def predict(self, X: spa.csr_matrix):
''' Predict class of X'''
if X.ndim == 1:
X = X.reshape(1, -1)
N, D = X.shape
class_probs = LR.logisticClassProbability(X, self.b, self.W)
y = class_probs.argmax(axis=1)
return self.encoder.inverse_transform(y)
def fit(self,
X: spa.csr_matrix,
Y: spa.csr_matrix,
X_val=None,
Y_val=None):
''' Fit model to data. X_val and Y_val are only used to report accuracy
during optimization they do not affect the fitted W,b parameters'''
if X.ndim == 1:
X = X.reshape(1, -1)
N, D = X.shape
self.encoder = LabelEncoder()
y = self.encoder.fit_transform(Y)
K = len(self.encoder.classes_)
Z = np.zeros((N, K), dtype=int)
Z[np.arange(N), y] = 1
if not (X_val is None):
N_val = len(X_val)
y_val = self.encoder.transform(Y_val)
Z_val = np.zeros((N_val, K), dtype=int)
Z_val[np.arange(N_val), y_val] = 1
else:
Z_val = None
b_guess = np.zeros(K)
W_guess = np.random.normal(0, 1, (K, D)) / np.sqrt(D)
self.b, self.W = LR.optimize_logistic_weights(
X,
Z,
b_guess,
W_guess,
X_val=X_val,
Z_val=Z_val,
penalty=self.penalty,
learning_rate=self.learning_rate,
batch_size=self.batch_size,
tol=self.tol,
max_iter=self.max_iter,
verbose=self.verbose)
