def predict(self, descs=None, to_file=False):
'''
Call package functions to predict
'''
if descs == None:
self.descs = self.__readDescs(self.directory + '/descriptors/output.txt')
else:
self.descs = descs
self.predictor = Predictor(network_name=self.networks,
predict_descs=self.descs,
training_descs=self.trn,
featureNorm=True)
# feature Normalization
normedDescs = self.predictor.featureNorm(self.predictor.descsForPred, self.predictor.X_scalar)
prediction_results = self.predictor.predict(self.predictor.network, normedDescs)
# Transform back
self.prediction_results = np.exp(prediction_results)
# if to_file:
# # save to file
# np.savetxt('prediction_results.csv',self.prediction_results,delimiter=',')
# print "Prediction results saved [prediction_results]. Make a copy if you want to keep it."
results_array = self.prediction_results.tolist()
confidence = self.predict_conf()
results = []
for idx, smiles in enumerate(self.smiles):
results.append([smiles, results_array[idx], confidence[idx]])
return results
class NetPrediction:
def __init__(self):
class_directory = os.path.dirname(os.path.abspath(inspect.getfile(inspect.currentframe())))
self.directory = class_directory
self.networks = self.directory + '/nets/CED_new.xml'
self.trn = self.__readTrn(self.directory + '/descriptors/trn_data_30.csv')
# try:
# opts, args = getopt.getopt(argv,"hn:d:t:",["net=","descs=","trn="])
# except getopt.GetoptError:
# print 'predict.py -n [Networks] -d [Descs] -t [TrnData]'
# sys.exit(2)
#
# # if len(args) == 0:
# # print 'Please provide networks, descriptors and training data, use -h for help'
# # sys.exit(2)
#
# for opt, arg in opts:
# if opt == '-h':
# print 'predict.py -n <network> -s <SMILEs>'
# sys.exit()
# elif opt in ("-n", "--net"):
# self.networks = arg
# elif opt in ("-d", "--descs"):
# self.descs = self._readDescs(arg)
# elif opt in ("-t", "--trn"):
# self.trn = self._readTrn(arg)
# else:
# print "Wrong arguments, use -h for helps."
#
# try:
# self.predictor = ann.predicting(network_name=self.networks,
# predict_descs=self.descs,
# training_descs=self.trn,
# featureNorm=True)
# except AttributeError:
# print 'Worng input files, please provide networks, descriptors and training data, use -h for help'
# sys.exit(2)
def __readDescs(self,raw_file):
'''
clean up descs file use pandas
remove header and col name
Input: raw_file: raw descriptors file with header and two useless colnums
return pure numpy arrary of descriptors values
'''
df = pd.read_csv(raw_file,delim_whitespace=True)
df = df.fillna(df.mean())
df = df.drop('No.', 1)
self.smiles = df['NAME']
df = df.drop('NAME',1)
val = df.values
return val
def __readTrn(self,trn_data):
'''
read training data
'''
# trnData = np.loadtxt(trn_data,skiprows=1)
df = pd.read_csv(trn_data)
trnData = df.values
return trnData
def predict(self, descs=None, to_file=False):
'''
Call package functions to predict
'''
if descs == None:
self.descs = self.__readDescs(self.directory + '/descriptors/output.txt')
else:
self.descs = descs
self.predictor = Predictor(network_name=self.networks,
predict_descs=self.descs,
training_descs=self.trn,
featureNorm=True)
# feature Normalization
normedDescs = self.predictor.featureNorm(self.predictor.descsForPred, self.predictor.X_scalar)
prediction_results = self.predictor.predict(self.predictor.network, normedDescs)
# Transform back
self.prediction_results = np.exp(prediction_results)
# if to_file:
# # save to file
# np.savetxt('prediction_results.csv',self.prediction_results,delimiter=',')
# print "Prediction results saved [prediction_results]. Make a copy if you want to keep it."
results_array = self.prediction_results.tolist()
confidence = self.predict_conf()
results = []
for idx, smiles in enumerate(self.smiles):
results.append([smiles, results_array[idx], confidence[idx]])
return results
def predict_conf(self):
'''
Prediction Confidence Level,
Base on the distance to the centroid of training data
'''
assert self.prediction_results is not None
assert self.trn is not None
self.dist = self.predictor.calcDist(self.descs, self.trn)
conf = []
for eachDist in self.dist:
if eachDist > 100:
conf.append('Low')
else:
conf.append('High')
self.prediction_conf = conf
# print self.smiles,self.dist
return conf
def to_file(self):
'''
combine outputs to file
'''
assert self.prediction_results is not None
outputs = pd.DataFrame(self.smiles)
outputs['prediction results'] = self.prediction_results
outputs['prediction confidence'] = self.prediction_conf
outputs.to_csv('prediction_results.csv')
print 'Prediction results output to "prediction_results.csv". Make a copy if you want to save the results'
