def test_verbose_int(verbose):
expected_line_count = {5: 3, False: 0, True: 10}
pool = Pool(TRAIN_FILE, column_description=CD_FILE)
tmpfile = 'test_data_dumps'
with LogStdout(open(tmpfile, 'w')):
cv(pool, {"iterations": 10, "random_seed": 0, "loss_function": "Logloss"}, verbose=verbose)
with open(tmpfile, 'r') as output:
assert(sum(1 for line in output) == expected_line_count[verbose])
with LogStdout(open(tmpfile, 'w')):
train(pool, {"iterations": 10, "random_seed": 0, "loss_function": "Logloss"}, verbose=verbose)
with open(tmpfile, 'r') as output:
assert(sum(1 for line in output) == expected_line_count[verbose])
return local_canonical_file(remove_time_from_json(JSON_LOG_PATH))
def test_cv_query():
pool = Pool(QUERYWISE_TRAIN_FILE, column_description=QUERYWISE_CD_FILE)
results = cv(pool, {"iterations": 5, "random_seed": 0, "loss_function": "QueryRMSE"})
assert "train-QueryRMSE-mean" in results
prev_value = results["train-QueryRMSE-mean"][0]
for value in results["train-QueryRMSE-mean"][1:]:
assert value < prev_value
prev_value = value
def test_cv_pairs():
pool = Pool(QUERYWISE_TRAIN_FILE, column_description=QUERYWISE_CD_FILE, pairs=QUERYWISE_TRAIN_PAIRS_FILE)
results = cv(pool, {"iterations": 5, "random_seed": 8, "loss_function": "PairLogit"})
assert "train-PairLogit-mean" in results
prev_value = results["train-PairLogit-mean"][0]
for value in results["train-PairLogit-mean"][1:]:
assert value < prev_value
prev_value = value
def test_cv():
pool = Pool(TRAIN_FILE, column_description=CD_FILE)
results = cv(pool, {"iterations": 5, "random_seed": 0, "loss_function": "Logloss"})
assert "train-Logloss-mean" in results
prev_value = results["train-Logloss-mean"][0]
for value in results["train-Logloss-mean"][1:]:
assert value < prev_value
prev_value = value
def test_cv():
pool = Pool(TRAIN_FILE, column_description=CD_FILE)
results = cv({"iterations": 5, "random_seed": 0, "loss_function": "Logloss"}, pool)
assert isinstance(results, dict)
assert "Logloss_train_avg" in results
prev_value = results["Logloss_train_avg"][0]
for value in results["Logloss_train_avg"][1:]:
assert value < prev_value
prev_value = value
def test_cv_pairs():
pool = Pool(QUERYWISE_TRAIN_FILE,
column_description=QUERYWISE_CD_FILE,
pairs=QUERYWISE_TRAIN_PAIRS_FILE)
results = cv(pool, {
"iterations": 5,
"random_seed": 8,
"loss_function": "PairLogit"
})
assert "train-PairLogit-mean" in results
prev_value = results["train-PairLogit-mean"][0]
for value in results["train-PairLogit-mean"][1:]:
assert value < prev_value
prev_value = value
def fit(self, X_train, y_train):
bst = cv(
Pool(X_train, y_train),
self.params
)
best_rounds = int(bst['test-{}-mean'.format(self.metric)].idxmax() * 1.5) + 1
print('Best Iteration: {}'.format(best_rounds))
self.params['iterations'] = best_rounds
self.model = CatBoostClassifier(**self.params)
self.model.fit(
X_train, y_train
)
def cv_model(params: dict, positions: tuple, date: pd.Timestamp,
data_pool_func):
"""Кросс-валидирует модель по RMSE, нормированному на СКО набора данных
Осуществляется проверка, что не достигнут максимум итераций, возвращается RMSE, R2 и параметры модели с оптимальным
количеством итераций в формате целевой функции hyperopt
Parameters
----------
params
Словарь с параметрами модели: ключ 'data' - параметры данных, ключ 'model' - параметры модели
positions
Кортеж тикеров, для которых необходимо осуществить кросс-валидацию
date
