def getTemplates(self):
""" Chip out the templates """
for maps in self.info:
# Determine if it's an H1 or an H0 template
if maps['class'] == 'H1':
P, freqs, bins = self.train.H1Sample(maps['file'],params=self.params)
else:
P, freqs, bins = self.train.H0Sample(maps['file'],params=self.params)
# Get the image and demean it
m, n = P.shape
P = metrics.slidingWindowV(P,maxM=m)
x0, xn, y0, yn = maps['x0'], maps['xn'], maps['y0'], maps['yn']
tmpl = P[x0:xn,y0:yn].astype('Float32')
# Normalize the image and turn into a binary mask
m_, s_ = np.mean(tmpl), np.std(tmpl)
min_ = tmpl.min()
tmpl[ tmpl < m_ + 0.5*s_] = min_
tmpl[ tmpl > min_ ] = 1
tmpl[tmpl < 0] = 0
# Add meta-data to the list
self.templates.append(tmpl)
m, n = tmpl.shape
self.limits.append([bins[y0], bins[yn], freqs[x0], freqs[xn]])
def getTemplates(self):
""" Chip out the templates """
for maps in self.info:
# Determine if it's an H1 or an H0 template
if maps['class'] == 'H1':
P, freqs, bins = self.train.H1Sample(maps['file'],
params=self.params)
else:
P, freqs, bins = self.train.H0Sample(maps['file'],
params=self.params)
# Get the image and demean it
m, n = P.shape
P = metrics.slidingWindowV(P, maxM=m)
x0, xn, y0, yn = maps['x0'], maps['xn'], maps['y0'], maps['yn']
tmpl = P[x0:xn, y0:yn].astype('Float32')
# Normalize the image and turn into a binary mask
m_, s_ = np.mean(tmpl), np.std(tmpl)
min_ = tmpl.min()
tmpl[tmpl < m_ + 0.5 * s_] = min_
tmpl[tmpl > min_] = 1
tmpl[tmpl < 0] = 0
# Add meta-data to the list
self.templates.append(tmpl)
m, n = tmpl.shape
self.limits.append([bins[y0], bins[yn], freqs[x0], freqs[xn]])
import metrics
import fileio
import numpy
from matplotlib import mlab, pyplot as plt
import pandas
df = pandas.DataFrame(
numpy.zeros(30000 * 5901).reshape(30000, 5901), columns=(["label"] + map(lambda a: str(a), range(100 * 59)))
)
labels = pandas.read_csv("../data/train.csv")
for i in range(30000):
img = fileio.ReadAIFF("../data/train/train{}.aiff".format(i + 1))
P, freqs, bins = mlab.specgram(img, NFFT=256, Fs=2000, noverlap=192)
Q = metrics.slidingWindowV(P, inner=3, maxM=40)
W = metrics.slidingWindowH(P, inner=3, outer=32, maxM=60)
s = numpy.vstack((Q, W))
df.ix[i] = numpy.append(labels["label"][i], s.reshape(100 * 59))
if i % 100 == 0:
print i
df.to_csv("total.csv", index=False)
def slidingWindowV(img):
w = metrics.slidingWindowV(img, inner=3, maxM=40)
print w.shape
return w
