AuG_type = 'leaky_AuG'
print('Leaky-AuGMEnT')
else:
AuG_type = 'AuGMEnT'
print('Standard AuGMEnT')
folder = 'DATA'
Q_m = np.zeros((N_sim, N_train * (d + 1), A))
RPE_v = np.zeros((N_sim, N_train))
resp_v = np.zeros((N_sim, N_train * (d + 1)))
for n in np.arange(N_sim):
print('SIMULATION ', n + 1)
model = AuGMEnT(S, R, M, A, alpha, beta, discount, eps, g, leak, rew,
dic_stim, dic_resp)
conv_ep[n], Q_m[n, :, :], RPE_v[n, :], _ = model.training_seq_pred_2(
S_train, O_train, dic_stim, dic_resp, verb)
print(np.mean(conv_ep))
str_conv = folder + '/' + AuG_type + '_' + task + '_ALL_CONV_2.txt'
np.savetxt(str_conv, conv_ep)
str_Q = folder + '/' + AuG_type + '_' + task + '_ALL_Q_2.txt'
np.savetxt(str_Q, np.reshape(Q_m, (N_train * (d + 1) * N_sim, A)))
str_RPE = folder + '/' + AuG_type + '_' + task + '_ALL_RPE_2.txt'
np.savetxt(str_RPE, RPE_v)
stop = True
criterion = 'strong'
verb = 0
do_test = True
perc_expl = np.zeros((N_sim))
perc_no_expl = np.zeros((N_sim))
perc_soft = np.zeros((N_sim))
for n in np.arange(N_sim):
print('SIMULATION ', n + 1)
model = AuGMEnT(S, R, M, A, alpha, beta, discount, eps, g, leak, rew,
dic_stim, dic_resp, prop)
E[n, :], conv_ep[n], REW[n, :] = model.training_12AX(
N_trial, p_target, criterion, stop, verb, policy, stoc, t_weighted,
e_weighted)
print('\t CONVERGED AT TRIAL ', conv_ep[n])
if do_test:
N_test = 1000
perc_expl[n], perc_no_expl[n], perc_soft[n] = model.test(
N_test, p_target)
print('Percentage of correct trials during test (exploration): ',
perc_expl[n], '%')
print('Percentage of correct trials during test (no exploration): ',
perc_no_expl[n], '%')
print('Length = ', d + 2)
dic_stim, dic_resp = get_dictionary(d)
## CONSTRUCTION OF THE AuGMEnT NETWORK
S = d + 2 # dimension of the input = number of possible stimuli
R = 3 # dimension of the regular units
M = 4 # dimension of the memory units
A = 2 # dimension of the activity units = number of possible responses
for n in np.arange(N_vec[d_cont]):
print('SIMULATION ', n + 1)
model = AuGMEnT(S, R, M, A, alpha, beta, discount, eps, g, leak, rew,
dic_stim, dic_resp)
E[n, :], conv_ep[n] = model.training_seq_pred(N_train, d, stop, verb)
#print('\t CONVERGED AT TRIAL ', conv_ep[n])
if N_test != 0:
perc[n] = model.test_seq_pred(N_test, d, verb)
print('Percentage of correct responses during test: ', perc, '%')
conv_ep_reduced = np.delete(conv_ep, np.where(conv_ep == 0))
AVG_CONV[d_cont] = np.mean(conv_ep_reduced)
SD_CONV[d_cont] = np.std(conv_ep_reduced)
print('Average convergence time for trials with length ', d + 2, ':\t',
AVG_CONV[d_cont], ' (', SD_CONV[d_cont], ')')
d_cont += 1
N_sim = 1
E = np.zeros((N_sim, N_trial))
conv_ep = np.zeros((N_sim))
perc = np.zeros((N_sim))
do_weight_plots = True
S_tot = np.shape(S_train)[0]
Q_m = np.zeros((N_sim, S_tot, A))
