def MakeData(self, N=10000):
S1 = np.asarray([[100, 0], [0, 0.01]])
Sigma = np.zeros((2, 2, 4))
Sigma[:, :, 0] = S1
Sigma[:, :, 1] = RandUtil.rotateCovMat(S1, theta=np.pi / 4)
Sigma[:, :, 2] = RandUtil.rotateCovMat(S1, theta=2 * np.pi / 4)
Sigma[:, :, 3] = RandUtil.rotateCovMat(S1, theta=3 * np.pi / 4)
self.Sigma = Sigma
Xlist = list()
Rlist = list()
for k in range(Sigma.shape[2]):
curX = RandUtil.mvnrand([0, 0], Sigma[:, :, k], N)
curresp = np.zeros((N, 4))
curresp[:, k] = 1.0
Xlist.append(curX)
Rlist.append(curresp)
X = np.vstack(Xlist)
self.Data = XData(X=X)
self.trueresp = np.vstack(Rlist)
def MakeData(self, N=10000):
S1 = np.asarray([[100, 0], [0, 0.01]])
Sigma = np.zeros( (2,2,4))
Sigma[:,:,0] = S1
Sigma[:,:,1] = RandUtil.rotateCovMat(S1, theta=np.pi/4)
Sigma[:,:,2] = RandUtil.rotateCovMat(S1, theta=2*np.pi/4)
Sigma[:,:,3] = RandUtil.rotateCovMat(S1, theta=3*np.pi/4)
self.Sigma = Sigma
Xlist = list()
Rlist = list()
for k in range(Sigma.shape[2]):
curX = RandUtil.mvnrand([0,0], Sigma[:,:,k], N)
curresp = np.zeros((N,4))
curresp[:,k] = 1.0
Xlist.append(curX)
Rlist.append(curresp)
X = np.vstack(Xlist)
self.Data = XData(X=X)
self.trueresp = np.vstack(Rlist)
def CreateToyDataFromMixModel(cls, seed=101,
nDocTotal=None,
nWordsPerDoc=None,
nWordsPerDocFunc=None,
beta=None,
topics=None,
**kwargs):
''' Generates BagOfWordsData dataset via mixture generative model.
Returns
------
Data : BagOfWordsData object
'''
from bnpy.util import RandUtil
PRNG = np.random.RandomState(seed)
K = topics.shape[0]
V = topics.shape[1]
# Make sure topics sum to one
topics = topics / topics.sum(axis=1)[:, np.newaxis]
assert K == beta.size
doc_range = np.zeros(nDocTotal + 1)
wordIDsPerDoc = list()
wordCountsPerDoc = list()
resp = np.zeros((nDocTotal, K))
Ks = list(range(K))
# startPos : tracks start index for current doc within corpus-wide
# lists
startPos = 0
for d in range(nDocTotal):
# Draw single topic assignment for this doc
k = RandUtil.choice(Ks, beta, PRNG)
resp[d, k] = 1
# Draw the observed words for this doc
# wordCountBins: V x 1 vector, entry v counts appearance of word v
wordCountBins = RandUtil.multinomial(nWordsPerDoc,
topics[k, :], PRNG)
# Record word_id, word_count, doc_range
wIDs = np.flatnonzero(wordCountBins > 0)
wCounts = wordCountBins[wIDs]
assert np.allclose(wCounts.sum(), nWordsPerDoc)
wordIDsPerDoc.append(wIDs)
wordCountsPerDoc.append(wCounts)
doc_range[d] = startPos
startPos += wIDs.size
# Package up all data
word_id = np.hstack(wordIDsPerDoc)
word_count = np.hstack(wordCountsPerDoc)
doc_range[-1] = word_count.size
# Make TrueParams dict
TrueParams = dict(K=K, topics=topics, beta=beta, resp=resp)
Data = BagOfWordsData(
word_id, word_count, doc_range, V, TrueParams=TrueParams)
return Data
def CreateToyDataFromLDAModel(cls, seed=101,
nDocTotal=None,
nWordsPerDoc=None,
nWordsPerDocFunc=None,
topic_prior=None, topics=None,
alpha=None, proba_K=None,
**kwargs):
''' Generates BagOfWordsData dataset via LDA generative model.
Returns
------
Data : BagOfWordsData object
'''
if topic_prior is None:
topic_prior = alpha * proba_K
from bnpy.util import RandUtil
K = topics.shape[0]
V = topics.shape[1]
# Make sure topics sum to one
topics = topics / topics.sum(axis=1)[:, np.newaxis]
assert K == topic_prior.size
doc_range = np.zeros(nDocTotal + 1)
wordIDsPerDoc = list()
wordCountsPerDoc = list()
Pi = np.zeros((nDocTotal, K))
respPerDoc = list()
# startPos : tracks start index for current doc within corpus-wide
# lists
startPos = 0
for d in range(nDocTotal):
# Need docseed to be type int, have non-zero value for all d
docseed = (seed * d + seed) % (100000000)
PRNG = np.random.RandomState(docseed)
# Draw topic appearance probabilities for this document
Pi[d, :] = PRNG.dirichlet(topic_prior)
if nWordsPerDocFunc is not None:
nWordsPerDoc = nWordsPerDocFunc(PRNG)
# Draw the topic assignments for this doc
# Npercomp : K-vector, Npercomp[k] counts appearance of topic k
Npercomp = RandUtil.multinomial(nWordsPerDoc, Pi[d, :], PRNG)
# Draw the observed words for this doc
# wordCountBins: V x 1 vector, entry v counts appearance of word v
wordCountBins = np.zeros(V)
for k in range(K):
wordCountBins += RandUtil.multinomial(Npercomp[k],
topics[k, :], PRNG)
# Record word_id, word_count, doc_range
wIDs = np.flatnonzero(wordCountBins > 0)
wCounts = wordCountBins[wIDs]
assert np.allclose(wCounts.sum(), nWordsPerDoc)
wordIDsPerDoc.append(wIDs)
wordCountsPerDoc.append(wCounts)
doc_range[d] = startPos
startPos += wIDs.size
# Record expected local parameters (LP)
curResp = (topics[:, wIDs] * Pi[d, :][:, np.newaxis]).T
respPerDoc.append(curResp)
word_id = np.hstack(wordIDsPerDoc)
word_count = np.hstack(wordCountsPerDoc)
doc_range[-1] = word_count.size
# Make TrueParams dict
resp = np.vstack(respPerDoc)
resp /= resp.sum(axis=1)[:, np.newaxis]
TrueParams = dict(K=K, topics=topics,
beta=topic_prior / topic_prior.sum(),
topic_prior=topic_prior, resp=resp)
Data = BagOfWordsData(
word_id, word_count, doc_range, V, TrueParams=TrueParams)
return Data