def expected_final_entropy(self, action, data=None):
return entropy_gains.hypotheses_expected_final_entropy(action, data)
def expected_final_entropy(self, action, data=None): return entropy_gains.hypotheses_expected_final_entropy(action, data)
def main(player, n):
#random.seed(0)
starttime = time.clock()
# if player=='adults':
# data=Data.Data(parameters.inputfile_adults)
# data.read(astext=False)
# elif player=='kids':
# data=Data.Data(parameters.inputfile_kids)
# data.read(astext=False)
#data=Data.Data(parameters.inputfile_kids)
data = Data.Data(parameters.inputfile_adults)
data.read(astext=False)
n_kids = parameters.n_kids
truncate = int(n)
#n_r_theo=parameters.n_r_theo
n_r_random = parameters.n_r_random
eg = np.zeros(len(data.get_kids()[:n_kids]))
if player in ('kids', 'adults'):
n_r = 1
for k, kid in enumerate(data.get_kids()[:n_kids]):
kidseq = data.data[kid][:truncate]
keg = entropy_gains.ave_theory_expected_entropy_gain(kidseq)[0]
#print kid, keg
eg[k] = keg
#nkegs
elif player == 'random':
n_r = n_r_random
egall = np.zeros((len(data.get_kids()[:n_kids]), n_r))
for k, kid in enumerate(data.get_kids()[:n_kids]):
for r in range(n_r_random):
rl = learners.RandomLearner()
rlseq = rl.play(min(data.get_kid_nactions(kid), truncate))
reg = entropy_gains.ave_theory_expected_entropy_gain(rlseq)[0]
#print kid, reg
eg[k] += reg
egall[k, r] = reg
eg[k] /= n_r
#[d.display() for d in rlseq]
#print reg
elif player == 'theory':
n_r = n_r_theo
egall = np.zeros((len(data.get_kids()[:n_kids]), n_r))
for k, kid in enumerate(data.get_kids()[:n_kids]):
for r in range(n_r_theo):
tl = learners.TheoryLearner()
tlseq = tl.play(min(data.get_kid_nactions(kid), truncate))
teg = entropy_gains.ave_theory_expected_entropy_gain(tlseq)[0]
#print kid, teg
eg[k] += teg
egall[k, r] = teg
eg[k] /= n_r
elif player == 'theoryfull':
n_long_kids=sum([data.get_kid_nactions(kid)>=truncate \
for kid in data.get_kids()])
eig = np.zeros((n_long_kids, 3))
tlactions = []
rlactions = []
k = 0
for ki, kid in enumerate(data.get_kids()[:n_kids]):
print 'Run for {0} actions, processing kid {1} out of {2}'.format(
truncate, ki + 1, n_kids)
if data.get_kid_nactions(kid) < truncate:
continue
#get kid's action sequence
kidseq = data.data[kid][:truncate]
#keg, kents=entropy_gains.ave_theory_expected_entropy_gain(kidseq)
#keg=entropy_gains.theory_expected_entropy_gain(kidseq[-1].action,kidseq[:-1])
keg = entropy_gains.theory_expected_final_entropy(
kidseq[-1].action, kidseq[:-1])
#print 'kid {0} entropies: {1}'.format(k,kents)
#compute optimal choice entropy gain with kid's action sequence
tl = learners.TheoryLearner()
tlaction = tl.choose_action(kidseq[:truncate - 1])
tlactions.append(tlaction)
yokedseq = kidseq[:-1] + [
Datapoint.Datapoint(tlaction, False)
] #this False is generic, shouldn't be taken into account
#tleg, tlents=entropy_gains.ave_theory_expected_entropy_gain(yokedseq)
#tleg=entropy_gains.theory_expected_entropy_gain(tlaction, kidseq[:-1])
tleg = entropy_gains.theory_expected_final_entropy(
tlaction, kidseq[:-1])
#print tlents
reg = 0
rlactions.append([])
for r in range(n_r_random):
rl = learners.RandomLearner()
#rseq=rl.play(truncate)
rlaction = rl.choose_action(kidseq[:truncate - 1])
rlactions[k].append(rlaction)
yokedseqr = kidseq[:-1] + [
Datapoint.Datapoint(rlaction, False)
