def topic_distribution(name = None, study = None, order = None, **options):
'''Given a model p_z,p_w_z,p_d_z, we can plot the document's distribution
using p(z|d) = normalized((p(d|z)*p(z))) '''
m = microbplsa.MicrobPLSA()
m.open_model(name = name, study = study, **options) #get model from the results file
#return document's distribution
p_z_d = m.model.document_topics()
Z,N =p_z_d.shape #number of samples
if order is not None:
p_z_d = p_z_d[:,order]
n = np.arange(N)
width = 25.0/float(N) #scale width of bars by number of samples
p = [] #list of plots
colors = plt.cm.rainbow(np.linspace(0, 1, Z))
Lab = Labelling(m, ignore_continuous = False)
Lab.metadata(non_labels = ['BarcodeSequence'])
R = Lab.correlate()
labels_r = Lab.assignlabels(R,num_labels = 1)
labels, r = zip(*labels_r)
labels = [l.replace('(','\n(') for l in labels]
#sort and organize labels and topics so they are always plotted in the same order
labelsUnsorted = zipper(labels,range(0,Z))
labelsUnsorted.sort()
labels, Zrange = zip(*labelsUnsorted)
Zrange = list(Zrange)
p.append(plt.bar(n, p_z_d[Zrange[0],:], width, color=colors[0], linewidth = 0))
height = p_z_d[Zrange[0],:]
for i,z in enumerate(Zrange[1:]):
p.append(plt.bar(n, p_z_d[z,:], width, color=colors[i+1], bottom=height, linewidth = 0))
height += p_z_d[z,:]
plt.ylabel('Probability P(z|d)')
plt.xlabel('Sample')
plt.title('Sample\'s topic distribution')
#plt.xticks(np.arange(0,width/2.0,N*width), ['S'+str(n) for n in range(1,N)])
topiclegend = ['Topic' + str(Zrange[labels.index(l)]+1) + ': '+ l + '\n ('+ str(r[Zrange[labels.index(l)]]) + ')' for l in labels]
fontP = FontProperties()
if N >60:
fontP.set_size('xx-small')
else: fontP.set_size('small')
ax = plt.subplot(111)
ratio = float(N)*0.5
ax.set_aspect(ratio)
ax.tick_params(axis = 'x', colors='w') #remove tick labels by setting them the same color as background
box = ax.get_position()
ax.set_position([box.x0, box.y0, 0.5, box.height])
if order is not None:
plt.xticks(n, order, size = 'xx-small')
if Z > 12:
columns = 2
else: columns = 1
plt.legend(p, topiclegend, prop = fontP, title = 'Topic Label', loc='center left', bbox_to_anchor=(1, 0.5), ncol = columns)
return plt
def topiclabel_scatter(X,Y,factor,z, colorlabel):
'''Given a study, a topic and a label we make scatter plot'''
plt.figure(1, figsize=(8,8))
labelset = list(set(colorlabel))
#print labelset
colors = plt.cm.rainbow(np.linspace(0, 1,len(labelset)+1))
#colors = ['r','b','y','m']
for x,y,c in zipper(X,Y,colorlabel):
plt.scatter(x,y, color=colors[labelset.index(c)])
plt.ylabel(factor)
plt.xlabel('Topic proportion')
plt.title('Scatter plot of proportion of Topic ' + str(z+1) + ' and ' + factor)
plt.legend(labelset)
return plt
def loglikelihood_curve(study = None, name = None, run = 'all', useC = False, save = False):
'''Given a dataset we can find the models p_z,p_w_z,p_d_z for
different number of topics Z and can plot the loglikelihood vs. Z curve'''
logl = []
topic = []
files_found = False
if type(run) == int : run = "_run" + str(run)
elif run == 'all': run = "_*"
add = ''
if useC:
add = '_with_C'
if study is not None:
study = str(study)
RESULT_FILE_TEMPLATE = 'study_' + study + '*_topics' + add + run + '.txt' #what a result file looks like
else:
RESULT_FILE_TEMPLATE = 'study_' + name + '*_topics' + add + run + '.txt' #what a result file looks like
m = microbplsa.MicrobPLSA()
m.open_data(study = study, name = name)
for file in os.listdir(os.path.join(_root_dir, RESULTS_LOCATION)):
if fnmatch.fnmatch(file, RESULT_FILE_TEMPLATE):
print 'Calculating LogLikelihood of model from the file: ', file
files_found = True
Z = int(re.findall(r'\d+_topic', file)[0].split('_')[0])
print 'Getting model...'
m.open_model(modelFile = os.path.join(_root_dir, RESULTS_LOCATION, file)) #get model from the results file
print 'Calculating...'
L = m.loglikelihood()
if Z in topic:
logl[topic.index(Z)].append(L)
else:
topic.append(Z)
logl.append([L])
print "Z = {0}, L = {1}".format(Z,L)
logl_std = [np.std(x) for x in logl]
log_count = [len(x) for x in logl]
logl = [np.mean(x) for x in logl]
if not files_found:
print "There were no files found using the template:" + RESULT_FILE_TEMPLATE
sys.exit()
topic.sort()
logl, logl_std = zip(*sorted(zipper(logl,logl_std)))
plt.plot(topic,logl,'b-')
#plt.plot(topic,logl,'m.')
plt.errorbar(topic, logl, yerr=logl_std, fmt='m.')
plt.ylabel('LogLikelihood')
if study:
plt.title('Loglikelihood versus number of topics for study ' + study )
else:
plt.title('Loglikelihood versus number of topics for study ' + str(name) )
plt.xlabel('Z, number of topics')
if save:
topic = [str(round(t,2)) for t in topic]
count = [str(t) for t in log_count]
logl = [str(round(t,2)) for t in logl]
logl_std = [str(round(t,1)) for t in logl_std]
if study:
logfile = open(os.path.join(_root_dir, RESULTS_LOCATION, 'Loglikelihoods', 'logs_' + study + '.txt'), 'w')
elif name:
logfile = open(os.path.join(_root_dir, RESULTS_LOCATION, 'Loglikelihoods', 'logs_' + str(name) + '.txt'), 'w')
logfile.write('\t'.join(topic))
logfile.write('\n')
logfile.write('\t'.join(count))
logfile.write('\n')
logfile.write('\t'.join(logl))
logfile.write('\n')
logfile.write('\t'.join(logl_std))
return plt
