def computeScoresOverride(self, query, docs):
queryText = self.concatTextEntryWords(query)
if self.debugPrint:
print('getScores', queryText)
queryIdArr = []
queryTextArr = []
docTextArr = []
docIdArr = []
labelArr = []
for e in docs:
docIdArr.append(e.id)
docTextArr.append(self.concatTextEntryWords(e))
queryTextArr.append(queryText)
queryIdArr.append('fake_qid')
labelArr.append(0)
dataRaw = pd.DataFrame({
'id_left': queryIdArr,
'text_left': queryTextArr,
'id_right': docIdArr,
'text_right': docTextArr,
'label': labelArr
})
dataTestPacked = pack(dataRaw)
dataTestProc = self.prep.transform(dataTestPacked)
dataForModel, _ = dataTestProc.unpack()
preds = self.model.predict(dataForModel)
sampleRet = {}
for k in range(len(docs)):
e = docs[k]
score = preds[k]
if self.debugPrint:
print(score, self.textEntryToStr(e))
sampleRet[e.id] = score
return sampleRet
dataFileTrain = os.path.join('collections', 'derived_data', colName,
'match_zoo_train', 'tran_neg10.tsv')
dataFileTest = os.path.join('collections', 'derived_data', colName,
'match_zoo_train', 'dev1_allCand.tsv')
print(
f'Collection: {colName} # of epochs: {epochQty} model file: {modelFile} data transform file: {dataTranFile}'
)
if os.path.exists(modelFile):
# Stupid hack for now, b/c save will fail if the model exists
print('Model already exists, exiting!')
sys.exit(1)
# Note dtype! don't let Pandas guess column data types!
dataTrainPacked = pack(readWhiteSpacedMatchZooData(dataFileTrain))
dataTestPacked = pack(readWhiteSpacedMatchZooData(dataFileTest))
#prep = mz.preprocessors.BasicPreprocessor()
prep = WhiteSpacePreprocessor()
import pdb, sys
#try:
if True:
rankingTask = mz.tasks.Ranking(loss=mz.losses.RankHingeLoss())
rankingTask.metrics = [
mz.metrics.NormalizedDiscountedCumulativeGain(k=3),
mz.metrics.NormalizedDiscountedCumulativeGain(k=5),
mz.metrics.MeanAveragePrecision()