def fit(self, X, y, max_iter=MAX_ITER):
"""
Trains with supplied data.
Args:
:param X: each position contains the data of a task as an (m, n)
np.array with data (rows are samples, cols are features).
:param y: each position contains the labels of a task as an (m)
np.array.
Returns:
W (np.array): (n, T) array with estimated parameters of all tasks.
cost (np.array): cost at the end of each iteration.
time (float): number of seconds spent in training.
"""
n_tasks = len(X)
n_feats = X[0].shape[1]
W = np.random.randn(n_feats, n_tasks)
start = time.time()
cost_function = 0
X = self.normalize_data(X)
X = self.add_bias(X)
for t in range(n_tasks):
#print('Training {} task with lasso regression'.format(t))
lasso = Fista(self, self.lambda_1)
w = lasso.fit(xk=W[:, t], A=X[t], b=y[t], ind=self.groups,
max_iter=max_iter)
W[:, t] = w
stop = time.time() - start
self.W = W
return W, np.array([cost_function]), stop
def fit(self, X, y):
"""
Trains with supplied data.
Args:
:param X: each position contains the data of a task as an (m, n)
np.array with data (rows are samples, cols are features).
:param y: each position contains the labels of a task as an (m)
np.array.
Returns:
W (np.array): (n, T) array with estimated parameters of all tasks.
cost (np.array): cost at the end of each iteration.
time (float): number of seconds spent in training.
Uses optimization/solve_fista.py.
"""
X = self.normalize_data(X)
X = self.add_bias(X)
n_tasks = len(X)
n_feats = X[0].shape[1]
W = np.random.randn(n_feats, n_tasks)
cost_function = 0
start = time.time()
for t in range(n_tasks):
#print('Training task {} with group lasso'.format(t))
fista = Fista(self, self.lambda_1)
w_opt = fista.fit(W[:, t], X[t], y[t], self.groups,
max_iter=self.max_iter)
W[:, t] = w_opt
cost_function += self.cost(X[t], y[t], W[:, t])
stop = time.time() - start
self.W = W
return W, np.array([cost_function]), stop
def fit(self, w, X, y):
"""
Uses optimization/fista.py to optimize Wt.
See file for help.
"""
fista = Fista(self, self.lambda_3)
w_opt = fista.fit(w, X, y, self.inds, max_iter=self.max_iter)
return w_opt