def test_theta_0():
rng.seed(0)
n_samples = 100
Y = rng.randn(n_samples, 5)
X = rng.randn(n_samples, 5)
sgcrf = SparseGaussianCRF(lamL=0.01, lamT=0.01)
sgcrf.fit(X, Y)
assert np.allclose(sgcrf.Lam, np.eye(5), .1, .2)
def sgcrf(task):
import sys
sys.path.insert(0, '/home/sikun/Documents/sgcrfpy/')
from sgcrf import SparseGaussianCRF
vec_s, vec_f = data_prep(task)
sgcrf = SparseGaussianCRF(learning_rate=0.1)
sgcrf.fit(vec_s, vec_f)
# loss = sgcrf.lnll
pred_f = sgcrf.predict(vec_s)
for k in range(vec_s.shape[0]):
print(pearsonr(pred_f[k], vec_f[k]))
def train_sgcrf(feature, gt, n_iter):
"""
Input: normalized feature and gt, number of iterations.
Output: SGCRF model
"""
print("Train SGCRF based on the output of RNN.")
model = SparseGaussianCRF(learning_rate=0.1,
lamL=0.01,
lamT=0.001,
n_iter=n_iter)
model.fit(feature, gt)
return model
def crf_models(df, df1, train_start_date, train_end_date, test_start_date,
test_end_date, request_type, CD, predictor_num):
"""
Trains 2 GCRF models on data from specified CD and Request Type which is assigned to fulfill request.
Uses specified start and end dates for training and testing to creat train and test sets.
Put this in a file called LAGCRF.py
"""
#Create Training and Testing Sets
dftrain = preprocessing(df, train_start_date, train_end_date)
dftrain = dftrain.reset_index(drop=True)
dftest = preprocessing(df1, test_start_date, test_end_date)
dftest = dftest.reset_index(drop=True)
#Reserve test set for training on all 3 models.
y_train, y_test = CreateTestSet(dftest, predictor_num)
y_test = y_test.reshape((-1, 1))
## 2 Models
#Model1: CD
modelCD = SparseGaussianCRF(lamL=0.1, lamT=0.1, n_iter=10000)
dftrainCD = dftrain[dftrain['CD'] == CD].reset_index(drop=True)
X_trainCD, X_testCD = CreateTrainSet(dftrainCD, predictor_num)
X_testCD = X_testCD.reshape((-1, 1))
modelCD.fit(X_trainCD, X_testCD)
y_predCD = modelCD.predict(y_train)
#Model2: Request_type
modelRT = SparseGaussianCRF(lamL=0.1, lamT=0.1, n_iter=10000)
dftrainRT = dftrain[dftrain['RequestType'] == request_type].reset_index(
drop=True)
X_trainRT, X_testRT = CreateTrainSet(dftrainRT, predictor_num)
X_testRT = X_testRT.reshape((-1, 1))
modelRT.fit(X_trainRT, X_testRT)
y_predRT = modelRT.predict(y_train)
#Average out all predictions
y_predFinal = (y_predCD + y_predRT) / 2
# Return models
return modelCD, modelRT
def Strukturni_predict_fun(train_index, test_index, ModelSTNo):
atribute = pd.read_csv('atribute')
output = pd.read_csv('output')
atribute = atribute.ix[:,1:].values
output = output.ix[:,1:].values
timeST = np.zeros([ModelSTNo])
R2 = np.zeros([ModelSTNo])
x_train, x_test = atribute[train_index,:], atribute[test_index,:]
y_train, y_test = output[train_index,:], output[test_index,:]
std_scl = StandardScaler()
std_scl.fit(x_train)
x_train = std_scl.transform(x_train)
x_test = std_scl.transform(x_test)
model = SparseGaussianCRF()
start_time = time.time()
model.fit(x_train, y_train)
y_SGCRF = model.predict(x_test).reshape(-1)
timeST[0] = time.time() - start_time
start_time = time.time()
model2 = Sequential()
model2.add(Dense(30, input_dim = x_train.shape[1], activation='relu'))
model2.add(Dense(25, activation='relu'))
model2.add(Dense(20, activation='relu'))
model2.add(Dense(7, activation='linear'))
model2.compile(loss='mean_absolute_error', optimizer='SGD')
ES = keras.callbacks.EarlyStopping(monitor='val_loss', min_delta=0, patience=50, verbose=0, mode='auto', baseline=None)
model2.fit(x_train, y_train, epochs=1200, batch_size=200,validation_data=(x_test, y_test), callbacks=[ES])
y_NN1 = model2.predict(x_test).reshape(-1)
timeST[1] = time.time() - start_time
start_time = time.time()
model3 = Sequential()
model3.add(Dense(35, input_dim = x_train.shape[1], activation='relu'))
model3.add(Dense(26, activation='relu'))
model3.add(Dense(22, activation='relu'))
model3.add(Dense(7, activation='linear'))