Дата, для которой необходимо осуществить кросс-валидацию
data_pool_func
Функция для получения catboost.Pool с данными
Returns
-------
dict
Словарь с результатом в формате hyperopt:
ключ 'loss' - нормированная RMSE на кросс-валидации (для hyperopt),
ключ 'status' - успешного прохождения (для hyperopt),
ключ 'std' - RMSE на кросс-валидации,
ключ 'r2' - 1- нормированная RMSE на кросс-валидации в квадрате,
ключ 'data' - параметры данных,
ключ 'model' - параметры модели, в которые добавлено оптимальное количество итераций градиентного бустинга на
кросс-валидации и общие настройки
"""
data_params = params["data"]
data = data_pool_func(positions, date, **data_params)
pool_std = np.array(data.get_label()).std()
model_params = make_model_params(params)
scores = catboost.cv(pool=data,
params=model_params,
fold_count=FOLDS_COUNT)
if len(scores) == MAX_ITERATIONS:
raise ValueError(
f"Необходимо увеличить MAX_ITERATIONS = {MAX_ITERATIONS}")
index = scores["test-RMSE-mean"].idxmin()
model_params["iterations"] = index + 1
return dict(
loss=scores.loc[index, "test-RMSE-mean"] / pool_std,
status=hyperopt.STATUS_OK,
std=scores.loc[index, "test-RMSE-mean"],
r2=1 - (scores.loc[index, "test-RMSE-mean"] / pool_std)**2,
data=data_params,
model=model_params,
)
def objective(space):
global best_score, trials_count
# if os.path.isdir('./catboost_info'):
# shutil.rmtree('./catboost_info', ignore_errors=True)
trials_count += 1
if (trials_count % 5) == 0 and is_quit_pressed():
raise co.TennisAbortError
args_dct = dict(**space)
params = {
"eval_metric": metric_name,
# 'eval_metric': 'Logloss',
"random_seed": random_state,
"logging_level": "Silent",
}
params.update(args_dct)
if how == "cv":
cv_data = cv(pools.train, params, stratified=True)
scr_val = np.max(cv_data[f"test-{metric_name}-mean"])
elif how == "sklearn":
mdl = CatBoostClassifier(**params)
mdl.fit(pools.train)
pred = mdl.predict_proba(pools.eval)[:, 1]
scr_val = roc_auc_score(pools.eval.y, pred)
elif how == "native":
mdl = CatBoost(params)
mdl.fit(
pools.train,
eval_set=None, # pools.eval if pools.eval else None,
silent=True,
) # eval_set=pools.eval
pred = mdl.predict(pools.eval, prediction_type="Probability")[:, 1]
scr_val = roc_auc_score(pools.eval.get_label(), pred)
else:
raise Exception("bad how arg {}".format(how))
# pred = mdl.predict(data.X_test)
# scr_val = precision_score(data.y_test, pred)
if scr_val > best_score:
if how == "cv":
cco.out("achieved best {} at {}".format(scr_val, params))
else:
cco.out("achieved best {} at {} lrate: {} ntrees: {}".format(
scr_val, mdl.get_params(), mdl.learning_rate_,
mdl.tree_count_))
best_score = scr_val
return {"loss": 1.0 - scr_val, "status": STATUS_OK}
def evaluate_model(self):
validation_scores = catboost.cv(
catboost.Pool(self.X_train,
self.y_train,
cat_features=self.categorical_columns_indices),
self.model.get_params(),
nfold=self.n_fold,
stratified=self.is_stratified,
seed=self.seed,
early_stopping_rounds=self.early_stopping_rounds,
shuffle=self.is_shuffle,
# metrics= 'RMSE',
plot=False)
self.scores = validation_scores
test_scores = validation_scores.iloc[:, 2]
best_metric = test_scores.min()
return best_metric
def hyperopt_objective(params):
model = CatBoostClassifier(
l2_leaf_reg=int(params['l2_leaf_reg']),
learning_rate=params['learning_rate'],
iterations=1000,
eval_metric='F1',
random_seed=42,
verbose=False,
loss_function='Logloss',
)
cv_data = cv(
Pool(data, data_label, cat_features=categorical_features_indices),