RPE_v = np.zeros((N_sim, S_tot - 1))
for n in np.arange(N_sim):
print('SIMULATION ', n + 1)
model = AuGMEnT(S, R, M, A, alpha, beta, discount, eps, g, leak, rew,
dic_stim, dic_resp, prop)
conv_ep[n], Q_m[n, :], RPE_v[n, :] = model.training_12AX_2(
S_train, O_train)
print('\t CONVERGED AT TRIAL ', conv_ep[n])
cues_vec = []
values_vec = list(dic_stim.values())
for l in values_vec:
cues_vec.append(l + '+')
for l in values_vec:
cues_vec.append(l + '-')
mem_vec = []
for i in range(M):
mem_vec.append('M' + str(i + 1))
N_sim = 10
E = np.zeros((N_sim, N_trial))
conv_ep = np.zeros((N_sim))
perc = np.zeros((N_sim))
stop = True
criterion = 'strong'
do_test = True
for n in np.arange(N_sim):
print('SIMULATION ', n + 1)
model = AuGMEnT(S, R, M, A, alpha, beta, discount, eps, g, leak, rew,
dic_stim, dic_resp, prop)
E[n, :], conv_ep[n] = model.training_12AX(N_trial, p_target, criterion,
stop)
print('\t CONVERGED AT TRIAL ', conv_ep[n])
if do_test:
N_test = 1000
perc[n] = model.test(N_test, p_target)
print('Percentage of correct trials during test: ', perc, '%')
folder = 'DATA'
str_conv = folder + '/AuGMEnT_' + task + '_conv.txt'
np.savetxt(str_conv, conv_ep)
E_mean = np.mean(np.reshape(E, (-1, 50)), axis=1)
AuG_type = 'AuGMEnT'
print('Standard AuGMEnT')
# reward settings
rew = 'RL'
prop = 'std'
N_sim = 1
E = np.zeros((N_sim,N_trial))
conv_ep = np.zeros((N_sim))
stop = True
verb = 1
for n in np.arange(N_sim):
print('SIMULATION ', n+1)
model = AuGMEnT(S,R,M,A,alpha,beta,discount,eps,g,leak,rew,dic_stim,dic_resp,prop)
E[n,:],conv_ep[n] = model.training_tXOR(N_trial,stop,verb)
print('\t CONVERGED AT TRIAL ', conv_ep[n])
folder = 'DATA'
str_conv = folder+'/'+AuG_type+'_'+task+'_conv.txt'
np.savetxt(str_conv,conv_ep)
E_mean = np.mean(np.reshape(E,(-1,50)),axis=1)
str_err = folder+'/'+AuG_type+'_'+task+'_error.txt'
np.savetxt(str_err,E_mean)
E_fix = np.zeros((N_sim, N_trial))
E_go = np.zeros((N_sim, N_trial))
conv_ep = np.zeros((N_sim))
perc_go = np.zeros((N_sim))
perc_fix = np.zeros((N_sim))
stop = True
S_tr, O_tr, _, _, dic_stim, dic_resp = data_construction(N=1, perc_training=1)
for n in np.arange(N_sim):
print('SIMULATION ', n + 1)
S_tr, O_tr, _, _, _, _ = data_construction(N=N_trial, perc_training=1)
model = AuGMEnT(S, R, M, A, alpha, beta, discount, eps, g, leak, rew,
dic_stim, dic_resp, prop)
_, E_go[n, :], conv_ep[n] = model.training_saccade(N_trial, S_tr, O_tr,
reset_cond, verb,
shape_fac, stop)
print('\t CONVERGED AT TRIAL ', conv_ep[n])
if do_test == True:
S_test, O_test, _, _, _, _ = data_construction(N=100, perc_training=1)
perc_fix[n], perc_go[n] = model.test_saccade(S_test, O_test,
reset_cond)
print('Percentage of correct FIX responses during test: ', perc_fix,
'%')
print('Percentage of correct GO responses during test: ', perc_go, '%')