def test_dict_completion_normalise(backend):
# Generate some toy data.
rng = np.random.RandomState(0)
U = rng.rand(50, 3)
V = rng.rand(3, 20)
X = np.dot(U, V)
mf = DictCompleter(n_components=3, max_n_iter=100, alpha=1e-3,
random_state=0,
backend=backend,
verbose=0, detrend=True)
mf.fit(X)
Y = np.dot(mf.P_, mf.Q_)
Y += mf.col_mean_[np.newaxis, :]
Y += mf.row_mean_[:, np.newaxis]
Y2 = mf.predict(X).toarray()
assert_array_almost_equal(Y, Y2)
rmse = np.sqrt(np.mean((X - Y) ** 2))
rmse2 = mf.score(X)
assert_almost_equal(rmse, rmse2)
def test_dict_completion_missing(backend):
# Generate some toy data.
rng = np.random.RandomState(0)
U = rng.rand(100, 4)
V = rng.rand(4, 20)
X = np.dot(U, V)
X = sp.csr_matrix(X)
X_tr, X_te = train_test_split(X, train_size=0.95)
X_tr = sp.csr_matrix(X_tr)
X_te = sp.csr_matrix(X_te)
mf = DictCompleter(
n_components=4,
max_n_iter=400,
alpha=1,
random_state=0,
backend=backend,
detrend=True,
verbose=0,
)
mf.fit(X_tr)
X_pred = mf.predict(X_te)
rmse = sqrt(np.sum((X_te.data - X_pred.data)**2) / X_te.data.shape[0])
X_te_c, _, _ = csr_center_data(X_te)
rmse_c = sqrt(np.sum((X_te.data - X_te_c.data)**2) / X_te.data.shape[0])
assert (rmse < rmse_c)
def test_dict_completion_missing(backend):
# Generate some toy data.
rng = np.random.RandomState(0)
U = rng.rand(100, 4)
V = rng.rand(4, 20)
X = np.dot(U, V)
X = sp.csr_matrix(X)
X_tr, X_te = train_test_split(X, train_size=0.95)
X_tr = sp.csr_matrix(X_tr)
X_te = sp.csr_matrix(X_te)
mf = DictCompleter(
n_components=4,
max_n_iter=400,
alpha=1,
random_state=0,
backend=backend,
detrend=True,
verbose=0,
)
mf.fit(X_tr)
X_pred = mf.predict(X_te)
rmse = sqrt(np.sum((X_te.data - X_pred.data)**2) / X_te.data.shape[0])
row_mean_, col_mean_ = compute_biases(X, beta=mf.beta, inplace=False)
X_te_c = copy.deepcopy(X_te)
for i in range(X_te_c.shape[0]):
X_te_c.data[X_te_c.indptr[i]:X_te_c.indptr[i + 1]] -= row_mean_[i]
X_te_c.data -= col_mean_.take(X_te_c.indices, mode='clip')
rmse_c = sqrt(np.sum((X_te.data - X_te_c.data)**2) / X_te.data.shape[0])
assert (rmse < rmse_c)
def test_dict_completion_missing(backend):
# Generate some toy data.
rng = np.random.RandomState(0)
U = rng.rand(100, 4)
V = rng.rand(4, 20)
X = np.dot(U, V)
X = sp.csr_matrix(X)
X_tr, X_te = train_test_split(X, train_size=0.95)
X_tr = sp.csr_matrix(X_tr)
X_te = sp.csr_matrix(X_te)
mf = DictCompleter(n_components=4, max_n_iter=400, alpha=1,
random_state=0,
backend=backend,
detrend=True,
verbose=0, )
mf.fit(X_tr)
X_pred = mf.predict(X_te)
rmse = sqrt(np.sum((X_te.data - X_pred.data) ** 2) / X_te.data.shape[0])
row_mean_, col_mean_ = compute_biases(X, beta=mf.beta, inplace=False)
X_te_c = copy.deepcopy(X_te)
for i in range(X_te_c.shape[0]):
X_te_c.data[X_te_c.indptr[i]:X_te_c.indptr[i + 1]] -= row_mean_[i]
X_te_c.data -= col_mean_.take(X_te_c.indices, mode='clip')
rmse_c = sqrt(np.sum((X_te.data - X_te_c.data) ** 2) / X_te.data.shape[0])
assert (rmse < rmse_c)
def test_dict_completion_normalise(backend):
# Generate some toy data.
rng = np.random.RandomState(0)
U = rng.rand(50, 3)
V = rng.rand(3, 20)
X = np.dot(U, V)
mf = DictCompleter(n_components=3,
max_n_iter=100,
alpha=1e-3,
random_state=0,
backend=backend,
verbose=0,
detrend=True)
mf.fit(X)
Y = np.dot(mf.P_, mf.Q_)
Y += mf.col_mean_[np.newaxis, :]
Y += mf.row_mean_[:, np.newaxis]
Y2 = mf.predict(X).toarray()
assert_array_almost_equal(Y, Y2)
rmse = np.sqrt(np.mean((X - Y)**2))
rmse2 = mf.score(X)
assert_almost_equal(rmse, rmse2)
def test_dict_completion(backend):
# Generate some toy data.
rng = np.random.RandomState(0)
U = rng.rand(50, 3)
V = rng.rand(3, 20)
X = np.dot(U, V)
mf = DictCompleter(
n_components=3,
max_n_iter=100,
alpha=1e-3,
random_state=0,
detrend=False,
backend=backend,
verbose=0,
)
mf.fit(X)
Y = np.dot(mf.code_, mf.components_)
Y2 = mf.predict(X).toarray()
assert_array_almost_equal(Y, Y2)
rmse = np.sqrt(np.mean((X - Y)**2))
rmse2 = mf.score(X)
assert_almost_equal(rmse, rmse2)
def test_dict_completion_missing(backend):