model.get_params())
best_f1 = np.max(cv_data['test-F1-mean'])
return 1 - best_f1
def cv_catboost(data):
data = pd.DataFrame(data)
X_tr = data.drop(3, axis=1)
y_tr = data[3]
params = {
"iterations": 200,
"depth": 2,
"loss_function": "RMSE",
"verbose": True
}
cv_dataset = Pool(data=X_tr, label=y_tr)
scores = cv(cv_dataset, params, fold_count=5, plot="True")
return scores
def eval_train(self):
cv_params = self.model.get_params()
cv_params.update({'loss_function': 'Logloss'})
cv_data = cv(
Pool(self.X, self.y),
cv_params,
#plot=True
)
print(
'Best validation accuracy score: {:.2f}±{:.2f} on step {}'.format(
np.max(cv_data['test-Accuracy-mean']),
cv_data['test-Accuracy-std'][np.argmax(
cv_data['test-Accuracy-mean'])],
np.argmax(cv_data['test-Accuracy-mean'])))
print('Precise validation accuracy score: {}'.format(
np.max(cv_data['test-Accuracy-mean'])))
def _model_cross_validation(self):
self.cv_data = cv(Pool(self.X,
self.y,
cat_features=self.categorical_features_indices),
self.model.get_params(),
plot=False)
# Now we have values of our loss functions at each boosting step averaged by 10 folds,
# which should provide us with a more accurate estimation of our model performance:
print(
'Best validation accuracy score: {:.2f}±{:.2f} on step {}'.format(
np.max(self.cv_data['test-Accuracy-mean']),
self.cv_data['test-Accuracy-std'][np.argmax(
self.cv_data['test-Accuracy-mean'])],
np.argmax(self.cv_data['test-Accuracy-mean'])))
print('Precise validation accuracy score: {}'.format(
np.max(self.cv_data['test-Accuracy-mean'])))
def objective(space_params):
#cast integer params from float to int
for param in integer_params:
space_params[param] = int(space_params[param])
#extract nested conditional parameters
if space_params['bootstrap_type']['bootstrap_type'] == 'Bayesian':
bagging_temp = space_params['bootstrap_type'].get(
'bagging_temperature')
space_params['bagging_temperature'] = bagging_temp
if space_params['grow_policy']['grow_policy'] == 'LossGuide':
max_leaves = space_params['grow_policy'].get('max_leaves')
space_params['max_leaves'] = int(max_leaves)
space_params['bootstrap_type'] = space_params['bootstrap_type'][
'bootstrap_type']
space_params['grow_policy'] = space_params['grow_policy'][
'grow_policy']
#random_strength cannot be < 0
space_params['random_strength'] = max(
space_params['random_strength'], 0)
#fold_len_multiplier cannot be < 1
space_params['fold_len_multiplier'] = max(
space_params['fold_len_multiplier'], 1)
#for classification set stratified=True
cv_results = cb.cv(train,
space_params,
fold_count=N_FOLDS,
early_stopping_rounds=25,
stratified=False,
partition_random_seed=42)
best_loss = cv_results['test-MAE-mean'].iloc[
-1] #'test-RMSE-mean' for RMSE
#for classification, comment out the line above and uncomment the line below:
#best_loss = cv_results['test-Logloss-mean'].iloc[-1]
#if necessary, replace 'test-Logloss-mean' with 'test-[your-preferred-metric]-mean'
return {'loss': best_loss, 'status': STATUS_OK}
def CAT(data, label):
pool = cat.Pool(data, label, has_header=False)
params = {
"loss_function": 'MultiClassOneVsAll', "eval_metric": 'MultiClassOneVsAll', "max_depth": 7,
"learning_rate": 0.2, "classes_count": num_class, "task_type": 'CPU', "thread_count": 6, "verbose_eval": False}
before = datetime.datetime.now()
results = cat.cv(pool=pool, params=params, num_boost_round=boost_rounds, fold_count=cv_fold, shuffle=True,
