def plot_embedding(X, y, outtag, title=None, alpha=0.6):
xmin, xmax = np.min(X,0), np.max(X,0)
X = (X-xmin)/(xmax-xmin)
fig = plt.figure()
for i in range(len(y)):
plt.text(X[i,0],X[i,1],
ur'$\mathrm{%s}$' % sampa_to_unicode(vowels[int(y[i])]),
color=colors[int(y[i])],
alpha=alpha)
# plt.scatter(X[i,0],X[i,1], color=colors[int(y[i])], alpha=alpha)
plt.xticks([]), plt.yticks([])
if not title is None:
plt.title(title)
plt.xlim(np.min(X,0)[0]-0.01, np.max(X,0)[0]+0.01)
plt.ylim(np.min(X,0)[1]-0.01, np.max(X,0)[1]+0.01)
plt.savefig(os.path.join(cfg_figdir, outtag))
def plot_vowels(outfiletag,
#nsamples=1000,
vowels=None,
scale='log',
percentile=99,
speakers=None,
verbose=True,
minsamples=1):
if vowels is None:
vowels = list(vowels_sampa)
allowed_scales = ['log', 'linear', 'bark','mel']
if not scale in allowed_scales:
raise ValueError, 'scale must be one of [%s]' % ', '.join(allowed_scales)
cgn_corpus = cgn.CGN()
ifa_corpus = ifa.IFA()
corpus = MergedCorpus([ifa_corpus, cgn_corpus])
#corpus = MergedCorpus([ifa_corpus])
fm = formants.FormantsMeasure(corpus)
fm.info()
forms = fm.sample(vowels, equal_samples=True,scale='hertz')
vowels = filter(lambda x:forms[x].shape[0] >= minsamples, vowels)
# plot the static F1, F2
fig = plt.figure()
colors = ['b','g','r','c','m','y','k']
xs = []
ys = []
min_x=np.inf
max_x=np.NINF
min_y=np.inf
max_y=np.NINF
means = {}
for n in range(len(vowels)):
f2 = forms[vowels[n]][:,4]
f1 = forms[vowels[n]][:,3]
# filter out values outside specified percentile score
f2_bottom_perc = stats.scoreatpercentile(f2, 100-percentile)
f2_top_perc = stats.scoreatpercentile(f2, percentile)
f1_bottom_perc = stats.scoreatpercentile(f1, 100-percentile)
f1_top_perc = stats.scoreatpercentile(f1, percentile)
f2_mask = np.logical_and(f2 > f2_bottom_perc, f2 < f2_top_perc)
f1_mask = np.logical_and(f1 > f1_bottom_perc, f1 < f1_top_perc)
mask = np.logical_and(f2_mask, f1_mask)
f2 = f2[mask]
f1 = f1[mask]
xs.append(f2)
ys.append(f1)
min_f2 = np.min(f2)
max_f2 = np.max(f2)
min_f1 = np.min(f1)
max_f1 = np.max(f1)
if min_f2 < min_x:
min_x = min_f2
if max_f2 > max_x:
max_x = max_f2
if min_f1 < min_y:
min_y = min_f1
if max_f1 > max_y:
max_y = max_f1
f1_mean = np.mean(f1, axis=0)
f2_mean = np.mean(f2, axis=0)
means[vowels[n]] = (f1_mean, f2_mean)
if verbose:
print 'vowel: %s\tobserved: %d\t sample mean (f1,f2): (%.3f,%.3f)' % (vowels[n], f2.shape[0], f1_mean, f2_mean)
#nsamples = min(map(lambda x:x.shape[0], xs))
for n in range(len(vowels)):
#sample_ids = random.sample(range(xs[n].shape[0]), nsamples)
# xs_loc = xs[n][sample_ids,:]
# ys_loc = ys[n][sample_ids,:]
xs_loc = xs[n]
ys_loc = ys[n]
#print 'samplesize for %s: %d' % (vowels[n],xs_loc.shape[0])
plt.scatter(xs_loc, ys_loc,
color=colors[n % len(colors)],
label=ur'$\mathrm{%s}$' % sampa_to_unicode(vowels[n]),
alpha=0.2)
for n in range(len(vowels)):
plt.scatter(means[vowels[n]][1], means[vowels[n]][0],
s=80,
color='k',
marker=ur'$\mathrm{%s}$' % sampa_to_unicode(vowels[n]))
#print means
plt.xlim(max_x+100, min_x-100)
plt.ylim(max_y+100, min_y-100)
plt.xlabel(r'F2')
plt.ylabel(r'F1')
if scale == 'log':
plt.xscale('log')
plt.yscale('log')
plt.legend(loc='lower left')
plt.savefig(os.path.join(cfg_figdir, outfiletag+'.png'))