def getBigramMaxesAndMins(bigrams):
def ni(x):
return x[x != float("-Inf")]
maxes = Object()
maxes.fullLength = np.max([np.max(ni(b.fullLength)) for b in bigrams])
maxes.gap = np.max([np.max(ni(b.gap)) for b in bigrams])
maxes.startStart = np.max([np.max(ni(b.startStart)) for b in bigrams])
mins = Object()
mins.fullLength = np.min([np.min(ni(b.fullLength)) for b in bigrams])
mins.gap = np.min([np.min(ni(b.gap)) for b in bigrams])
mins.startStart = np.min([np.min(ni(b.startStart)) for b in bigrams])
return (maxes, mins)
def compileBigramModel(data):
COUNT_CUTOFF = 5
TIME_CUTOFF = 5
bigramModel = data.bigramModel
keycodes = sorted(list(KEYCODES.values()))
keyToIndex = dict((v, k) for (k, v) in enumerate(keycodes))
fullLength = np.zeros([len(keycodes)] * 2) + float("-Inf")
gap = np.zeros([len(keycodes)] * 2) + float("-Inf")
startStart = np.zeros([len(keycodes)] * 2) + float("-Inf")
for (k, v) in bigramModel.iteritems():
for (m, n) in v.iteritems():
(fl, g, ss) = zip(*n)
if len(fl) >= COUNT_CUTOFF and abs(np.mean(fl)) < TIME_CUTOFF:
fullLength[keyToIndex[k], keyToIndex[m]] = np.mean(fl)
if len(g) >= COUNT_CUTOFF and abs(np.mean(g)) < TIME_CUTOFF:
gap[keyToIndex[k], keyToIndex[m]] = np.mean(g)
if len(ss) >= COUNT_CUTOFF and abs(np.mean(ss)) < TIME_CUTOFF:
startStart[keyToIndex[k], keyToIndex[m]] = np.mean(ss)
bigram = Object()
bigram.fullLength = fullLength
bigram.gap = gap
bigram.startStart = startStart
return bigram