stratified=True,
verbose=False)
after = datetime.datetime.now()
print("CatBoost")
print("najlepsi priemer: " + str(1 - min(results['test-MultiClassOneVsAll-mean'])))
print("index najlepsieho: " + str(results['test-MultiClassOneVsAll-mean'][results['test-MultiClassOneVsAll-mean']
== min(
results['test-MultiClassOneVsAll-mean'])].index[0]))
print("najhorsi priemer: " + str(1 - max(results['test-MultiClassOneVsAll-mean'])))
print("finalny priemer: " + str(1 - results['test-MultiClassOneVsAll-mean'].iloc[-1]))
print("cas: " + str(after - before))
print('\n')
def cross_validation(
clf,
pool: Pool,
metric_name,
fold_count=5,
stratified=True,
early_stopping_rounds=None,
plot=False,
):
"""metric_name: 'AUC', 'Precision', 'Accuracy'... Output as sample:
Best test Precision score: 0.72+-0.25 on step 22
Best test AUC score: 0.57+-0.01 on step 23
Best validation Logloss score: 0.69+-0.00 on step 0
"""
assert isinstance(pool, Pool)
print(f"cross_validation start with metric {metric_name} "
f"fold_count {fold_count} stratified {stratified} "
f"early_stopping {early_stopping_rounds}")
cv_params = clf.get_params()
cv_params.update({"loss_function": "Logloss"})
cv_data = cv(
pool,
cv_params,
fold_count=fold_count,
stratified=stratified,
plot=plot,
early_stopping_rounds=early_stopping_rounds,
)
test_metric_pref = "test-" + metric_name
print("Best test {} score: {:.2f}+-{:.2f} on step {}".format(
metric_name,
np.max(cv_data[test_metric_pref + "-mean"]),
cv_data[test_metric_pref + "-std"][np.argmax(cv_data[test_metric_pref +
"-mean"])],
np.argmax(cv_data[test_metric_pref + "-mean"]),
))
print("Best validation Logloss score: {:.2f}+-{:.2f} on step {}".format(
np.max(cv_data["test-Logloss-mean"]),
cv_data["test-Logloss-std"][np.argmax(cv_data["test-Logloss-mean"])],
np.argmax(cv_data["test-Logloss-mean"]),
))
def make_catboost(conf: dict, df, log_path):
X_new = univariate_feature_selection(df['X'], df['y'], conf['feature_selection']['k_best_features'])
logger.info('{} is running. X shape = {}'.format('->'.join(log_path), X_new.shape))
cv_dataset = Pool(data=X_new,
label=df['y'])
params = {"loss_function": "Logloss",
"early_stopping_rounds": 30,
"verbose": True,
"custom_metric": ["Accuracy", "F1", "Recall", "Precision"],
"eval_metric": 'F1'}
scores = cv(cv_dataset,
params,
fold_count=5,
verbose=False)
logger.info('{} is done. output = {}'.format('->'.join(log_path), scores['test-F1-mean'].max()))
# clf.fit(X_new, y)
return scores['test-F1-mean'].max()
def train_all_save_catboost(self, X, y, categorical_features_indices):
"""train whole data and save the training to be use later in new predictions"""
model = CatBoostClassifier(loss_function='MultiClass',
eval_metric='TotalF1',
random_seed=42,
leaf_estimation_method='Newton')
cv_data = cv(Pool(X, y, cat_features=categorical_features_indices),
model.get_params())
print("precise validation accuracy score:{}".format(np.max(cv_data)))
model.fit(X, y, cat_features=categorical_features_indices)
#feature importance
print(model.get_feature_importance(prettified=True))
# train = Pool(X, y, cat_features=categorical_features_indices)
# feature_importances = model.get_feature_importance(train)
# feature_names = X.columns
# for score, name in sorted(zip(feature_importances, feature_names), reverse=True):
# print('{}: {}'.format(name, score))
model.save_model('catboost_model.dump')
print("Catboost model has been saved!")
