if i == k:
testUser = KshuffleUserId[i]
else:
trainUser.extend(KshuffleUserId[i])
data['rnd'] = np.random.random(data.shape[0])
trainInd = (list(map(lambda x: x in trainUser, data['userId'])) |
(data['rnd'] < 0.7))
dataTrain = data[trainInd]
dataTest = data[np.invert(trainInd)]
bob = graphALS(num_factors=5, num_iterations=20, verbose=True)
bob.fit(dataTrain[['userId', 'movieId', 'rating']], userContextSparseGraph,
itemContextSparseGraph)
Rhat = bob.user_vectors.dot(bob.item_vectors.T)
R_test = csr_matrix(
(dataTest['rating'], (dataTest['userId'], dataTest['movieId'])),
shape=(numUser, numItem))
W = R_test.nonzero()
rmse = RMSE(R_test, Rhat, W)
print(rmse)
R = csr_matrix(
(dataTrain['rating'], (dataTrain['userId'], dataTrain['movieId'])),
shape=(numUser, numItem))
bobALS = ALS(5)
bobALS.fit(R)
Rhat_ALS = bobALS.U.dot(bobALS.V.T)
rmseALS = RMSE(R_test, Rhat_ALS, W)
print(rmseALS)
for k in range(K):
print("cross-val", k)
trainUser = []
testUser = []
for i in range(K):
if i == k:
testUser = KshuffleUserId[i]
else:
trainUser.extend(KshuffleUserId[i])
data["rnd"] = np.random.random(data.shape[0])
trainInd = list(map(lambda x: x in trainUser, data["userId"])) | (data["rnd"] < 0.7)
dataTrain = data[trainInd]
dataTest = data[np.invert(trainInd)]
bob = graphALS(num_factors=5, num_iterations=20, verbose=True)
bob.fit(dataTrain[["userId", "movieId", "rating"]], userContextSparseGraph, itemContextSparseGraph)
Rhat = bob.user_vectors.dot(bob.item_vectors.T)
R_test = csr_matrix((dataTest["rating"], (dataTest["userId"], dataTest["movieId"])), shape=(numUser, numItem))
W = R_test.nonzero()
rmse = RMSE(R_test, Rhat, W)
print(rmse)
R = csr_matrix((dataTrain["rating"], (dataTrain["userId"], dataTrain["movieId"])), shape=(numUser, numItem))
bobALS = ALS(5)
bobALS.fit(R)
Rhat_ALS = bobALS.U.dot(bobALS.V.T)
rmseALS = RMSE(R_test, Rhat_ALS, W)
print(rmseALS)
#alphas = params['ALS-WR']['params']['alpha']
#epss = params['ALS-WR']['params']['eps']
regs = ["default"]
lambdas = [0.8]
rank = [20]
alphas = [10]
epss= [10]
current_perf = dict.fromkeys(['model','reg','lambda','rank','alpha','eps','crossval'])
for elt in itertools.product(*[regs,lambdas,rank,alphas,epss,[0]]):
reg, l, r, alpha, eps, k = elt
current_perf.update({'model':'ALS-WR','reg':reg,'lambda':l,
'rank':r,'alpha':alpha,'eps':eps,
'crossval':k})
print(current_perf)
bob = ALS(d=r,num_users=numUser,num_items=numItem,lbda=l,seed=0,
reg=reg,verbose=True)
t0 = time()
bob.fitImplicit(data[data['cv']!=k],alpha=alpha,c="log",eps=eps)
T = time()-t0
Rhat = bob.U.dot(bob.V.T)
R_test = sparseMatrix(data,k,include=True,names=list(data.columns)[:3])
rank = rankMeasure(R_test,Rhat)
print(rank)
ind = getLine_fromdict(perf,current_perf)
perf.loc[ind] = ['ALS-WR',reg,l,r,alpha,eps,k,rank,T]
print('-'*50)
#==============================================================================
# graphALS
#==============================================================================
for elt in itertools.product(*[regs, lambdas, rank, alphas, epss, [0]]):
reg, l, r, alpha, eps, k = elt
current_perf.update({
'model': 'ALS-WR',
'reg': reg,
'lambda': l,
'rank': r,
'alpha': alpha,
'eps': eps,
'crossval': k
})
print(current_perf)
bob = ALS(d=r,
num_users=numUser,
num_items=numItem,
lbda=l,
seed=0,
reg=reg,
verbose=True)
t0 = time()
bob.fitImplicit(data[data['cv'] != k], alpha=alpha, c="log", eps=eps)
T = time() - t0
Rhat = bob.U.dot(bob.V.T)
R_test = sparseMatrix(data,
k,
include=True,
names=list(data.columns)[:3])
rank = rankMeasure(R_test, Rhat)
print(rank)
ind = getLine_fromdict(perf, current_perf)
perf.loc[ind] = ['ALS-WR', reg, l, r, alpha, eps, k, rank, T]
# ALS-WR
#==============================================================================
if params['ALS-WR']['learn']=="True":
print("Learn ALS-WR")
lambdas = params['ALS-WR']['params']['lambda']
rank = params['ALS-WR']['params']['rank']
current_perf = dict.fromkeys(['model','lambda','rank','crossval'])
for elt in itertools.product(*[lambdas,rank,range(K)]):
l, r, k = elt
current_perf['model'] = 'ALS-WR'
current_perf['lambda'] = l
current_perf['rank'] = r
current_perf['crossval'] = k
print(current_perf)
bob = ALS(d=r,num_users=numUser,num_items=numItem,lbda=l,parallel=True)
train = data[data['cv']!=k][['row','col','val']].to_dict(orient='list')
test = data[data['cv']==k][['row','col','val']].to_dict(orient='list')
t0 = time()
bob.fit(train)
T = time()-t0
Rhat = bob.U.dot(bob.V.T)
R_test = sparse_matrix(test,numUser,numItem)
rmse = RMSE(R_test,Rhat)
print(rmse)
ind = getLine_fromdict(perf,current_perf)
perf.loc[ind,['model','rank','lambda','crossval','rmse','runningTime']] = ['ALS-WR',r,l,k,rmse,T]
print('-'*50)