model3.compile(loss='mean_absolute_error', optimizer='SGD')
ES = keras.callbacks.EarlyStopping(monitor='val_loss', min_delta=0, patience=50, verbose=0, mode='auto', baseline=None)
model2.fit(x_train, y_train, epochs=1200, batch_size=200,validation_data=(x_test, y_test), callbacks=[ES])
y_NN2 = model3.predict(x_test).reshape(-1)
timeST[2] = time.time() - start_time
start_time = time.time()
model4 = Sequential()
model4.add(Dense(36, input_dim = x_train.shape[1], activation='relu'))
model4.add(Dense(27, activation='relu'))
model4.add(Dense(21, activation='relu'))
model4.add(Dense(7, activation='linear'))
model4.compile(loss='mean_absolute_error', optimizer='SGD')
ES = keras.callbacks.EarlyStopping(monitor='val_loss', min_delta=0, patience=50, verbose=0, mode='auto', baseline=None)
model4.fit(x_train, y_train, epochs=1200, batch_size=200,validation_data=(x_test, y_test), callbacks=[ES])
y_NN3 = model4.predict(x_test).reshape(-1)
timeST[3] = time.time() - start_time
R2[0] = r2_score(y_test.reshape(-1), y_SGCRF)
R2[1] = r2_score(y_test.reshape(-1), y_NN1)
R2[2] = r2_score(y_test.reshape(-1), y_NN2)
R2[3] = r2_score(y_test.reshape(-1), y_NN3)
return timeST, R2
iterations = 3 #number of GCRF models
data_path = '/home/mllab/Desktop/defazio-311/Data/SameDatapoints/Complaint/'
results_path = "/home/mllab/Desktop/defazio-311/Results/complaint/predicted_data_water.csv"
fh = open(results_path, "w")
fh.write("Predicted1,Real1,Predicted7,Real7\n")
filename = data_path + 'Water System.csv'
response_times, X_train, Y_train, X_test, Y_test = gs.getSamples(
filename, x_length, y_length)
model = SparseGaussianCRF(
lamL=0.1, lamT=0.1,
n_iter=10000) #lamL and lamT are regularization parameters
predictions = []
#run the model a few times and store predictions
for i in range(0, iterations):
model.fit(X_train, Y_train)
prediction = model.predict(X_test)
predictions.append(prediction)
predictions = np.array(predictions)
#average different GCRF's predictions
predictions = np.mean(predictions, axis=0)
def lacer(df, df1, train_start_date, train_end_date, test_start_date,
test_end_date, request_type, owner, CD,
predictor_num): #Once model is ready, replace df with csv
"""
Trains 3 GCRF models on data from specified CD, Request Type, and Owner which is assigned to fulfill request.
Uses specified start and end dates for training and testing to creat train and test sets.
"""
#Create Training and Testing Sets
dftrain = preprocessing(df, train_start_date, train_end_date)
dftrain = dftrain.reset_index(drop=True)
dftest = preprocessing(df1, test_start_date, test_end_date)
dftest = dftest.reset_index(drop=True)
#Reserve test set for training on all 3 models.
y_train, y_test = CreateTestSet(dftest, predictor_num)
y_test = y_test.reshape((-1, 1))
## 3 Models
#Model1: CD
modelCD = SparseGaussianCRF(lamL=0.1, lamT=0.1, n_iter=10000)
dftrainCD = dftrain[dftrain['CD'] == CD].reset_index(drop=True)
X_trainCD, X_testCD = CreateTrainSet(dftrainCD, predictor_num)
X_testCD = X_testCD.reshape((-1, 1))
modelCD.fit(X_trainCD, X_testCD)
y_predCD = modelCD.predict(y_train)
#Model2: Request_type
modelRT = SparseGaussianCRF(lamL=0.1, lamT=0.1, n_iter=10000)
dftrainRT = dftrain[dftrain['RequestType'] == request_type].reset_index(
drop=True)
X_trainRT, X_testRT = CreateTrainSet(dftrainRT, predictor_num)
X_testRT = X_testRT.reshape((-1, 1))
modelRT.fit(X_trainRT, X_testRT)
y_predRT = modelRT.predict(y_train)
#Model3: Owner
modelOwner = SparseGaussianCRF(lamL=0.1, lamT=0.1, n_iter=10000)
dftrainOwner = dftrain[dftrain['Owner'] == owner].reset_index(drop=True)
X_trainOwner, X_testOwner = CreateTrainSet(dftrainOwner, predictor_num)
X_testOwner = X_testOwner.reshape((-1, 1))
modelOwner.fit(X_trainOwner, X_testOwner)
y_predOwner = modelOwner.predict(y_train)
#Average out all predictions
y_predFinal = (y_predCD + y_predRT + y_predOwner) / 3
#Return metrics
return metrics(y_predFinal, y_test)