def train_test_dataset(X, y, cat_features=None, auto_class_weights=None,
loss_function='Logloss', iterations=2000, metrics=['AUC']):
pool = Pool(data=X,
label=y,
cat_features=cat_features)
params = {
'task_type': 'CPU',
'auto_class_weights': auto_class_weights,
'custom_metric': metrics,
'verbose': False,
'loss_function': loss_function,
'iterations': iterations
}
return cv(
pool,
params,
fold_count=5,
plot=True,
logging_level='Info'
)
def model_cv(self, model, folds):
'''Run model cross-validation'''
cv_params = model.get_params()
print('{}\nStart model crossvalidation proccess'.format(split_line))
cv_data = cv(
self.pool,
cv_params,
fold_count = folds,
#iterations = 800,
verbose = 200,
early_stopping_rounds = 20
)
print('Best validation accuracy score: {:.4f}±{:.4f} on step {}'\
.format(
np.min(cv_data['test-RMSE-mean']),
cv_data['test-RMSE-std'][np.argmax(cv_data['test-RMSE-mean'])],
np.argmax(cv_data['test-RMSE-mean'])
)
)
def cv(self,
params_model=None,
nfold=5,
num_boost_round=10000,
early_stopping_rounds=100,
**kwargs):
# If no params_model is given, take self.params_best_fit
if params_model is None:
params_model = self.params_best_fit
dtrain = self.get_train_set(as_cgb_pool=True)
eval_hist = cgb.cv(params=params_model,
dtrain=dtrain,
nfold=nfold,
verbose_eval=True,
num_boost_round=num_boost_round,
early_stopping_rounds=early_stopping_rounds,
**kwargs)
return eval_hist
def hyperopt_objective(params):
X_train = self.X_train
y_train = self.y_train
categorical_features_indices = self.cat_index
model = CatBoostRegressor(
l2_leaf_reg=int(params['l2_leaf_reg']),
learning_rate=params['learning_rate'],
depth=params['tree_depth'],
#iterations=500,
eval_metric='RMSE',
#use_best_model=True,
random_seed=42,
logging_level='Silent')
cv_data = cv(params=model.get_params(),
pool=Pool(X_train,
y_train,
cat_features=categorical_features_indices))
#print(cv_data)
best_rmse = np.min(cv_data['test-RMSE-mean'])
print('params is', params, 'rmse is ', best_rmse)
return best_rmse # as hyperopt minimises
def hyperopt_objective(params):
model = CatBoostClassifier(
l2_leaf_reg=int(params['l2_leaf_reg']),
#learning_rate=params['learning_rate'],
depth=params['depth'],
iterations=500,
eval_metric='Accuracy',
od_type='Iter',
od_wait=40,
random_seed=42,
logging_level='Silent',
allow_writing_files=False
)
cv_data = cv(
train_pool,
model.get_params()
)
best_accuracy = np.max(cv_data['test-Accuracy-mean'])
print(params, best_accuracy)
return 1 - best_accuracy # as hyperopt minimises
def ctb_crossval(self, params, optim_type):
'''catboost cross validation model
Paramters
---------
params: Hyper parameters in dict type from different optimization methods
optim_type: choose among Optuna, Hyperopt, RandomSearch
Returns
------
Loss, params, n_estimator, run_time'''
# initializing the timer
start = timer()
print('trial using : ', params)
cv_results = cb.cv(self.train_set,
params,
fold_count=N_FOLDS,
num_boost_round=NUM_BOOST_ROUNDS,
early_stopping_rounds=EARLY_STOPPING_ROUNDS,
stratified=True,
partition_random_seed=SEED,
verbose_eval=True,
plot=False)
# store the runtime
run_time = timer() - start
# Extract the best score
best_score = np.max(cv_results['test-F1-mean'])
# Loss must be minimized
loss = 1 - best_score
# Boosting rounds that returned the highest cv score
n_estimators = int(np.argmax(cv_results['test-F1-mean']) + 1)
if loss < self.loss:
self.estimator = n_estimators
self.loss = loss
#print(params)
return loss, params, n_estimators, run_time
def cross_val(self,
nfold=3,
shuffle=True,
stratified=None,
plot=True,
partition_random_seed: int = 14):
"""
:param nfold:
:param shuffle:
:param stratified:
:param plot:
:param partition_random_seed:
:return:
cv results : pandas.core.frame.DataFrame with cross-validation results
columns are: test-error-mean test-error-std train-error-mean train-error-std
"""
from catboost import Pool, cv
import numpy as np
features, labels, cat_cols = self._data_processor.cv_input_fn()
cv_data = Pool(data=features, label=labels, cat_features=cat_cols)
cv_result = cv(cv_data,
self._params,
nfold=nfold,
shuffle=shuffle,
stratified=stratified,
plot=plot,
partition_random_seed=partition_random_seed)
print('Best validation {} score: {:.2f}±{:.2f} on step {}'.format(
self._params['custom_metric'],
np.max(cv_result[f'test-{self._params["custom_metric"]}-mean']),
cv_result[f'test-{self._params["custom_metric"]}-std'][np.argmax(
cv_result[f'test-{self._params["custom_metric"]}-mean'])],
np.argmax(
cv_result[f'test-{self._params["custom_metric"]}-mean'])))
print('Precise validation {} score: {}'.format(
self._params['custom_metric'],
np.max(cv_result[f'test-{self._params["custom_metric"]}-mean'])))
return cv_result
def catboosttrainer(X,y,features,initparam,modelname,modelpath,docpath,cvfold = 5):
print ("searching for optimal iteration count...")