] #this False is generic, shouldn't be taken into account
#reg+=entropy_gains.ave_theory_expected_entropy_gain(yokedseqr)[0]
#reg+=entropy_gains.theory_expected_entropy_gain(rlaction, kidseq[:-1])
reg += entropy_gains.theory_expected_final_entropy(
rlaction, kidseq[:-1])
reg /= n_r_random
eig[k, 0] = tleg
eig[k, 1] = reg
eig[k, 2] = keg
#print 'k: {0}, r:{1}, t:{2}'.format(keg, reg, tleg)
k += 1
elif player == 'jointfull':
n_long_kids=sum([data.get_kid_nactions(kid)>=truncate \
for kid in data.get_kids()])
eig = np.zeros((n_long_kids, 3))
tlactions = []
rlactions = []
k = 0
for ki, kid in enumerate(data.get_kids()[:n_kids]):
if data.get_kid_nactions(kid) < truncate:
continue
#get kid's action sequence
kidseq = data.data[kid][:truncate]
#keg, kents=entropy_gains.ave_theory_expected_entropy_gain(kidseq)
#keg=entropy_gains.theory_expected_entropy_gain(kidseq[-1].action,kidseq[:-1])
keg = entropy_gains.joint_expected_final_entropy(
kidseq[-1].action, kidseq[:-1])
#print 'kid {0} entropies: {1}'.format(k,kents)
#compute optimal choice entropy gain with kid's action sequence
tl = learners.JointLearner()
tlaction = tl.choose_action(kidseq[:truncate - 1])
tlactions.append(tlaction)
yokedseq = kidseq[:-1] + [
Datapoint.Datapoint(tlaction, False)
] #this False is generic, shouldn't be taken into account
#tleg, tlents=entropy_gains.ave_theory_expected_entropy_gain(yokedseq)
#tleg=entropy_gains.theory_expected_entropy_gain(tlaction, kidseq[:-1])
tleg = entropy_gains.joint_expected_final_entropy(
tlaction, kidseq[:-1])
#print tlents
reg = 0
rlactions.append([])
for r in range(n_r_random):
rl = learners.RandomLearner()
#rseq=rl.play(truncate)
rlaction = rl.choose_action(kidseq[:truncate - 1])
rlactions[k].append(rlaction)
yokedseqr = kidseq[:-1] + [
Datapoint.Datapoint(rlaction, False)
] #this False is generic, shouldn't be taken into account
#reg+=entropy_gains.ave_theory_expected_entropy_gain(yokedseqr)[0]
#reg+=entropy_gains.theory_expected_entropy_gain(rlaction, kidseq[:-1])
reg += entropy_gains.joint_expected_final_entropy(
rlaction, kidseq[:-1])
reg /= n_r_random
eig[k, 0] = tleg
eig[k, 1] = reg
eig[k, 2] = keg
#print 'k: {0}, r:{1}, t:{2}'.format(keg, reg, tleg)
k += 1
elif player == 'hypfull':
n_long_kids=sum([data.get_kid_nactions(kid)>=truncate \
for kid in data.get_kids()])
eig = np.zeros((n_long_kids, 3))
tlactions = []
rlactions = []
k = 0
for ki, kid in enumerate(data.get_kids()[:n_kids]):
if data.get_kid_nactions(kid) < truncate:
continue
#get kid's action sequence
kidseq = data.data[kid][:truncate]
#keg, kents=entropy_gains.ave_theory_expected_entropy_gain(kidseq)
#keg=entropy_gains.theory_expected_entropy_gain(kidseq[-1].action,kidseq[:-1])
keg = entropy_gains.hypotheses_expected_final_entropy(
kidseq[-1].action, kidseq[:-1])
#print 'kid {0} entropies: {1}'.format(k,kents)
#compute optimal choice entropy gain with kid's action sequence
tl = learners.HypothesesLearner()
tlaction = tl.choose_action(kidseq[:truncate - 1])
tlactions.append(tlaction)
yokedseq = kidseq[:-1] + [
Datapoint.Datapoint(tlaction, False)
] #this False is generic, shouldn't be taken into account
#tleg, tlents=entropy_gains.ave_theory_expected_entropy_gain(yokedseq)
#tleg=entropy_gains.theory_expected_entropy_gain(tlaction, kidseq[:-1])
tleg = entropy_gains.hypotheses_expected_final_entropy(
tlaction, kidseq[:-1])
