppp = np.random.permutation(len(xxx))
ppp = popo[pzzz[ii]]
xxx = xxx[ppp]
yyy = yyy[ppp]
if((conna == True) and (ii == 1)):
current_pos = current_pos - array((5,0))
else:
current_pos = array((xxx[0],yyy[0]))
current_hpc_activity = np.zeros(hpc_numcells)
if ((ii!=1) or (conna == False)):
current_mec_activity = np.zeros(mec_numcells)
current_hpc_activity = np.zeros(hpc_numcells)
current_mec_activity = np.zeros(mec_numcells)
current_context = context_vec[ii]
current_shape = shape_vec[ii]
current_vector = lec_whichone(lec_type,lec_change,current_context,current_shape)
base_lec = np.zeros(current_vector.shape)
base_lec = lec_activity[0].copy()
base_lec[current_vector==2] = lec_activity[1][current_vector==2]
#set the random seed
np.random.seed(seed_path+ii)
tokikila = np.random.permutation(len(xxx)-1)+1;
tokikila = tokikila[0:nono[ii]];
for pp in arange(len(xxx)):
print("aaa %d of %d" % (pp,len(xxx)))
current_pos_old = current_pos
current_pos = array((xxx[pp],yyy[pp]))
current_speed = current_pos - current_pos_old
current_lec_activity = np.random.uniform(0,1,(lec_numcells))
if(size(np.where(tokikila==pp))==0):
current_lec_activity = base_lec[:,current_pos[0],current_pos[1]]
current_lec_noise = np.random.uniform(0.0,lec_noise,current_lec_activity.shape)
current_mec_noise = np.random.uniform(0.0,mec_noise,current_mec_activity.shape)
current_hpc_noise = np.random.uniform(0.0,hpc_noise,current_hpc_activity.shape)
lec_inact_vect[ii,:,xxx[pp],yyy[pp]] = current_lec_activity
for kk in arange(theta_cycles):
if (kk>0): current_speed = array((0,0))
current_mec_input = (current_mec_activity+current_mec_noise)
if (mec_ratio>0.0):
for jj in arange(mec_blocks):
gxx,gyy = meshgrid(arange(mec_blocksize[jj])+(-1)*current_speed[0],arange(mec_blocksize[jj])+(-1)*current_speed[1])
gyy[mod(divide(gxx-mod(gxx,mec_blocksize[jj]),mec_blocksize[jj]),2)>0] = gyy[mod(divide(gxx-mod(gxx,mec_blocksize[jj]),mec_blocksize[jj]),2)>0] + floor(mec_blocksize[jj]/2)
gxx = int0(mod(gxx,mec_blocksize[jj]))
gyy = int0(mod(gyy,mec_blocksize[jj]))
current_mec_input[mec_indexlist[jj]] = current_mec_input[mec_indexlist[jj]][gyy,gxx]
h_h = np.dot(current_hpc_activity+current_hpc_noise,hpc_mec_weights)
if(np.max(h_h)>0.0):
h_h = h_h/np.max(h_h)
h_h[isnan(h_h)] = 0.0
current_mec_input = (1-mec_ratio)*h_h + mec_ratio*current_mec_input
current_lec_noise = np.random.uniform(0.0,lec_noise,current_lec_activity.shape)
current_mec_noise = np.random.uniform(0.0,mec_noise,current_mec_activity.shape)
current_hpc_noise = np.random.uniform(0.0,hpc_noise,current_hpc_activity.shape)
for jj in arange(mec_blocks):
current_mec_activity[mec_indexlist[jj]] = (current_mec_input[mec_indexlist[jj]] - current_emax*np.max(current_mec_input[mec_indexlist[jj]]))
current_mec_activity[current_mec_activity<0] = 0.0
current_mec_activity[mec_indexlist[jj]] /= np.max(current_mec_activity[mec_indexlist[jj]])
current_mec_activity[isnan(current_mec_activity)] = 0.0
mec_inact_vect[ii,mec_indexlist[jj],xxx[pp],yyy[pp]] = current_mec_activity[mec_indexlist[jj]]
h_l = np.dot(current_lec_activity+current_lec_noise,lec_hpc_weights)
h_l = h_l/np.max(h_l)
