def runAlgorithm():
K,P = 10, 75
for modelName in [ Rtm ]: #ModelNames:
cmdline = '' \
+ ' --debug ' + "False" \
+ ' --model ' + modelName \
+ ' --dtype ' + 'f8:f8' \
+ ' --num-topics ' + str(K) \
+ ' --log-freq ' + '5' \
+ ' --eval ' + 'perplexity' \
+ ' --iters ' + '5' \
+ ' --query-iters ' + '5' \
+ ' --folds ' + '2' \
+ ' --words ' + AclWordPath \
+ ' --links ' + AclCitePath \
+ ' --limit-to ' + '100000' \
+ ' --eval ' + MeanAveragePrecAllDocs \
+ ' --out-model ' + '/Users/bryanfeeney/Desktop/acl-out'
# + ' --words ' + '/Users/bryanfeeney/Dropbox/Datasets/ACL/words.pkl' \
# + ' --words ' + '/Users/bryanfeeney/Desktop/NIPS-from-pryor-Sep15/W_ar.pkl'
# + ' --words ' + '/Users/bryanfeeney/Desktop/Dataset-Sep-2014/words.pkl' \
# + ' --feats ' + '/Users/bryanfeeney/Desktop/Dataset-Sep-2014/side.pkl'
# + ' --words ' + wordsFile \
# + ' --feats ' + featsFile
# + ' --words ' + '/Users/bryanfeeney/Desktop/Tweets600/words-by-author.pkl' \
# + ' --out-model ' + modelFileDir \
run(cmdline.strip().split(' '))
def testMainScript(self):
D, T, K, Q, F, P, avgWordsPerDoc = 200, 100, 10, 6, 12, 8, 500
tpcs, vocab, docLens, X, W = sampleFromModel(D, T, K, F, P, avgWordsPerDoc)
wordsFile, featsFile, modelFileDir = tmpFiles()
with open(wordsFile, 'wb') as f:
pkl.dump(W, f)
with open(featsFile, 'wb') as f:
pkl.dump(X, f)
print ("New Version")
RateDelay = 10
ForgettingRate = 0.55
BatchSize = 100
sgd_setups = [(b,r,f) for b in [1, 5, 10, 100] for r in [1, 10, 30] for f in [0.6, 0.75, 0.9]]
Folds, ExecutedFoldCount = 5,1
K,P,Q = 10, 150, 20
TrainIters, QueryIters, LogFreq = 2, 2, 1
PriorCov = 1
VocabPrior = 5
Debug = True
modelFileses = []
for DataSetName in [TweetsFreq]:
for k in [K]: # [10, 25, 50, 100]:
for p in [P]: #, [50, 100, 250, 500]:
#for (BatchSize, RetardationRate, ForgettingRate) in sgd_setups:
for modelName in [ StmYvBohning ]:
cmdline = '' \
+(' --debug ' + str(Debug) if Debug else "") \
+ ' --model ' + modelName \
+ ' --dtype ' + 'f8' \
+ ' --num-topics ' + str(k) \
+ ' --num-lat-feats ' + str(p) \
+ ' --num-lat-topics ' + str(Q) \
+ ' --log-freq ' + str(LogFreq) \
+ ' --eval ' + HashtagPrecAtM \
+ ' --gradient-batch-size ' + str(BatchSize) \
+ ' --gradient-rate-delay ' + str(RateDelay) \
+ ' --gradient-forgetting-rate ' + str(ForgettingRate) \
+ ' --iters ' + str(TrainIters) \
+ ' --query-iters ' + str(QueryIters) \
+ ' --folds ' + str(Folds) \
+ ' --truncate-folds ' + str(ExecutedFoldCount) \
+ (' --word-dict ' + DictsPath[DataSetName] if DictsPath[DataSetName] is not None else "") \
+ ' --words ' + WordsPath[DataSetName] \
+ (' --feats ' + FeatsPath[DataSetName] if FeatsPath[DataSetName] is not None else "") \
+ (' --links ' + CitesPath[DataSetName] if CitesPath[DataSetName] is not None else "") \
+ ' --topic-var ' + str(PriorCov) \
+ ' --feat-var ' + str(PriorCov) \
+ ' --lat-topic-var ' + str(PriorCov) \
+ ' --lat-feat-var ' + str(PriorCov) \
+ ' --vocab-prior ' + str(VocabPrior) \
+ ' --tag-recall-opts ' + "0:-1,0.75"
