def hotStartTrainRMSE(model_1,model_2,K, L, X1, X2, train_I, train_J, train_Y, train_op):
M = X1.shape[0]
N = X2.shape[0]
op1 = train_op['op1']
op2 = train_op['op2']
if model_1 == 'SCOAL':
model = op1['model']
predictions = predict_scoal(train_I, train_J, train_Y, M, N, model)
if model_1 == 'MF':
model = op1['model']
predictions = mf_predict(train_I,train_J,M,N,model)
if model_2 == 'SCOAL':
model = op2['model']
predictions += predict_scoal(train_I, train_J, train_Y, M, N, model)
if model_2 == 'MF':
model = op2['model']
predictions += mf_predict(train_I,train_J,M,N,model)
Z = sp.csr_matrix((train_Y, (train_I,train_J)), shape=(M,N))
nonzero_Z = np.array(Z[(train_I,train_J)]).ravel()
hotStartTrainRMSE = np.sqrt(np.mean((predictions.flatten()-nonzero_Z)**2))
return hotStartTrainRMSE
def train_scoal(K, L, X1, X2, train_I, train_J, train_Y, reg,residue):
# Train SCOAL
learner = 'ridge'
scoal_op = run_scoal(K, L, X1, X2, train_I, train_J, residue, learner, reg)
# Predict using SCOAL
M = X1.shape[0]
N = X2.shape[0]
model = scoal_op['model']
predictions = predict_scoal(train_I, train_J, residue, M, N, model)
# Calculate residue
residue = train_Y - predictions
return scoal_op, residue
