def trainCG(distInit, accumulate, ps = d.getDefaultParamSpec(), length = -50, verbosity = 0):
"""Re-estimates a distribution using a conjugate gradient optimizer.
See the note in the docstring for this module for information on how the
log likelihood is scaled. This scaling is presumed to have only a small
impact on the dist returned by this function.
"""
# (FIXME : investigate how large the effect of the scale factor is for
# a few example dists?)
def negLogLike_derivParams(params):
dist = ps.parseAll(distInit, params)
acc = ps.createAccG(dist)
accumulate(acc)
count = acc.count()
count = max(count, 1.0)
return -acc.logLike() / count, -ps.derivParams(acc) / count
params = ps.params(distInit)
if verbosity >= 2:
print 'trainCG: initial params =', params
print 'trainCG: initial derivParams =', -negLogLike_derivParams(params)[1]
params, negLogLikes, lengthUsed = minimize(negLogLike_derivParams, params, length = length, verbosity = verbosity)
if verbosity >= 3:
print 'trainCG: logLikes =', map(lambda x: -x, negLogLikes)
if verbosity >= 2:
print 'trainCG: final params =', params
print 'trainCG: final derivParams =', -negLogLike_derivParams(params)[1]
if verbosity >= 1:
print 'trainCG: logLike %s -> %s (delta = %s)' % (-negLogLikes[0], -negLogLikes[-1], negLogLikes[0] - negLogLikes[-1])
print 'trainCG: (used', lengthUsed, 'function evaluations)'
dist = ps.parseAll(distInit, params)
return dist
def trainCGandEM(distInit, accumulate, ps = d.getDefaultParamSpec(), createAccEM = d.getDefaultCreateAcc(), estimateTotAux = d.getDefaultEstimateTotAux(), iterations = 5, length = -50, afterEst = None, verbosity = 0):
"""Re-estimates a distribution using conjugate gradients and EM.
See the note in the docstring for this module for information on how the
log likelihood is scaled. This scaling is presumed to have only a small
impact on the dist returned by this function (via its impact on trainCG).
"""
assert iterations >= 1
dist = distInit
for it in range(1, iterations + 1):
if verbosity >= 1:
print 'trainCGandEM: starting it =', it, 'of CG and EM'
dist = (timed(trainCG) if verbosity >= 2 else trainCG)(dist, accumulate, ps = ps, length = length, verbosity = verbosity)
dist, _, _, _ = expectationMaximization(dist, accumulate, createAcc = createAccEM, estimateTotAux = estimateTotAux, verbosity = verbosity)
if afterEst is not None:
afterEst(dist = dist, it = it)
if verbosity >= 1:
print 'trainCGandEM: finished it =', it, 'of CG and EM'
print 'trainCGandEM:'
return dist
def trainCG(distInit,
accumulate,
ps=d.getDefaultParamSpec(),
length=-50,
verbosity=0):
"""Re-estimates a distribution using a conjugate gradient optimizer.
See the note in the docstring for this module for information on how the
log likelihood is scaled. This scaling is presumed to have only a small
impact on the dist returned by this function.
"""
# (FIXME : investigate how large the effect of the scale factor is for
# a few example dists?)
def negLogLike_derivParams(params):
dist = ps.parseAll(distInit, params)
acc = ps.createAccG(dist)
accumulate(acc)
count = acc.count()
count = max(count, 1.0)
return -acc.logLike() / count, -ps.derivParams(acc) / count
params = ps.params(distInit)
if verbosity >= 2:
print 'trainCG: initial params =', params
print 'trainCG: initial derivParams =', -negLogLike_derivParams(
params)[1]
params, negLogLikes, lengthUsed = minimize(negLogLike_derivParams,
params,
length=length,
verbosity=verbosity)
if verbosity >= 3:
print 'trainCG: logLikes =', map(lambda x: -x, negLogLikes)
if verbosity >= 2:
print 'trainCG: final params =', params
print 'trainCG: final derivParams =', -negLogLike_derivParams(
params)[1]
if verbosity >= 1:
print 'trainCG: logLike %s -> %s (delta = %s)' % (
-negLogLikes[0], -negLogLikes[-1],
negLogLikes[0] - negLogLikes[-1])
print 'trainCG: (used', lengthUsed, 'function evaluations)'
dist = ps.parseAll(distInit, params)
return dist
def trainCGandEM(distInit,
accumulate,
ps=d.getDefaultParamSpec(),
createAccEM=d.getDefaultCreateAcc(),
estimateTotAux=d.getDefaultEstimateTotAux(),
iterations=5,
length=-50,
afterEst=None,
verbosity=0):
"""Re-estimates a distribution using conjugate gradients and EM.
See the note in the docstring for this module for information on how the
log likelihood is scaled. This scaling is presumed to have only a small
impact on the dist returned by this function (via its impact on trainCG).
"""
assert iterations >= 1
dist = distInit
for it in range(1, iterations + 1):
if verbosity >= 1:
print 'trainCGandEM: starting it =', it, 'of CG and EM'
dist = (timed(trainCG) if verbosity >= 2 else trainCG)(
dist, accumulate, ps=ps, length=length, verbosity=verbosity)
dist, _, _, _ = expectationMaximization(dist,
accumulate,
createAcc=createAccEM,
estimateTotAux=estimateTotAux,
verbosity=verbosity)
if afterEst is not None:
afterEst(dist=dist, it=it)
if verbosity >= 1:
print 'trainCGandEM: finished it =', it, 'of CG and EM'
print 'trainCGandEM:'
return dist
def __init__(self, mdlFactor, paramSpec=d.getDefaultParamSpec()):
self.mdlFactor = mdlFactor
self.paramSpec = paramSpec
def __init__(self, mdlFactor, paramSpec = d.getDefaultParamSpec()):
self.mdlFactor = mdlFactor
self.paramSpec = paramSpec