h_l[isnan(h_l)] = 0.0
h_m = np.dot(current_mec_activity+current_mec_noise,mec_hpc_weights)
if(np.max(h_m)>0.0):
h_m = h_m/np.max(h_m)
h_m[isnan(h_m)] = 0.0
current_hpc_input = (1-hpc_ratio)*h_l + hpc_ratio*h_m
if (kk>0):
ddd = current_hpc_activity * 0
for mm in arange(len(hpc_memories)):
ccc = corrcoef(hpc_memories[mm],current_hpc_activity+current_hpc_noise)[0][1]
if ccc<hpc_pcompl_th:
ccc=0
else:
ddd += hpc_memories[mm] #ccc=1
if (np.max(ddd) > 0):
ddd = ddd/np.max(ddd)
ddd[isnan(ddd)] = 0.0
current_hpc_input = (1-mec_ratio)*current_hpc_input + mec_ratio*ddd
current_hpc_activity = (current_hpc_input - current_emax*np.max(current_hpc_input))
current_hpc_activity[current_hpc_activity<0] = 0.0
current_hpc_activity /= np.max(current_hpc_activity)
current_hpc_activity[current_hpc_activity<current_emax_plast] = 0
#
hpc_inact_vect[ii,:,xxx[pp],yyy[pp]] = current_hpc_activity
if (lllf[ii]>0):
lec_hpc_weights = normalize_weight(learn_weight(lec_hpc_weights,current_lec_activity+current_lec_noise,current_hpc_activity+current_hpc_noise,current_lrate_lec_hpc),lec_hpc_weights_mean)
mec_hpc_weights = normalize_weight(learn_weight(mec_hpc_weights,current_mec_activity+current_mec_noise,current_hpc_activity+current_hpc_noise,current_lrate_mec_hpc),mec_hpc_weights_mean)
hpc_mec_weights = normalize_weight(learn_weight(hpc_mec_weights,current_hpc_activity+current_hpc_noise,current_mec_activity+current_mec_noise,current_lrate_hpc_mec),hpc_mec_weights_mean)
if ((lllf[ii]>0) and (hpc_pcompl_th<1.0)):
hpc_memories.append(current_hpc_activity)
lec_act[:,xxx[pp],yyy[pp]] = current_lec_activity
mec_act[:,xxx[pp],yyy[pp]] = current_mec_activity
hpc_act[:,xxx[pp],yyy[pp]] = current_hpc_activity
mec_act_vect.append(mec_act)
lec_act_vect.append(lec_act)
hpc_act_vect.append(hpc_act)
# 0 : learn #1
# 1 : learn #2
# 2-17 : try #1 grid cells
# 18-33 : try #1 regular
# 34-49 : try #2 grid cells
# 50-65 : try #1 regular
ooo1a = np.zeros((16,16))
ooo2a = np.zeros((16,16))
ooo3a = np.zeros((16,16))
ooo4a = np.zeros((16,16))
ooo5a = np.zeros((16))
ooo1b = np.zeros((16,16))
ooo2b = np.zeros((16,16))
ooo3b = np.zeros((16,16))
ooo4b = np.zeros((16,16))
ooo5b = np.zeros((16))
ooo1c = np.zeros((16,16))
ooo2c = np.zeros((16,16))
ooo3c = np.zeros((16,16))
ooo4c = np.zeros((16,16))
pfdist1 = np.zeros((16))
pfdist2 = np.zeros((16))
#ooo1d = np.zeros((16,16))
#ooo2d = np.zeros((16,16))
#ooo3d = np.zeros((16,16))
for xx in arange(16):
pfdist1 = pfdist1 + np.histogram(np.sum(np.sum(hpc_inact_vect[xx+18,:,:,:]>0,axis=1),axis=1),arange(17))[0]
pfdist2 = pfdist2 + np.histogram(np.sum(np.sum(hpc_inact_vect[xx+50,:,:,:]>0,axis=1),axis=1),arange(17))[0]
vvv5a = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv5b = np.zeros((arena_binsize[0],arena_binsize[1]))
for ii in arange(arena_binsize[0]):
for jj in arange(arena_binsize[1]):
vvv5a[ii,jj] = np.corrcoef(mec_inact_vect[xx+2,:,ii,jj],mec_inact_vect[xx+18,:,ii,jj])[0,1]
vvv5b[ii,jj] = np.corrcoef(mec_inact_vect[xx+34,:,ii,jj],mec_inact_vect[xx+50,:,ii,jj])[0,1]
ooo5a[xx] = np.mean(vvv5a)
ooo5b[xx] = np.mean(vvv5b)
for yy in arange(xx,16):
