def train_test(self, x_train, y_train, x_test=None, cuisines=None, k=15):
torch.cuda.empty_cache()
x_train = np.array(x_train)
testing = x_test is not None
if testing:
x_tr = torch.tensor(x_train)
y_tr = torch.tensor(y_train)
x_val = torch.tensor(x_test)
dist = self.get_dist(x_tr, x_val)
y_pred = self.predict(dist, y_tr, k)
ids = [cuisine.id for cuisine in cuisines]
pred_cuisines = [
self.dataset.id2cuisine[label] for label in y_pred
]
self._write2csv(ids, pred_cuisines)
else:
shuffle_idx = torch.randperm(x_train.shape[0])
x_train = torch.tensor(x_train).float()
y_train = torch.tensor(y_train)
x_train = x_train[shuffle_idx]
y_train = y_train[shuffle_idx]
x_val = x_train[35000:]
x_tr = x_train[:35000]
y_val = y_train[35000:]
y_tr = y_train[:35000]
use_DML = False
if use_DML:
x_val = x_train[5000:6000]
x_tr = x_train[:5000]
y_val = y_train[5000:6000]
y_tr = y_train[:20000]
x_tr, x_val = self.PCA(x_tr, x_val, 64)
lmnn = LMNN(k=15, learn_rate=1e-6, min_iter=50, max_iter=100)
lmnn.fit(x_tr.numpy(), y_tr.numpy())
M = lmnn.get_mahalanobis_matrix()
M = torch.tensor(M).float()
n, d = x_val.shape
m = x_tr.shape[0]
x0 = x_tr.unsqueeze(1).expand(-1, n,
-1).contiguous().view(-1, d)
x1 = x_val.unsqueeze(0).expand(m, -1,
-1).contiguous().view(-1, d)
x = x0 - x1
dist0 = torch.mm(M, x.t().contiguous())
dists = dist0.t().contiguous() * x
dist = dists.sum(1).view(m, n)
else:
x_tr, x_val = self.PCA(x_tr, x_val, 500)
dist = self.get_dist(x_tr, x_val).cpu()
for k in [1, 3, 5, 8, 10, 15, 20, 25, 30]:
y_pred = self.predict(dist, y_tr, k)
acc = (y_pred == y_val).sum().float().numpy() / y_val.shape[0]
print("K=", k, " acc=", acc)
torch.cuda.empty_cache()
def test_lmnn(self): lmnn = LMNN(k=5, learn_rate=1e-6, verbose=False) lmnn.fit(self.X, self.y) L = lmnn.components_ assert_array_almost_equal(L.T.dot(L), lmnn.get_mahalanobis_matrix())
def main(params):
initialize_results_dir(params.get('results_dir'))
backup_params(params, params.get('results_dir'))
print('>>> loading data...')
X_train, y_train, X_test, y_test = LoaderFactory().create(
name=params.get('dataset'),
root=params.get('dataset_dir'),
random=True,
seed=params.getint('split_seed'))()
print('<<< data loaded')
print('>>> computing psd matrix...')
if params.get('algorithm') == 'identity':
psd_matrix = np.identity(X_train.shape[1], dtype=X_train.dtype)
elif params.get('algorithm') == 'nca':
nca = NCA(init='auto',
verbose=True,
random_state=params.getint('algorithm_seed'))
nca.fit(X_train, y_train)
psd_matrix = nca.get_mahalanobis_matrix()
elif params.get('algorithm') == 'lmnn':
lmnn = LMNN(init='auto',
verbose=True,
random_state=params.getint('algorithm_seed'))
lmnn.fit(X_train, y_train)
psd_matrix = lmnn.get_mahalanobis_matrix()
elif params.get('algorithm') == 'itml':
itml = ITML_Supervised(verbose=True,
random_state=params.getint('algorithm_seed'))
itml.fit(X_train, y_train)
psd_matrix = itml.get_mahalanobis_matrix()
elif params.get('algorithm') == 'lfda':
lfda = LFDA()
lfda.fit(X_train, y_train)
psd_matrix = lfda.get_mahalanobis_matrix()
elif params.get('algorithm') == 'arml':
learner = TripleLearner(
optimizer=params.get('optimizer'),
optimizer_params={
'lr': params.getfloat('lr'),
'momentum': params.getfloat('momentum'),
'weight_decay': params.getfloat('weight_decay'),
},
criterion=params.get('criterion'),
criterion_params={'calibration': params.getfloat('calibration')},
n_epochs=params.getint('n_epochs'),
batch_size=params.getint('batch_size'),
random_initialization=params.getboolean('random_initialization',
fallback=False),
update_triple=params.getboolean('update_triple', fallback=False),
device=params.get('device'),
seed=params.getint('learner_seed'))
psd_matrix = learner(X_train,
y_train,
n_candidate_mins=params.getint('n_candidate_mins',
fallback=1))
else:
raise Exception('unsupported algorithm')
print('<<< psd matrix got')
np.savetxt(os.path.join(params.get('results_dir'), 'psd_matrix.txt'),
psd_matrix)
def test_lmnn(self): lmnn = LMNN(k=5, learn_rate=1e-6, verbose=False) lmnn.fit(self.X, self.y) L = lmnn.transformer_ assert_array_almost_equal(L.T.dot(L), lmnn.get_mahalanobis_matrix())