#print tlents
reg = 0
rlactions.append([])
for r in range(n_r_random):
rl = learners.RandomLearner()
#rseq=rl.play(truncate)
rlaction = rl.choose_action(kidseq[:truncate - 1])
rlactions[k].append(rlaction)
yokedseqr = kidseq[:-1] + [
Datapoint.Datapoint(rlaction, False)
] #this False is generic, shouldn't be taken into account
#reg+=entropy_gains.ave_theory_expected_entropy_gain(yokedseqr)[0]
#reg+=entropy_gains.theory_expected_entropy_gain(rlaction, kidseq[:-1])
reg += entropy_gains.hypotheses_expected_final_entropy(
rlaction, kidseq[:-1])
reg /= n_r_random
eig[k, 0] = tleg
eig[k, 1] = reg
eig[k, 2] = keg
#print 'k: {0}, r:{1}, t:{2}'.format(keg, reg, tleg)
k += 1
if player in ['random', 'theory', 'kids', 'adults']:
filename = parameters.output_directory + 'out-' + player + '-' + str(
truncate) + '_tru-' + str(n_r) + '_real.txt'
np.savetxt(filename, eg)
if player in ['random', 'theory']:
filenameall = parameters.output_directory + 'all-' + player + '-' + str(
truncate) + '_tru-' + str(n_r) + '_real.txt'
np.savetxt(filenameall, egall)
if player in ['theoryfull', 'jointfull', 'hypfull']:
filename=parameters.output_directory+player+'-'+str(truncate)+'_tru-'\
+str(n_r_random)+'_rreal.txt'
np.savetxt(filename, eig)
with open(parameters.output_directory+player+'-modelactions-'+str(truncate)+'_tru-'+\
str(n_r_random)+'_rreal.txt','w') as f:
for kact in tlactions:
f.write(str(kact) + '\n')
with open(parameters.output_directory+player+'-randomactions-'+str(truncate)+'_tru-'+\
str(n_r_random)+'_rreal.txt','w') as f:
for kact in rlactions:
f.write(str(kact) + '\n')
print 'time elapsed for run {0}: {1:.0f} s'.format(
filename,
time.clock() - starttime)
def main(player, n):
#random.seed(0)
starttime=time.clock()
data=Data.Data()
data.read(astext=False)
n_kids=parameters.n_kids
truncate=int(n)
#n_r_theo=parameters.n_r_theo
n_r_random=parameters.n_r_random
eg=np.zeros(len(data.get_kids()[:n_kids]))
if player=='kids':
n_r=1
for k,kid in enumerate(data.get_kids()[:n_kids]):
kidseq=data.data[kid][:truncate]
keg=entropy_gains.ave_theory_expected_entropy_gain(kidseq)[0]
#print kid, keg
eg[k]=keg
#nkegs
elif player=='random':
n_r=n_r_random
egall=np.zeros((len(data.get_kids()[:n_kids]), n_r))
for k,kid in enumerate(data.get_kids()[:n_kids]):
for r in range(n_r_random):
rl=learners.RandomLearner()
rlseq=rl.play(min(data.get_kid_nactions(kid),truncate))
reg=entropy_gains.ave_theory_expected_entropy_gain(rlseq)[0]
#print kid, reg
eg[k]+=reg
egall[k,r]=reg
eg[k]/=n_r
#[d.display() for d in rlseq]
#print reg
elif player=='theory':
n_r=n_r_theo
egall=np.zeros((len(data.get_kids()[:n_kids]), n_r))
for k,kid in enumerate(data.get_kids()[:n_kids]):
for r in range(n_r_theo):
tl=learners.TheoryLearner()
tlseq=tl.play(min(data.get_kid_nactions(kid),truncate))
teg=entropy_gains.ave_theory_expected_entropy_gain(tlseq)[0]
#print kid, teg
eg[k]+=teg
egall[k,r]=teg
eg[k]/=n_r
elif player=='theoryfull':
n_long_kids=sum([data.get_kid_nactions(kid)>=truncate \
for kid in data.get_kids()])
eig=np.zeros((n_long_kids,3))
tlactions=[]
rlactions=[]
k=0
for ki,kid in enumerate(data.get_kids()[:n_kids]):
if data.get_kid_nactions(kid)<truncate:
continue
#get kid's action sequence
kidseq=data.data[kid][:truncate]
#keg, kents=entropy_gains.ave_theory_expected_entropy_gain(kidseq)
