def fit(self, X, y):
"""パラメータの推定を行う.
Parameters
----------
X : array-like, shape = (n_samples, n_features)
Data matrix
y : structured array, shape = (n_samples,)
[(打切り変数, 観測時間), ...]
Returns
-------
self
"""
X, event, time = check_arrays_survival(X, y)
if self.alpha < 0:
raise ValueError('alpha must be positive, but was %r' % self.alpha)
log_likelihood = WeibullRMLogLikelihood(X, event, time, self.alpha)
verbose = self.verbose
n_params = X.shape[1] + 1
parameters = np.zeros(n_params)
parameters[-1] = 0.5 # sigmaは分母として用いるので0
i = 0
while True:
if i >= self.n_iter:
if verbose:
print('iter {:>6d}: reached maximum number of iterations. Stopping.'.format(i + 1))
print('Optimization did not converge: Maximum number of iterations has been exceeded.')
break
log_likelihood.update(parameters)
delta = solve(log_likelihood.hessian, log_likelihood.gradient,
overwrite_a=False, overwrite_b=False, check_finite=False)
if not np.all(np.isfinite(delta)):
raise ValueError('search direction contains NaN or infinite values')
parameters_new = parameters - delta
if verbose:
print('iter {: > 6d}: update={}'.format(i + 1, delta))
parameters = parameters_new
res = np.abs(delta)
if np.all(res < self.tol):
break
else:
i += 1
self.w_ = parameters[:-1]
self.sigma_ = parameters[-1]
# self.hessian, self.gradient は attribute で残しときたい.
self.hessian = log_likelihood.hessian
self.gradient = log_likelihood.gradient
return self
def fit(self, X, y):
"""パラメータの推定を行う.
Parameters
----------
X : pandas.DataFrame, shape = (n_samples, n_features)
Data matrix
y : structured array, shape = (n_samples,)
[(打切り変数, 観測時間), ...]
Returns
-------
self
"""
# lifelines.CoxPHFitter の fit() に渡すためのデータフレームを整形.
feature_name_list = X.columns.values.tolist()
name_event, name_time = y.dtype.names
x, event, time = check_arrays_survival(X, y)
y_df = pd.DataFrame([time, event]).T
y_df.columns = [name_time, name_event]
x_df = pd.DataFrame(x, columns=feature_name_list)
df = pd.merge(x_df, y_df, left_index=True, right_index=True)
super().fit(df,
duration_col=name_time,
event_col=name_event,
strata=self.strat_list)
# self.w_ は attribute で残しときたい.
self.w_ = self.hazards_.values[0]
self.feature_name_list = feature_name_list
return self
def fit(self, X, y, id_):
"""モデルの学習処理.
Parameters
----------
X : array-like, shape = (n_samples, n_features)
Data matrix
y : structured array, shape = (n_samples,)
[(打切り変数, 観測時間), ...]
Returns
-------
self
"""
X, event, time = check_arrays_survival(X, y)
if self.alpha < 0:
raise ValueError('alpha must be positive, but was %r' % self.alpha)
init_parameters = np.zeros(X.shape[1])
parameter = optimize.minimize(logistic_log_likelihood, init_parameters, args=(X, event, time, id_), method='Nelder-Mead')
print(parameter)
return self
def fit(self, X, y, sample_weight=None):
X, _, time = check_arrays_survival(X, y)
return super().fit(X, time)