trainpool = cat.Pool(X[features],y)
cvresult = cat.cv(params= initparam, fold_count=cvfold, pool=trainpool,stratified = True)
initparam['iterations'] = (len(cvresult)) - (initparam['od_wait']+1)
del initparam['od_wait']
del initparam['od_type']
print ("optimal iteration count is ", initparam['iterations'])
print ("fitting model...")
clf = cat.CatBoostClassifier(** initparam)
clf.fit(trainpool)
imp = clf.get_feature_importance(trainpool,fstr_type='FeatureImportance')
dfimp = pd.DataFrame(imp,columns = ['CatBoostImportance'])
dfimp.insert(0,column='Feature', value=features)
dfimp = dfimp.sort_values(['CatBoostImportance','Feature'], ascending= False)
xlsxpath = os.path.join(docpath,modelname+".xlsx")
dfimp.to_excel(xlsxpath)
print ("pickling model...")
picklepath = os.path.join(modelpath,modelname)
with open(picklepath,'wb') as fout:
pickle.dump(clf, fout)
return cvresult,clf,initparam,dfimp
def objetive(trial):
params.update({
"boosting_type":
trial.suggest_categorical("boosting_type",
['Ordered', 'Plain']),
"learning_rate":
trial.suggest_loguniform("learning_rate", 0.005, 0.1),
"max_depth":
trial.suggest_int("max_depth", 4, 12),
"l2_leaf_reg":
trial.suggest_loguniform("l2_leaf_reg", 1e-4, 1e4),
"border_count":
trial.suggest_int('border_count', 1, 255),
"random_strength":
trial.suggest_loguniform("random_strength", 1e-4, 1e4),
"bagging_temperature":
trial.suggest_loguniform("bagging_temperature", 1e-4, 1e4),
})
cv_results = catboost.cv(params=params,
pool=d_train,
iterations=10000,
early_stopping_rounds=50,
folds=folds,
verbose_eval=None,
as_pandas=False)
rmetric_name = list(cv_results.keys())[1]
score = cv_results[rmetric_name][
-1] # np.min(cv_results[rmetric_name])
print("Num_boost_round: " + str(len(cv_results[rmetric_name])))
if save_study_as is not None:
joblib.dump(study, save_study_as)
return score
def cgb_fit(config, X_train, y_train):
"""模型（交叉验证）训练，并返回最优迭代次数和最优的结果。
Args:
config: xgb 模型参数 {params, max_round, cv_folds, early_stop_round, seed, save_model_path}
X_train：array like, shape = n_sample * n_feature
y_train: shape = n_sample * 1
Returns:
best_model: 训练好的最优模型
best_auc: float, 在测试集上面的 AUC 值。
best_round: int, 最优迭代次数。
"""
params = config.params
max_round = config.max_round
cv_folds = config.cv_folds
seed = config.seed
save_model_path = config.save_model_path
if cv_folds is not None:
dtrain = cgb.Pool(X_train, label=y_train)
cv_result = cgb.cv(dtrain, params, num_boost_round=max_round, nfold=cv_folds, seed=seed, logging_level='Verbose')
# 最优模型，最优迭代次数
auc_test_avg = cv_result['AUC_test_avg']
best_round = np.argmax(auc_test_avg)
best_auc = np.max(auc_test_avg) # 最好的 auc 值
best_model = cgb.train(dtrain, params, num_boost_round=best_round)
else:
X_train, X_valid, y_train, y_valid = train_test_split(X_train, y_train, test_size=0.2, random_state=100)
dtrain = cgb.Pool(X_train, label=y_train)
dvalid = cgb.Pool(X_valid, label=y_valid)
best_model = cgb.train(params, dtrain, num_boost_round=max_round, eval_set=dvalid)
best_round = best_model.best_iteration
best_auc = best_model.best_score