vvv1a = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv2a = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv3a = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv4a = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv1b = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv2b = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv3b = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv4b = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv1c = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv2c = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv3c = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv4c = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv1d = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv2d = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv3d = np.zeros((arena_binsize[0],arena_binsize[1]))
vvv4d = np.zeros((arena_binsize[0],arena_binsize[1]))
for ii in arange(arena_binsize[0]):
for jj in arange(arena_binsize[1]):
vvv1a[ii,jj] = np.corrcoef(mec_inact_vect[xx+2,:,ii,jj],mec_inact_vect[yy+2,:,ii,jj])[0,1]
vvv2a[ii,jj] = np.corrcoef(hpc_inact_vect[xx+2,:,ii,jj],hpc_inact_vect[yy+2,:,ii,jj])[0,1]
vvv3a[ii,jj] = np.corrcoef(mec_inact_vect[xx+18,:,ii,jj],mec_inact_vect[yy+18,:,ii,jj])[0,1]
vvv4a[ii,jj] = np.corrcoef(hpc_inact_vect[xx+18,:,ii,jj],hpc_inact_vect[yy+18,:,ii,jj])[0,1]
vvv1b[ii,jj] = np.corrcoef(mec_inact_vect[xx+34,:,ii,jj],mec_inact_vect[yy+34,:,ii,jj])[0,1]
vvv2b[ii,jj] = np.corrcoef(hpc_inact_vect[xx+34,:,ii,jj],hpc_inact_vect[yy+34,:,ii,jj])[0,1]
vvv3b[ii,jj] = np.corrcoef(mec_inact_vect[xx+50,:,ii,jj],mec_inact_vect[yy+50,:,ii,jj])[0,1]
vvv4b[ii,jj] = np.corrcoef(hpc_inact_vect[xx+50,:,ii,jj],hpc_inact_vect[yy+50,:,ii,jj])[0,1]
vvv1c[ii,jj] = np.corrcoef(mec_inact_vect[xx+2,:,ii,jj],mec_inact_vect[yy+34,:,ii,jj])[0,1]
vvv2c[ii,jj] = np.corrcoef(hpc_inact_vect[xx+2,:,ii,jj],hpc_inact_vect[yy+34,:,ii,jj])[0,1]
vvv3c[ii,jj] = np.corrcoef(mec_inact_vect[xx+18,:,ii,jj],mec_inact_vect[yy+50,:,ii,jj])[0,1]
vvv4c[ii,jj] = np.corrcoef(hpc_inact_vect[xx+18,:,ii,jj],hpc_inact_vect[yy+50,:,ii,jj])[0,1]
vvv1d[ii,jj] = np.corrcoef(mec_inact_vect[xx+34,:,ii,jj],mec_inact_vect[yy+2,:,ii,jj])[0,1]
vvv2d[ii,jj] = np.corrcoef(hpc_inact_vect[xx+34,:,ii,jj],hpc_inact_vect[yy+2,:,ii,jj])[0,1]
vvv3d[ii,jj] = np.corrcoef(mec_inact_vect[xx+50,:,ii,jj],mec_inact_vect[yy+18,:,ii,jj])[0,1]
vvv4d[ii,jj] = np.corrcoef(hpc_inact_vect[xx+50,:,ii,jj],hpc_inact_vect[yy+18,:,ii,jj])[0,1]
ooo1a[xx,yy] = np.mean(vvv1a)
ooo1a[yy,xx] = np.mean(vvv1a)
ooo2a[xx,yy] = np.mean(vvv2a)
ooo2a[yy,xx] = np.mean(vvv2a)
ooo3a[xx,yy] = np.mean(vvv3a)
ooo3a[yy,xx] = np.mean(vvv3a)
ooo4a[xx,yy] = np.mean(vvv4a)
ooo4a[yy,xx] = np.mean(vvv4a)
ooo1b[xx,yy] = np.mean(vvv1b)
ooo1b[yy,xx] = np.mean(vvv1b)
ooo2b[xx,yy] = np.mean(vvv2b)
ooo2b[yy,xx] = np.mean(vvv2b)
ooo3b[xx,yy] = np.mean(vvv3b)
ooo3b[yy,xx] = np.mean(vvv3b)
ooo4b[xx,yy] = np.mean(vvv4b)
ooo4b[yy,xx] = np.mean(vvv4b)
ooo1c[xx,yy] = np.mean(vvv1c)
ooo1c[yy,xx] = np.mean(vvv1d)
ooo2c[xx,yy] = np.mean(vvv2c)
ooo2c[yy,xx] = np.mean(vvv2d)
ooo3c[xx,yy] = np.mean(vvv3c)
ooo3c[yy,xx] = np.mean(vvv3d)
ooo4c[xx,yy] = np.mean(vvv4c)