# Generate some toy data.
rng = np.random.RandomState(0)
U = rng.rand(100, 4)
V = rng.rand(4, 20)
X = np.dot(U, V)
X = sp.csr_matrix(X)
X_tr, X_te = train_test_split(X, train_size=0.95)
X_tr = sp.csr_matrix(X_tr)
X_te = sp.csr_matrix(X_te)
mf = DictCompleter(n_components=4, max_n_iter=400, alpha=1,
random_state=0,
backend=backend,
detrend=True,
verbose=0, )
mf.fit(X_tr)
X_pred = mf.predict(X_te)
rmse = sqrt(np.sum((X_te.data - X_pred.data) ** 2) / X_te.data.shape[0])
X_te_c, _, _ = csr_center_data(X_te)
rmse_c = sqrt(np.sum((X_te.data - X_te_c.data) ** 2) / X_te.data.shape[0])
assert(rmse < rmse_c)
def test_dict_completion(backend):
# Generate some toy data.
rng = np.random.RandomState(0)
U = rng.rand(50, 3)
V = rng.rand(3, 20)
X = np.dot(U, V)
mf = DictCompleter(n_components=3, max_n_iter=100, alpha=1e-3,
random_state=0,
detrend=False,
backend=backend,
verbose=0, )
mf.fit(X)
Y = np.dot(mf.code_, mf.components_)
Y2 = mf.predict(X).toarray()
assert_array_almost_equal(Y, Y2)
rmse = np.sqrt(np.mean((X - Y) ** 2))
rmse2 = mf.score(X)
assert_almost_equal(rmse, rmse2)
X_pred = mf.predict(self.X_te)
rmse = np.sqrt(np.mean((X_pred.data - self.X_te.data)**2))
self.rmse.append(rmse)
print('Test RMSE: ', rmse)
self.test_time += time.clock() - test_time
self.times.append(time.clock() - self.start_time - self.test_time)
random_state = 0
mf = DictCompleter(n_components=30,
alpha=.001,
beta=0,
verbose=3,
batch_size=1000,
detrend=True,
offset=0,
projection='partial',
random_state=0,
learning_rate=.9,
n_epochs=5,
backend='python')
# Need to download from spira
X = load_movielens('10m')
X_tr, X_te = train_test_split(X, train_size=0.75, random_state=random_state)
X_tr = X_tr.tocsr()
X_te = X_te.tocsr()
cb = Callback(X_tr, X_te)
mf.set_params(callback=cb)
t0 = time.time()
rmse = np.sqrt(np.mean((X_pred.data - self.X_te.data) ** 2))
self.rmse.append(rmse)
print("Test RMSE: ", rmse)
self.test_time += time.clock() - test_time
self.times.append(time.clock() - self.start_time - self.test_time)
random_state = 0
mf = DictCompleter(
n_components=30,
alpha=0.001,
beta=0,
verbose=3,
batch_size=1000,
detrend=True,
offset=0,
projection="partial",
random_state=0,
learning_rate=0.9,
n_epochs=5,
backend="python",
)
# Need to download from spira
X = load_movielens("10m")
X_tr, X_te = train_test_split(X, train_size=0.75, random_state=random_state)
X_tr = X_tr.tocsr()
X_te = X_te.tocsr()
cb = Callback(X_tr, X_te)
mf.set_params(callback=cb)
rmse_tr = np.sqrt(np.mean((X_pred.data - self.X_tr.data) ** 2))
self.rmse.append(rmse)
self.rmse_tr.append(rmse_tr)
self.q.append(mf.Q_[1, :10].copy())
self.test_time += time.clock() - test_time
self.times.append(time.clock() - self.start_time - self.test_time)
random_state = 0
mf = DictCompleter(n_components=30, alpha=.8, verbose=5,
batch_size=60, detrend=True,
offset=0,
impute=False,
fit_intercept=True,
random_state=0,
learning_rate=0.8,
max_n_iter=60000,
backend='c')
# Need to download from spira
X = load_movielens('1m')
X_tr, X_te = train_test_split(X, train_size=0.75,
random_state=random_state)
X_tr = X_tr.tocsr()
X_te = X_te.tocsr()
cb = Callback(X_tr, X_te)
mf.set_params(callback=cb)
mf.fit(X_tr)
self.rmse.append(rmse)
self.rmse_tr.append(rmse_tr)
self.q.append(mf.Q_[1, :10].copy())
self.test_time += time.clock() - test_time
self.times.append(time.clock() - self.start_time - self.test_time)
random_state = 0
mf = DictCompleter(n_components=30,
alpha=.8,
verbose=5,
batch_size=60,
detrend=True,
offset=0,
impute=False,
fit_intercept=True,
random_state=0,
learning_rate=0.8,
max_n_iter=60000,
backend='c')
# Need to download from spira
X = load_movielens('1m')
X_tr, X_te = train_test_split(X, train_size=0.75, random_state=random_state)
X_tr = X_tr.tocsr()
X_te = X_te.tocsr()
cb = Callback(X_tr, X_te)
mf.set_params(callback=cb)
mf.fit(X_tr)
trace_dir = expanduser('~/output/modl/recsys_bias')
estimator_grid = {
'cd': {
'estimator': ExplicitMF(
n_components=30,
detrend=True,
),
'name': 'Coordinate descent'
},
'dl': {
'estimator':
DictCompleter(n_components=30,
detrend=True,
projection='full',
fit_intercept=True,
backend='c'),
'name':
'Proposed online masked MF'
},
'dl_partial': {
'estimator':
DictCompleter(n_components=30,
detrend=True,
projection='partial',
fit_intercept=True,
backend='c'),
'name':
'Proposed algorithm'
' (with partial projection)'