def _learnFF(seq, X, Xblur, Lmin, Lmax, Lfilt, generateSeedsStep=.1):
padLen = (len(seq) - X.shape[1]) // 2
X = ar.addZeroCols(X, padLen, prepend=True)
X = ar.addZeroCols(X, padLen, prepend=False)
Xblur = ar.addZeroCols(Xblur, padLen, prepend=True)
Xblur = ar.addZeroCols(Xblur, padLen, prepend=False)
timeStartSeed = time.clock()
# find seeds; i.e., candidate instance indices from which to generalize
numShifts = int(1. / generateSeedsStep) + 1
stepLen = int(Lmax * generateSeedsStep)
windowLen = Lmax + stepLen
# score all subseqs based on how much they don't look like random walks
# when examined using different sliding window lengths
scores = np.zeros(len(seq))
for dim in range(seq.shape[1]):
# compute these just once, not once per length
dimData = seq[:, dim].flatten()
std = np.std(dimData[1:] - dimData[:-1])
for divideBy in [1, 2, 4, 8]:
partialScores = feat.windowScoresRandWalk(dimData, Lmin // divideBy, std)
scores[:len(partialScores)] += partialScores
# figure out optimal pair based on scores of all subseqs
bestIdx = np.argmax(scores)
start = max(0, bestIdx - Lmin)
end = min(len(scores), start + Lmin)
scores[start:end] = -1 # disqualify idxs within Lmin of bestIdx
secondBestIdx = np.argmax(scores)
# compute all seed idxs from this pair
seedIdxs = [bestIdx, secondBestIdx]
maxIdx = X.shape[1] - windowLen - 1
seedStartIdxs = _computeAllSeedIdxsFromPair(seedIdxs, numShifts, stepLen, maxIdx)
seedEndIdxs = seedStartIdxs + windowLen
timeEndSeed = time.clock()
bsfScore, bsfLocs, bsfFilt = _findInstancesUsingSeedLocs(X, Xblur,
seedStartIdxs, seedEndIdxs, Lmin, Lmax, Lfilt, windowLen=windowLen)
startIdxs, endIdxs = _extractTrueLocs(X, Xblur, bsfLocs, bsfFilt,
windowLen, Lmin, Lmax)
timeEndFF = time.clock()
print "learnFF(): seconds to find seeds, regions, total =\n\t{:.3f}\t{:.3f}\t{:.3f}".format(
timeEndSeed - timeStartSeed, timeEndFF - timeEndSeed, timeEndFF - timeStartSeed)
return startIdxs, endIdxs, bsfFilt
def _seedScores(seq, Lmin):
scores = np.zeros(len(seq))
for dim in range(seq.shape[1]):
# compute these just once, not once per length
dimData = seq[:, dim].flatten()
std = np.std(dimData[1:] - dimData[:-1])
for divideBy in [1, 2, 4, 8]:
length = Lmin // divideBy
# for length in [8, 16, 32]:
partialScores = feat.windowScoresRandWalk(dimData, length, std)
# Lmin // divideBy, std, 1000) # TODO remove 1k
scores[: len(partialScores)] += partialScores
# # center partial scores within full seq
# end = min(len(scores), len(partialScores) + length // 2)
# start = end - len(partialScores)
# scores[start:end] += partialScores
return scores
def _learnFF(seq, X, Xblur, Lmin, Lmax, Lfilt, generateSeedsStep=.1):
padLen = (len(seq) - X.shape[1]) // 2
X = ar.addZeroCols(X, padLen, prepend=True)
X = ar.addZeroCols(X, padLen, prepend=False)
Xblur = ar.addZeroCols(Xblur, padLen, prepend=True)
Xblur = ar.addZeroCols(Xblur, padLen, prepend=False)
timeStartSeed = time.clock()
# find seeds; i.e., candidate instance indices from which to generalize
numShifts = int(1. / generateSeedsStep) + 1
stepLen = int(Lmax * generateSeedsStep)
windowLen = Lmax + stepLen
# score all subseqs based on how much they don't look like random walks
# when examined using different sliding window lengths
scores = np.zeros(len(seq))
for dim in range(seq.shape[1]):
# compute these just once, not once per length
dimData = seq[:, dim].flatten()
std = np.std(dimData[1:] - dimData[:-1])
for divideBy in [1, 2, 4, 8]:
partialScores = feat.windowScoresRandWalk(dimData,
Lmin // divideBy, std)
scores[:len(partialScores)] += partialScores
# figure out optimal pair based on scores of all subseqs
bestIdx = np.argmax(scores)
start = max(0, bestIdx - Lmin)
end = min(len(scores), start + Lmin)
scores[start:end] = -1 # disqualify idxs within Lmin of bestIdx
secondBestIdx = np.argmax(scores)
# compute all seed idxs from this pair
seedIdxs = [bestIdx, secondBestIdx]
maxIdx = X.shape[1] - windowLen - 1
seedStartIdxs = _computeAllSeedIdxsFromPair(seedIdxs, numShifts, stepLen,
maxIdx)
seedEndIdxs = seedStartIdxs + windowLen
timeEndSeed = time.clock()
bsfScore, bsfLocs, bsfFilt = _findInstancesUsingSeedLocs(
X,
Xblur,
seedStartIdxs,
seedEndIdxs,
Lmin,
Lmax,
Lfilt,
windowLen=windowLen)
startIdxs, endIdxs = _extractTrueLocs(X, Xblur, bsfLocs, bsfFilt,
windowLen, Lmin, Lmax)
timeEndFF = time.clock()
print "learnFF(): seconds to find seeds, regions, total =\n\t{:.3f}\t{:.3f}\t{:.3f}".format(
timeEndSeed - timeStartSeed, timeEndFF - timeEndSeed,
timeEndFF - timeStartSeed)
return startIdxs, endIdxs, bsfFilt