cv_result = None
if save_model_path:
check_path(save_model_path)
pickle.dump(best_model, open(save_model_path, 'wb'))
return best_model, best_auc, best_round, cv_result
def cbfunc(border_count, l2_leaf_reg, depth, learning_rate):
params = {
'eval_metric': 'MAE', # using MAE here, could also be RMSE or MSE
'early_stopping_rounds': esrounds,
'num_boost_round': brounds,
'use_best_model': True,
'task_type': "GPU"
}
params['border_count'] = round(border_count, 0)
params['l2_leaf_reg'] = l2_leaf_reg
params['depth'] = round(depth, 0)
params['learning_rate'] = learning_rate
# Cross validation
cv_results = cb.cv(cb.Pool(xtrain, ytrain, cat_features=cat_features),
params=params,
fold_count=3,
inverted=False,
partition_random_seed=5,
shuffle=True,
logging_level='Silent')
# bayes_opt MAXIMISES: In order to minimise MAE, I use 1/MAE as target value
return 1 / cv_results['test-MAE-mean'].min()
def do_cv(learning_rate, depth, l2_leaf_reg):
param = {
'iterations': 3000,
'od_type': 'Iter',
'od_wait': 50,
'learning_rate': learning_rate,
'depth': depth,
'l2_leaf_reg': l2_leaf_reg,
'loss_function': 'MAE',
'eval_metric': 'MAE',
}
eval_hist = catboost.cv(param, catboost.Pool(train_x, train_y,
cat_inds), 5)
res_list = [
eval_hist['MAE_test_avg'][-1],
eval_hist['MAE_test_stddev'][-1],
len(eval_hist['MAE_test_avg']),
param['learning_rate'],
param['depth'],
param['l2_leaf_reg'],
]
line = '%.7f,%.7f,%.0f,%.6f,%.0f,%.0f' % tuple(res_list)
write_to_file(line)
return res_list
def cv(self, data, clf=None):
setseed(self.seed)
if clf:
train_X, train_y = \
data[self.features_name].values,\
data['validRevenue'].values
else:
train_X, train_y = \
data[self.features_name].values,\
data['totals_transactionRevenue'].values
cat_train = cat.Pool(data=train_X,
label=train_y,
feature_names=self.features_name,
cat_features=self.categorical_feature)
cat_cv_hist = cat.cv(
pool=cat_train,
params=self.params,
# num_boost_round = self.num_boost_round,
nfold=self.nfold,
seed=self.seed)
return cat_cv_hist
def test_verbose_int():
pool = Pool(TRAIN_FILE, column_description=CD_FILE)
tmpfile = 'test_data_dumps'
with LogStdout(open(tmpfile, 'w')):
cv(pool, {"iterations": 10, "random_seed": 0, "loss_function": "Logloss"}, verbose=5)
with open(tmpfile, 'r') as output:
assert(sum(1 for line in output) == 2)
with LogStdout(open(tmpfile, 'w')):
cv(pool, {"iterations": 10, "random_seed": 0, "loss_function": "Logloss"}, verbose=False)
with open(tmpfile, 'r') as output:
assert(sum(1 for line in output) == 0)
with LogStdout(open(tmpfile, 'w')):
cv(pool, {"iterations": 10, "random_seed": 0, "loss_function": "Logloss"}, verbose=True)
with open(tmpfile, 'r') as output:
assert(sum(1 for line in output) == 10)
log_files = []
for i in range(3):
log_files.append(JSON_LOG_PATH[:-5]+str(i)+JSON_LOG_PATH[-5:])
with LogStdout(open(tmpfile, 'w')):
train(pool, {"iterations": 10, "random_seed": 0, "loss_function": "Logloss", "json_log": log_files[0]}, verbose=5)
with open(tmpfile, 'r') as output:
assert(sum(1 for line in output) == 2)
