def test_squared_loss(make_whas500):
whas500_data = make_whas500(with_std=False, to_numeric=True)
model = GradientBoostingSurvivalAnalysis(loss="squared",
n_estimators=100,
max_depth=3,
random_state=0)
model.fit(whas500_data.x, whas500_data.y)
time_predicted = model.predict(whas500_data.x)
time_true = whas500_data.y["lenfol"]
event_true = whas500_data.y["fstat"]
rmse_all = numpy.sqrt(mean_squared_error(time_true, time_predicted))
assert round(abs(rmse_all - 580.23345259002951), 7) == 0
rmse_uncensored = numpy.sqrt(
mean_squared_error(time_true[event_true],
time_predicted[event_true]))
assert round(abs(rmse_uncensored - 383.10639243317951), 7) == 0
cindex = model.score(whas500_data.x, whas500_data.y)
assert round(abs(cindex - 0.9021810004), 7) == 0
with pytest.raises(
ValueError,
match="`fit` must be called with the loss option set to 'coxph'"
):
model.predict_survival_function(whas500_data.x)
with pytest.raises(
ValueError,
match="`fit` must be called with the loss option set to 'coxph'"
):
model.predict_cumulative_hazard_function(whas500_data.x)
def fit_and_score_features(X, y):
n_features = X.shape[1]
scores = np.empty(n_features)
m = GradientBoostingSurvivalAnalysis(verbose=True, n_estimators=500)
for j in range(n_features):
Xj = X[:, j:j + 1]
m.fit(Xj, y)
scores[j] = m.score(Xj, y)
return scores
def test_squared_loss(make_whas500):
whas500_data = make_whas500(with_std=False, to_numeric=True)
model = GradientBoostingSurvivalAnalysis(loss="squared", n_estimators=100, max_depth=3, random_state=0)
model.fit(whas500_data.x, whas500_data.y)
time_predicted = model.predict(whas500_data.x)
time_true = whas500_data.y["lenfol"]
event_true = whas500_data.y["fstat"]
rmse_all = numpy.sqrt(mean_squared_error(time_true, time_predicted))
assert round(abs(rmse_all - 580.23345259002951), 7) == 0
rmse_uncensored = numpy.sqrt(mean_squared_error(time_true[event_true], time_predicted[event_true]))
assert round(abs(rmse_uncensored - 383.10639243317951), 7) == 0
cindex = model.score(whas500_data.x, whas500_data.y)
assert round(abs(cindex - 0.9021810004), 7) == 0
def test_ipcwls_loss(make_whas500):
whas500_data = make_whas500(with_std=False, to_numeric=True)
model = GradientBoostingSurvivalAnalysis(loss="ipcwls", n_estimators=100, max_depth=3, random_state=0)
model.fit(whas500_data.x, whas500_data.y)
time_predicted = model.predict(whas500_data.x)
time_true = whas500_data.y["lenfol"]
event_true = whas500_data.y["fstat"]
rmse_all = numpy.sqrt(mean_squared_error(time_true, time_predicted))
assert round(abs(rmse_all - 590.5441693629117), 7) == 0
rmse_uncensored = numpy.sqrt(mean_squared_error(time_true[event_true], time_predicted[event_true]))
assert round(abs(rmse_uncensored - 392.97741487479743), 7) == 0
cindex = model.score(whas500_data.x, whas500_data.y)
assert round(abs(cindex - 0.8979161399), 7) == 0
verbose=args.verbose,
minibatch_frac=1.0,
Base=base_name_to_learner[args.base],
Score=eval(args.score),
)
train_losses = ngb.fit(X_train, Y_train, E_train)
forecast = ngb.pred_dist(X_test)
train_forecast = ngb.pred_dist(X_train)
print("NGB score: %.4f (val), %.4f (train)" % (
concordance_index_censored(E_test.astype(bool), Y_test,
-forecast.mean())[0],
concordance_index_censored(E_train.astype(bool), Y_train,
-train_forecast.mean())[0],
))
##
## sksurv
##
gbsa = GBSA(
n_estimators=args.n_est,
learning_rate=args.lr,
subsample=args.minibatch_frac,
verbose=args.verbose,
)
gbsa.fit(X_train, Y_join(Y_train, E_train))
print("GBSA score: %.4f (val), %.4f (train)" % (
gbsa.score(X_test, Y_join(Y_test, E_test)),
gbsa.score(X_train, Y_join(Y_train, E_train)),
))
X_train, X_val, Y_train, Y_val, E_train, E_val = train_test_split(X_train, Y_train, E_train, test_size=0.2)
ngb = NGBSurvival(Dist=eval(args.distn),
n_estimators=args.n_est,
learning_rate=args.lr,
natural_gradient=args.natural,
verbose=args.verbose,
minibatch_frac=1.0,
Base=base_name_to_learner[args.base],
Score=eval(args.score))
train_losses = ngb.fit(X_train, Y_train, E_train)
forecast = ngb.pred_dist(X_test)
train_forecast = ngb.pred_dist(X_train)
print('NGB score: %.4f (val), %.4f (train)' % (concordance_index_censored(E_test.astype(bool), Y_test, -forecast.mean())[0],
concordance_index_censored(E_train.astype(bool), Y_train, -train_forecast.mean())[0]
))
##
## sksurv
##
gbsa = GBSA(n_estimators=args.n_est,
learning_rate=args.lr,
subsample=args.minibatch_frac,
verbose=args.verbose)
gbsa.fit(X_train, Y_join(Y_train, E_train))
print('GBSA score: %.4f (val), %.4f (train)' % (gbsa.score(X_test, Y_join(Y_test, E_test)),
gbsa.score(X_train, Y_join(Y_train, E_train))))
n_estimators=args.n_est,
learning_rate=args.lr,
natural_gradient=args.natural,
verbose=args.verbose,
minibatch_frac=1.0,
Base=base_name_to_learner[args.base],
Score=eval(args.score)())
train_losses = ngb.fit(X_train, Y_train) #, X_val, Y_val)
forecast = ngb.pred_dist(X_test)
train_forecast = ngb.pred_dist(X_train)
print('NGB score: %.4f (val), %.4f (train)' %
(concordance_index_censored(Y_test['Event'], Y_test['Time'],
-forecast.mean())[0],
concordance_index_censored(Y_train['Event'], Y_train['Time'],
-train_forecast.mean())[0]))
#logger.tick(forecast, Y_test)
##
## sksurv
##
gbsa = GBSA(n_estimators=args.n_est,
learning_rate=args.lr,
subsample=args.minibatch_frac,
verbose=args.verbose)
gbsa.fit(X_train, Y_train)
print('GBSA score: %.4f (val), %.4f (train)' %
(gbsa.score(X_test, Y_test), gbsa.score(X_train, Y_train)))
#logger.save()
df2['T'] = T
X, y = get_x_y(df2, ['E', 'T'], pos_label=True)
for c in X.columns.values:
if c != 'AGE AT DOC':
X[c] = X[c].astype('category')
data_x_numeric = OneHotEncoder().fit_transform(X)
#%%
estimator = GradientBoostingSurvivalAnalysis(verbose=True,
n_estimators=500)
estimator.fit(data_x_numeric, y)
print(estimator.score(data_x_numeric, y))
print()
scores = fit_and_score_features(data_x_numeric.values, y)
print(
pd.Series(scores,
index=data_x_numeric.columns).sort_values(ascending=False))
pickle.dump(estimator, open('GradientRegressor.pkl', 'wb'))
#%%
from sklearn.feature_selection import SelectKBest
from sklearn.pipeline import Pipeline
pipe = Pipeline([('encode', OneHotEncoder()),
