def putFeatures(data):
""" Saves features to be later used for prediction and learning"""
if not isinstance(data['data'], list) and not len(data['data']) == 1:
raise Exception(data['data'])
problem = data['problem']
features = db.Features(data['user'], data['hub'], data['track'], problem['chrom'], problem['chromStart'])
features.put(pd.Series(data['data'][0]))
def doPrediction(data, problem, txn=None):
features = db.Features(data['user'], data['hub'], data['track'],
problem['chrom'], problem['chromStart']).get()
if features.empty:
return False
model = db.Prediction('model').get()
if not isinstance(model, dict):
return False
colsToDrop = db.Prediction('badCols').get()
featuresDropped = features.drop(labels=colsToDrop)
prediction = predictWithFeatures(featuresDropped, model)
if prediction is None:
return False
return prediction
def getDataPoints():
if not check():
return
dataPoints = pd.DataFrame()
for key in db.ModelSummaries.db_key_tuples():
modelSum = db.ModelSummaries(*key).get()
if modelSum.empty:
continue
if modelSum['regions'].max() < 1:
continue
withPeaks = modelSum[modelSum['numPeaks'] > 0]
noError = withPeaks[withPeaks['errors'] < 1]
logPenalties = np.log10(noError['penalty'].astype(float))
featuresDb = db.Features(*key)
featuresTxn = db.getTxn()
features = featuresDb.get(txn=featuresTxn, write=True)
for penalty in logPenalties:
datapoint = features.copy()
datapoint['logPenalty'] = penalty
dataPoints = dataPoints.append(datapoint, ignore_index=True)
featuresTxn.commit()
# TODO: Save datapoints, update ones which have changed, not all of them every time
Y = dataPoints['logPenalty']
X = dataPoints.drop('logPenalty', 1)
return dropBadCols(X), Y
def getFeatures(data):
problem = data['problem']
features = db.Features(data['user'], data['hub'], data['track'], problem['chrom'], problem['chromStart'])
return features.get()