#keg=entropy_gains.theory_expected_entropy_gain(kidseq[-1].action,kidseq[:-1])
keg=entropy_gains.theory_expected_final_entropy(kidseq[-1].action,kidseq[:-1])
#print 'kid {0} entropies: {1}'.format(k,kents)
#compute optimal choice entropy gain with kid's action sequence
tl=learners.TheoryLearner()
tlaction=tl.choose_action(kidseq[:truncate-1])
tlactions.append(tlaction)
yokedseq=kidseq[:-1]+[Datapoint.Datapoint(tlaction, False)]#this False is generic, shouldn't be taken into account
#tleg, tlents=entropy_gains.ave_theory_expected_entropy_gain(yokedseq)
#tleg=entropy_gains.theory_expected_entropy_gain(tlaction, kidseq[:-1])
tleg=entropy_gains.theory_expected_final_entropy(tlaction, kidseq[:-1])
#print tlents
reg=0
rlactions.append([])
for r in range(n_r_random):
rl=learners.RandomLearner()
#rseq=rl.play(truncate)
rlaction=rl.choose_action(kidseq[:truncate-1])
rlactions[k].append(rlaction)
yokedseqr=kidseq[:-1]+[Datapoint.Datapoint(rlaction, False)]#this False is generic, shouldn't be taken into account
#reg+=entropy_gains.ave_theory_expected_entropy_gain(yokedseqr)[0]
#reg+=entropy_gains.theory_expected_entropy_gain(rlaction, kidseq[:-1])
reg+=entropy_gains.theory_expected_final_entropy(rlaction, kidseq[:-1])
reg/=n_r_random
eig[k,0]=tleg
eig[k,1]=reg
eig[k,2]=keg
#print 'k: {0}, r:{1}, t:{2}'.format(keg, reg, tleg)
k+=1
elif player=='jointfull':
n_long_kids=sum([data.get_kid_nactions(kid)>=truncate \
for kid in data.get_kids()])
eig=np.zeros((n_long_kids,3))
tlactions=[]
rlactions=[]
k=0
for ki,kid in enumerate(data.get_kids()[:n_kids]):
if data.get_kid_nactions(kid)<truncate:
continue
#get kid's action sequence
kidseq=data.data[kid][:truncate]
#keg, kents=entropy_gains.ave_theory_expected_entropy_gain(kidseq)
#keg=entropy_gains.theory_expected_entropy_gain(kidseq[-1].action,kidseq[:-1])
keg=entropy_gains.joint_expected_final_entropy(kidseq[-1].action,kidseq[:-1])
#print 'kid {0} entropies: {1}'.format(k,kents)
#compute optimal choice entropy gain with kid's action sequence
tl=learners.JointLearner()
tlaction=tl.choose_action(kidseq[:truncate-1])
tlactions.append(tlaction)
yokedseq=kidseq[:-1]+[Datapoint.Datapoint(tlaction, False)]#this False is generic, shouldn't be taken into account
#tleg, tlents=entropy_gains.ave_theory_expected_entropy_gain(yokedseq)
#tleg=entropy_gains.theory_expected_entropy_gain(tlaction, kidseq[:-1])
tleg=entropy_gains.joint_expected_final_entropy(tlaction, kidseq[:-1])
#print tlents
reg=0
rlactions.append([])
for r in range(n_r_random):
rl=learners.RandomLearner()
#rseq=rl.play(truncate)
rlaction=rl.choose_action(kidseq[:truncate-1])
rlactions[k].append(rlaction)
yokedseqr=kidseq[:-1]+[Datapoint.Datapoint(rlaction, False)]#this False is generic, shouldn't be taken into account
#reg+=entropy_gains.ave_theory_expected_entropy_gain(yokedseqr)[0]
#reg+=entropy_gains.theory_expected_entropy_gain(rlaction, kidseq[:-1])
reg+=entropy_gains.joint_expected_final_entropy(rlaction, kidseq[:-1])
reg/=n_r_random
eig[k,0]=tleg
eig[k,1]=reg
eig[k,2]=keg
#print 'k: {0}, r:{1}, t:{2}'.format(keg, reg, tleg)
k+=1
elif player=='hypfull':
n_long_kids=sum([data.get_kid_nactions(kid)>=truncate \
for kid in data.get_kids()])
eig=np.zeros((n_long_kids,3))
tlactions=[]
rlactions=[]
k=0
for ki,kid in enumerate(data.get_kids()[:n_kids]):
if data.get_kid_nactions(kid)<truncate:
continue
#get kid's action sequence
