def plotAllRates(children, analysis):
#plots rate of distuption events for users who used device for more than 21 days
#plots the rate how often device was used with best possible reaction
#figures are created with matplotlibe and saved as .jpg files in folder "figures2"
count=0
for cid in analysis.keys():
if len(analysis[cid].use)>20:
fig, ax=plt.subplots()
ax.set_title('rate of events for the last 7 days')
ax.set_xlabel('number of days')
ax.set_ylabel('rate')
rate=analysis[cid].rate_dst_event
use=analysis[cid].use
reactgood_rate=analysis[cid].complete_dev_rate['ReactGood']
dev_rate=analysis[cid].dev_use_rate
#print rate
#print rate, use
ax.plot(use, rate, 'bs-')
ax.plot(use, dev_rate, 'k,-')
ax.plot(use, reactgood_rate, 'go-')
init=children[cid].initial
value=init.month_count*init.night_count/30.0
ax.plot([min(use) - 1],value, 'ro')
ax.set_xlim((-2, max(use)+1))
ax.set_ylim((-0.2, max([value] + rate)*1.2))
#fig.savefig('figures2/dev_rate'+str(cid)+'.jpg')
#plt.close(fig)
plt.show()
count+=1
if count>10:
break
def clusterLongTermUsers(analysis):
#try clustering longterm users
X=[]
Y=[]
Y2=[]
Y3=[]
for cid in analysis.keys():
ch=analysis[cid]
if len(ch.biggest_improvement)>0:
vector=[ch.std_time_for_bed, ch.std_event1_time, ch.std_time_to_event1, ch.avg_time_to_event1, ch.std_offset_for_device_start, ch.avg_offset_for_device_start, ch.std_device_start_time, ch.std_device_interval, ch.use_rate, len(ch.use)]
valid=True
for el in vector:
if numpy.isnan(el):
#print 'child with id=', cid, 'invalid data point is', vector
valid=False
if valid:
if ch.corr_dev_reactgood is None or numpy.isnan(ch.corr_dev_reactgood):
continue
else:
Y3.append(ch.corr_dev_reactgood)
X.append(vector)
Y.append(ch.biggest_improvement[2])
Y2.append(ch.biggest_improvement[0])
X=numpy.array(X)
k_means = cluster.KMeans(n_clusters=3)
k_means.fit(X)
# print "label, rate change, final rate, correlation"
# y=k_means.predict(X)
# for i in range(len(y)):
# print y[i], Y[i], Y2[i], Y3[i]
X2=numpy.array(zip(Y3,numpy.array(X)[:,5], numpy.array(X)[:,5]))
k_means = cluster.KMeans(n_clusters=5)
k_means.fit(X2)
y=k_means.predict(X2)
plt.figure()
plt.scatter(X2[:,0],X2[:,1], c=y, edgecolor='face')
plt.xlabel('correlation')
plt.ylabel('std offset')
plt.title('long term user analysis')
plt.colorbar()
plt.figure()
plt.scatter(X2[:,0],X2[:,2], c=y, edgecolor='face')
plt.xlabel('correlation')
plt.ylabel('mean offset')
plt.title('long term user analysis')
plt.colorbar()
plt.show()
def dataForML(analysis):
#prepares data to be used for machine learning analysis
index=analysis.keys()
random.shuffle(index) #shuffle users to ensure that data for ML is fair (no hidden trends based on when users joined the system)
#print "Does shuffle work", index
#initialize machine learning variables
X=[]
Y=[]
Y2=[]
Y3=[]
Y4=[]
#print "problem with analysis data structure?"
for cid in index:
ch=analysis[cid]
#print cid, ch.std_time_for_bed
if len(ch.biggest_improvement)>0:
vector=[ch.std_time_for_bed, ch.std_event1_time,
ch.std_time_to_event1, ch.avg_time_to_event1,
ch.std_offset_for_device_start, ch.avg_offset_for_device_start,
ch.std_device_start_time, ch.std_device_interval,
ch.use_rate, len(ch.use),
len(ch.vibration_use)]
#vector=[ch.use_rate]
#print 'time_for_bed?', cid, ch.std_time_for_bed
valid=True
for el in vector:
if numpy.isnan(el):
#print 'child with id=', cid, 'invalid data point is', vector
valid=False
if ch.corr_dev_reactgood is None or numpy.isnan(ch.corr_dev_reactgood):
valid = False
if valid:
X.append(vector)
Y.append(ch.biggest_improvement[2])
Y2.append(ch.biggest_improvement[0])
Y3.append(ch.corr_dev_reactgood)
Y4.append(int(ch.works_flag))
return X, Y, Y2, Y3, Y4
def dataForInitialAnalysis(children, analysis):
#generates data for machine learning with "initial" data (prior history on sleep disorder) provided from the users
index=analysis.keys()
random.shuffle(index)
#print "Does shuffle work", index
#collect data in machine learning format
X=[]
Y=[]
for i in index:
init=children[i].initial
if analysis[i].start_date:
s_date=dateDiff(start, analysis[i].start_date)
#print i, s_date, analysis[i].start_date
X.append([init.month_count, init.night_count, s_date])
Y.append(len(analysis[i].use))
X=numpy.array(X)
Y=numpy.array(Y)
print 'mean', numpy.mean(X[:,0])
print 'sdt', numpy.std(X[:,0])
print 'sample X', X[:10]
plotInitData(X,Y)
return index, X, Y
def plotRate(children, analysis):
#plots rate of distuption events for users who used device for more than 21 days
#plots done with matplotlib and saved as .jpg's in folder "figures2"
count=0
for cid in analysis.keys():
if len(analysis[cid].use)>20:
fig, ax=plt.subplots()
ax.set_title('rate of distuption events for the last 7 days')
ax.set_xlabel('number of days')
ax.set_ylabel('dist event rate')
rate=analysis[cid].rate_dst_event
use=analysis[cid].use
#print rate
#print rate, use
ax.plot(use, rate, 'bs-')
init=children[cid].initial
value=init.month_count*init.night_count/30.0
ax.plot([min(use) - 1],value, 'ro')
ax.set_xlim((-2, max(use)+1))
ax.set_ylim((-0.2, max([value] + rate)*1.2))
fig.savefig('figures2/dev_rate'+str(cid)+'.jpg')
plt.close(fig)
#plt.show()
count+=1