ooo4c[yy,xx] = np.mean(vvv4d)
corrVectMECGRID1[sessions] = np.mean(ooo1a)
corrVectHPCGRID1[sessions] = np.mean(ooo2a)
corrVectMEC1[sessions] = np.mean(ooo3a)
corrVectHPC1[sessions] = np.mean(ooo4a)
corrVectMECvsGRID1[sessions] = np.mean(ooo5a)
corrVectMECGRID2[sessions] = np.mean(ooo1b)
corrVectHPCGRID2[sessions] = np.mean(ooo2b)
corrVectMEC2[sessions] = np.mean(ooo3b)
corrVectHPC2[sessions] = np.mean(ooo4b)
corrVectMECvsGRID2[sessions] = np.mean(ooo5b)
corrVectMECGRIDx[sessions] = np.mean(ooo1c)
corrVectHPCGRIDx[sessions] = np.mean(ooo2c)
corrVectMECx[sessions] = np.mean(ooo3c)
corrVectHPCx[sessions] = np.mean(ooo4c)
dist_pf1[sessions,:] = pfdist1
dist_pf2[sessions,:] = pfdist2
for xx in arange(21):
ooo1a = np.zeros(arena_binsize)
ooo2a = np.zeros(arena_binsize)
ooo1b = np.zeros(arena_binsize)
ooo2b = np.zeros(arena_binsize)
ooo3a = np.zeros(arena_binsize)
ooo3b = np.zeros(arena_binsize)
for ii in arange(arena_binsize[0]):
for jj in arange(arena_binsize[1]):
ooo1a[ii,jj] = np.corrcoef(hpc_inact_vect[66,:,ii,jj],hpc_inact_vect[xx+66,:,ii,jj])[0,1]
ooo1b[ii,jj] = np.corrcoef(mec_inact_vect[66,:,ii,jj],mec_inact_vect[xx+66,:,ii,jj])[0,1]
ooo2a[ii,jj] = np.corrcoef(hpc_inact_vect[86,:,ii,jj],hpc_inact_vect[86-xx,:,ii,jj])[0,1]
ooo2b[ii,jj] = np.corrcoef(mec_inact_vect[86,:,ii,jj],mec_inact_vect[86-xx,:,ii,jj])[0,1]
ooo3a[ii,jj] = np.mean((ooo1a[ii,jj],ooo2a[ii,jj]))
ooo3b[ii,jj] = np.mean((ooo1b[ii,jj],ooo2b[ii,jj]))
pvCorrelationCurveHPC1[sessions,xx] = np.mean(ooo1a)
pvCorrelationCurveMEC1[sessions,xx] = np.mean(ooo1b)
pvCorrelationCurveHPC2[sessions,xx] = np.mean(ooo2a)
pvCorrelationCurveMEC2[sessions,xx] = np.mean(ooo2b)
pvCorrelationCurveHPC[sessions,xx] = np.mean(ooo3a)
pvCorrelationCurveMEC[sessions,xx] = np.mean(ooo3b)
if (acts==True):
actvLec1 = lec_inact_vect[66,0:100,:,:]
actvLec2 = lec_inact_vect[86,0:100,:,:]
actvMec1 = mec_inact_vect[66,0:100,:,:]
actvMec2 = mec_inact_vect[86,0:100,:,:]
actvHpc1 = hpc_inact_vect[66,:,:,:]
actvHpc2 = hpc_inact_vect[86,:,:,:]
with gzip.open(filenames.fileRunPickle(listofvalues,simulation_num,9)+'z', 'wb') as ff:
pickle.dump([actvLec1,actvLec2,actvMec1,actvMec2,actvHpc1,actvHpc2] , ff)
with gzip.open(filenames.fileRunPickle(listofvalues,simulation_num,0)+'z', 'wb') as ff:
pickle.dump([corrVectMECGRID1,corrVectHPCGRID1,corrVectMEC1,corrVectHPC1,corrVectMECvsGRID1] , ff)
with gzip.open(filenames.fileRunPickle(listofvalues,simulation_num,1)+'z', 'wb') as ff:
pickle.dump([corrVectMECGRID2,corrVectHPCGRID2,corrVectMEC2,corrVectHPC2,corrVectMECvsGRID2] , ff)
with gzip.open(filenames.fileRunPickle(listofvalues,simulation_num,2)+'z', 'wb') as ff:
pickle.dump([corrVectMECGRIDx,corrVectHPCGRIDx,corrVectMECx,corrVectHPCx] , ff)
with gzip.open(filenames.fileRunPickle(listofvalues,simulation_num,3)+'z', 'wb') as ff:
pickle.dump([dist_pf1,dist_pf2] , ff)
with gzip.open(filenames.fileRunPickle(listofvalues,simulation_num,4)+'z', 'wb') as ff:
pickle.dump([pvCorrelationCurveHPC,pvCorrelationCurveHPC1,pvCorrelationCurveHPC2,pvCorrelationCurveMEC,pvCorrelationCurveMEC1,pvCorrelationCurveMEC2] , ff)
if __name__ == "__main__":
main(sys.argv[1:])