kidseq=data.data[kid][:truncate]
#keg, kents=entropy_gains.ave_theory_expected_entropy_gain(kidseq)
#keg=entropy_gains.theory_expected_entropy_gain(kidseq[-1].action,kidseq[:-1])
keg=entropy_gains.hypotheses_expected_final_entropy(kidseq[-1].action,kidseq[:-1])
#print 'kid {0} entropies: {1}'.format(k,kents)
#compute optimal choice entropy gain with kid's action sequence
tl=learners.HypothesesLearner()
tlaction=tl.choose_action(kidseq[:truncate-1])
tlactions.append(tlaction)
yokedseq=kidseq[:-1]+[Datapoint.Datapoint(tlaction, False)]#this False is generic, shouldn't be taken into account
#tleg, tlents=entropy_gains.ave_theory_expected_entropy_gain(yokedseq)
#tleg=entropy_gains.theory_expected_entropy_gain(tlaction, kidseq[:-1])
tleg=entropy_gains.hypotheses_expected_final_entropy(tlaction, kidseq[:-1])
#print tlents
reg=0
rlactions.append([])
for r in range(n_r_random):
rl=learners.RandomLearner()
#rseq=rl.play(truncate)
rlaction=rl.choose_action(kidseq[:truncate-1])
rlactions[k].append(rlaction)
yokedseqr=kidseq[:-1]+[Datapoint.Datapoint(rlaction, False)]#this False is generic, shouldn't be taken into account
#reg+=entropy_gains.ave_theory_expected_entropy_gain(yokedseqr)[0]
#reg+=entropy_gains.theory_expected_entropy_gain(rlaction, kidseq[:-1])
reg+=entropy_gains.hypotheses_expected_final_entropy(rlaction, kidseq[:-1])
reg/=n_r_random
eig[k,0]=tleg
eig[k,1]=reg
eig[k,2]=keg
#print 'k: {0}, r:{1}, t:{2}'.format(keg, reg, tleg)
k+=1
if player in ['random', 'theory', 'kids']:
filename=parameters.output_directory+'out-'+player+'-'+str(truncate)+'_tru-'+str(n_r)+'_real.txt'
np.savetxt(filename, eg)
if player in ['random', 'theory']:
filenameall=parameters.output_directory+'all-'+player+'-'+str(truncate)+'_tru-'+str(n_r)+'_real.txt'
np.savetxt(filenameall, egall)
if player in ['theoryfull', 'jointfull', 'hypfull']:
filename=parameters.output_directory+player+'-'+str(truncate)+'_tru-'\
+str(n_r_random)+'_rreal.txt'
np.savetxt(filename, eig)
with open(parameters.output_directory+player+'-modelactions-'+str(truncate)+'_tru-'+\
str(n_r_random)+'_rreal.txt','w') as f:
for kact in tlactions:
f.write(str(kact)+'\n')
with open(parameters.output_directory+player+'-randomactions-'+str(truncate)+'_tru-'+\
str(n_r_random)+'_rreal.txt','w') as f:
for kact in rlactions:
f.write(str(kact)+'\n')
print 'time elapsed for run {0}: {1:.0f} s'.format(filename, time.clock()-starttime)
today='141119/' #batch='ep-0.05/' batch='varyep/' data_directory=output_directory+today#+batch n_act=2 model='jointfull' filename=model+'-'+str(n_act)+'_tru-20'+'_rreal.txt' datajoi=np.loadtxt(data_directory+filename) print data_directory+filename model='hypfull' filename=model+'-'+str(n_act)+'_tru-20'+'_rreal.txt' datahyp=np.loadtxt(data_directory+filename) print data_directory+filename for i in range(len(datahyp)): print datahyp[i,2]-datahyp[i,0], datajoi[i,2]-datajoi[i,0] import entropy_gains as eg import world pa=world.possible_actions() print eg.hypotheses_expected_final_entropy(pa[0],[])-eg.hypotheses_expected_final_entropy(pa[1],[]) print eg.joint_expected_final_entropy(pa[0],[])-eg.hypotheses_expected_final_entropy(pa[1],[])
import low_model as model import world import entropy_gains as eg for action in world.possible_actions(): print action, eg.hypotheses_expected_final_entropy(action,[])