# + ' --out-model ' + '/Users/bryanfeeney/Desktop/acl-out-tm' \
# + ' --feats-mask ' + FeatsMask[DataSetName] \
# + ' --lda-model ' + PreBuiltVbTopics[DataSetName][k]
# + ' --words ' + '/Users/bryanfeeney/Dropbox/Datasets/ACL/words.pkl' \
# + ' --words ' + '/Users/bryanfeeney/Desktop/NIPS-from-pryor-Sep15/W_ar.pkl'
# + ' --words ' + '/Users/bryanfeeney/Desktop/Dataset-Sep-2014/words.pkl' \
# + ' --feats ' + '/Users/bryanfeeney/Desktop/Dataset-Sep-2014/side.pkl'
# + ' --words ' + wordsFile \
# + ' --feats ' + featsFile
# + ' --words ' + '/Users/bryanfeeney/Desktop/Tweets600/words-by-author.pkl' \
modelFileses.extend (run(cmdline.strip().split(' ')))
modelFileses.insert(0, wordsFile)
modelFileses.insert(1, featsFile)
print ("Files can be found in:" + "\n\t".join(modelFileses))
def testAlgorithms(self):
D, T, K, Q, F, P, avgWordsPerDoc = 200, 100, 10, 6, 12, 8, 500
tpcs, vocab, docLens, X, W = sampleFromModel(D, T, K, F, P, avgWordsPerDoc)
wordsFile, featsFile, modelFileDir = tmpFiles()
with open(wordsFile, "wb") as f:
pkl.dump(W, f)
with open(featsFile, "wb") as f:
pkl.dump(X, f)
print("New Version")
Folds, ExecutedFoldCount = 5, 5
K, P, Q = 50, 150, 20
TrainIters, QueryIters, LogFreq = 600, 500, 50
PriorCov = 1
VocabPrior = 5
Debug = False
modelFileses = []
for DataSetName in [TweetsFreq]:
for k in [50]:
for modelName in [StmUyvBohning]:
cmdline = (
""
+ (" --debug " + str(Debug) if Debug else "")
+ " --model "
+ modelName
+ " --dtype "
+ "f8:f8"
+ " --num-topics "
+ str(k)
+ " --num-lat-feats "
+ str(P)
+ " --num-lat-topics "
+ str(Q)
+ " --log-freq "
+ str(LogFreq)
+ " --eval "
+ Perplexity
+ " --iters "
+ str(TrainIters)
+ " --query-iters "
+ str(QueryIters)
+ " --folds "
+ str(Folds)
+ " --truncate-folds "
+ str(ExecutedFoldCount)
+ (" --word-dict " + DictsPath[DataSetName] if DictsPath[DataSetName] is not None else "")
+ " --words "
+ WordsPath[DataSetName]
+ (" --feats " + FeatsPath[DataSetName] if FeatsPath[DataSetName] is not None else "")
+ (" --links " + CitesPath[DataSetName] if CitesPath[DataSetName] is not None else "")
+ " --topic-var "
+ str(PriorCov)
+ " --feat-var "
+ str(PriorCov)
+ " --lat-topic-var "
+ str(PriorCov)
+ " --lat-feat-var "
+ str(PriorCov)
+ " --vocab-prior "
+ str(VocabPrior)
+ " --out-model "
+ "/Users/bryanfeeney/Desktop/acl-out-tm"
)
# + ' --feats-mask ' + FeatsMask[DataSetName] \
# + ' --lda-model ' + PreBuiltVbTopics[DataSetName][k]
# + ' --words ' + '/Users/bryanfeeney/Dropbox/Datasets/ACL/words.pkl' \
# + ' --words ' + '/Users/bryanfeeney/Desktop/NIPS-from-pryor-Sep15/W_ar.pkl'
# + ' --words ' + '/Users/bryanfeeney/Desktop/Dataset-Sep-2014/words.pkl' \
# + ' --feats ' + '/Users/bryanfeeney/Desktop/Dataset-Sep-2014/side.pkl'
# + ' --words ' + wordsFile \
# + ' --feats ' + featsFile
# + ' --words ' + '/Users/bryanfeeney/Desktop/Tweets600/words-by-author.pkl' \
modelFileses.extend(run(cmdline.strip().split(" ")))
modelFileses.insert(0, wordsFile)
modelFileses.insert(1, featsFile)
print("Files can be found in:" + "\n\t".join(modelFileses))
