def exercise1_3(N=200,P=50,tests=100):
print "Exercise 1.3:"
h = HopfieldNetwork()
error = zeros((P,tests),float)
for p in range(P):
for q in range(tests):
if p > 0:
r = np.random.randint(0,p)
else:
r = 0
error[p,q]=(h.run(N, P=p+1, mu=r, flip_ratio=0.1, bPlot=False))
print error;
stdev=np.std(error,axis=1)
meanvalues=np.mean(error,axis=1)
fig = figure()
ax1 = fig.add_subplot(111)
lp = errorbar(range(1,P+1),meanvalues,stdev,color='g')
bp = boxplot(transpose(error))
axhline(y=0.02,linewidth=1, color='r')
xticks(arange(0,P+1,P/10))
setp(lp,color='black')
ax1.set_ylim(0,0.5)
ax1.yaxis.grid(True,linestyle='-',which='major',
color=(0.2,0.2,0.2),alpha=0.5)
ax1.set_axisbelow(True)
ax1.set_title('Mean retrieval error over different'
'network realizations (N=%d, tests=%d)' % (N,tests) )
ax1.set_ylabel('Mean retrieval error')
ax1.set_xlabel('Dictonary size P')
savefig('../tex/dat/ex1_3-error_avg-N%d-P%d-Q%d.png' % (N,P,tests))
close()
def exercise1_1(N=200,P=5,c=0.2):
print "Exercise 1.1:"
h = HopfieldNetwork()
h.run(N=N, P=P, flip_ratio=c, bPlot=True)
savefig('../tex/dat/ex1_1-energy_overlap-N%d-P%d-c%d.png' % (N,P,c*100))
close()
