norepinephrine_wm/helper.py

   1 def make_addon(N):
   2     import string
   3     import random
   4     addon=str(''.join(random.choice(string.ascii_uppercase+string.digits) for _ in range(N)))
   5     return addon
   6
   7 def ch_dir():
   8     #change directory for data and plot outputs
   9     import os
  10     import sys
  11     root=os.getcwd()
  12     addon=make_addon(9)
  13     datadir=''
  14     if sys.platform == "linux" or sys.platform == "linux2" or sys.platform == "darwin":
  15         datadir=root+'/data/'+addon+'/' #linux or mac
  16     elif sys.platform == "win32":
  17         datadir=root+'\\data\\'+addon+'\\' #windows
  18     os.makedirs(datadir)
  19     os.chdir(datadir)
  20     return datadir
  21
  22 def empirical_dataframe():
  23     import numpy as np
  24     import pandas as pd
  25     columns=('time','drug','accuracy','trial')
  26     emp_times = [3.0,5.0,7.0,9.0]
  27     emp_dataframe = pd.DataFrame(columns=columns,index=np.arange(0, 12))
  28     pre_PHE=[0.972, 0.947, 0.913, 0.798]
  29     pre_GFC=[0.970, 0.942, 0.882, 0.766]
  30     post_GFC=[0.966, 0.928, 0.906, 0.838]
  31     post_PHE=[0.972, 0.938, 0.847, 0.666]
  32     q=0
  33     for t in range(len(emp_times)):
  34         emp_dataframe.loc[q]=[emp_times[t],'control (empirical)',np.average([pre_GFC[t],pre_PHE[t]]),0]
  35         emp_dataframe.loc[q+1]=[emp_times[t],'PHE (empirical)',post_PHE[t],0]
  36         emp_dataframe.loc[q+2]=[emp_times[t],'GFC (empirical)',post_GFC[t],0]
  37         q+=3
  38     return emp_dataframe
  39
  40 def make_cues(P):
  41     import numpy as np
  42     trials=np.arange(P['n_trials'])
  43     perceived=np.ones(P['n_trials']) #list of correctly perceived (not necessarily remembered) cues
  44     rng=np.random.RandomState(seed=P['seed'])
  45     cues=2*rng.randint(2,size=P['n_trials'])-1 #whether the cues is on the left or right
  46     for n in range(len(perceived)):
  47         if rng.rand()<P['misperceive']: perceived[n]=0
  48     return trials, perceived, cues
  49
  50
  51 '''drug approximations'''
  52 import nengo
  53 class MySolver(nengo.solvers.Solver):
  54     #When the simulator builds the network, it looks for a solver to calculate the decoders
  55     #instead of the normal least-squares solver, we define our own, so that we can return
  56     #the old decoders
  57     def __init__(self,weights): #feed in old decoders
  58         self.weights=False #they are not weights but decoders
  59         self.my_weights=weights
  60     def __call__(self,A,Y,rng=None,E=None): #the function that gets called by the builder
  61         return self.my_weights.T, dict()
  62
  63 def reset_gain_bias(P,model,sim,wm,wm_recurrent,wm_to_decision,drug):
  64     #set gains and biases as a constant multiple of the old values
  65     wm.gain = sim.data[wm].gain * P['drug_effect_biophysical'][drug][0]
  66     wm.bias = sim.data[wm].bias * P['drug_effect_biophysical'][drug][1]
  67     #set the solver of each of the connections coming out of wm using the custom MySolver class
  68     #with input equal to the old decoders. We use the old decoders because we don't want the builder
  69     #to optimize the decoders to the new gainbias/bias values, otherwise it would "adapt" to the drug
  70     wm_recurrent.solver = MySolver(sim.model.params[wm_recurrent].weights)
  71     wm_to_decision.solver=MySolver(sim.model.params[wm_to_decision].weights)
  72     #rebuild the network to affect the gain/bias change
  73     sim=nengo.Simulator(model,dt=P['dt'])
  74     return sim
  75
  76 '''dataframe initialization'''
  77 def primary_dataframe(P,sim,drug,trial,probe_wm,probe_output):
  78     import numpy as np
  79     import pandas as pd
  80     columns=('time','drug','wm','output','correct','trial')
  81     df_primary = pd.DataFrame(columns=columns, index=np.arange(0,len(P['timesteps'])))
  82     i=0
  83     for t in P['timesteps']:
  84         wm_val=np.abs(sim.data[probe_wm][t][0])
  85         output_val=sim.data[probe_output][t][0]
  86         correct=get_correct(P['cues'][trial],output_val)
  87         rt=t*P['dt_sample']
  88         df_primary.loc[i]=[rt,drug,wm_val,output_val,correct,trial]
  89         i+=1
  90     return df_primary
  91
  92 def firing_dataframe(P,sim,drug,trial,sim_wm,probe_spikes):
  93     import numpy as np
  94     import pandas as pd
  95     columns=('time','drug','neuron-trial','tuning','firing_rate')
  96     df_firing = pd.DataFrame(columns=columns, index=np.arange(0,len(P['timesteps'])*\
  97         int(P['neurons_wm']*P['frac'])))
  98     t_width = 0.2
  99     t_h = np.arange(t_width / P['dt']) * P['dt'] - t_width / 2.0
 100     h = np.exp(-t_h ** 2 / (2 * P['sigma_smoothing'] ** 2))
 101     h = h / np.linalg.norm(h, 1)
 102     j=0
 103     for nrn in range(int(P['neurons_wm']*P['frac'])):
 104         enc = sim_wm.encoders[nrn]
 105         tuning = get_tuning(P,trial,enc)
 106         spikes = sim.data[probe_spikes][:,nrn]
 107         firing_rate = np.convolve(spikes,h,mode='same')
 108         for t in P['timesteps']:
 109             rt=t*P['dt_sample']
 110             df_firing.loc[j]=[rt,drug,nrn+trial*P['neurons_wm'],tuning,firing_rate[t]]
 111             j+=1
 112         # print 'appending dataframe for neuron %s' %f
 113     return df_firing
 114
 115 def get_correct(cue,output_val):
 116     if (cue > 0.0 and output_val > 0.0) or (cue < 0.0 and output_val < 0.0): correct=1
 117     else: correct=0
 118     return correct
 119
 120 def get_tuning(P,trial,enc):
 121     cue=P['cues'][trial]
 122     enc_min_cutoff=P['enc_min_cutoff']
 123     enc_max_cutoff=P['enc_max_cutoff']
 124     if (cue > 0.0 and 0.0 < enc[0] < enc_min_cutoff) or \
 125         (cue < 0.0 and 0.0 > enc[0] > -1.0*enc_min_cutoff): tuning='superweak'
 126     if (cue > 0.0 and enc_min_cutoff < enc[0] < enc_max_cutoff) or \
 127         (cue < 0.0 and -1.0*enc_max_cutoff < enc[0] < -1.0*enc_min_cutoff): tuning='weak'
 128     elif (cue > 0.0 and enc[0] > enc_max_cutoff) or \
 129         (cue < 0.0 and enc[0] < -1.0*enc_max_cutoff): tuning='strong'
 130     else: tuning='nonpreferred'
 131     return tuning