with LogStdout(open(tmpfile, 'w')):
train(pool, {"iterations": 10, "random_seed": 0, "loss_function": "Logloss", "json_log": log_files[1]}, verbose=False)
with open(tmpfile, 'r') as output:
assert(sum(1 for line in output) == 0)
with LogStdout(open(tmpfile, 'w')):
train(pool, {"iterations": 10, "random_seed": 0, "loss_function": "Logloss", "json_log": log_files[2]}, verbose=True)
with open(tmpfile, 'r') as output:
assert(sum(1 for line in output) == 10)
canonical_files = []
for log_file in log_files:
canonical_files.append(local_canonical_file(remove_time_from_json(log_file)))
return canonical_files
y_pred = model.predict(test_pool)
# 模型评估
print(f'R2: {r2_score(y_test, y_pred)}')
# 特征重要性
print(f'feature_importance = {model.get_feature_importance(train_pool)}')
# 保存模型
model.save_model('regression.cbm', format="cbm")
# Calculate the RMSE metric for the objects in the given dataset.
print(f'score: {model.score(X_train, y_train)}')
# cv
args = config.pop('calc_feature_importance', None)
print(config)
config1 = {
'pool': train_pool,
'params': config,
'iterations': 1000,
'fold_count': 10,
'partition_random_seed': 120,
'logging_level': 'Verbose',
'stratified': False,
'as_pandas': True
}
scores = cv(**config1)
print(f'CV result is: {scores}')
X_train = train[features].fillna('')
y_train = train['Survived']
X_test = test[features].fillna('')
model = catboost.CatBoostClassifier(one_hot_max_size=4,
iterations=100,
random_seed=0,
verbose=False,
eval_metric='Accuracy')
pool = catboost.Pool(X_train, y_train, cat_features=[0, 2])
print(
'To see the Catboost plots, fork this kernel and run it in the editing mode.'
)
cv_scores = catboost.cv(pool, model.get_params(), fold_count=10, plot=True)
print('CV score: {:.5f}'.format(cv_scores['test-Accuracy-mean'].values[-1]))
# You can check yourself the public LB score of this model (0.77990) by submitting the file `submission2.csv` from the section Output of this kernel.
# In[ ]:
model.fit(pool)
pred = model.predict(X_test).astype('int')
output = pd.concat(
[test['PassengerId'],
pd.DataFrame(pred, columns=['Survived'])], axis=1)
output.to_csv('submission2.csv', index=False)
# Next I consider the generated feature `Boy`. It takes the value 1 if the title in `Name` is "Master" and the value 0 otherwise.
#
def test_cv_with_not_binarized_target():
train_file = data_file('adult_not_binarized', 'train_small')
cd = data_file('adult_not_binarized', 'train.cd')
pool = Pool(train_file, column_description=cd)
cv(pool, {"iterations": 5, "random_seed": 0, "loss_function": "Logloss"})
return local_canonical_file(remove_time_from_json(JSON_LOG_PATH))
def test_bad_params_in_cv():
pool = Pool(TRAIN_FILE, column_description=CD_FILE)
with pytest.warns(UserWarning):
cv({"iterations": 5, "random_seed": 0, "loss_function": "Logloss", "use_best_model": True}, pool)
def test_cv_with_not_binarized_target():
train_file = data_file('adult_not_binarized', 'train_small')
cd = data_file('adult_not_binarized', 'train.cd')
pool = Pool(train_file, column_description=cd)
cv(pool, {"iterations": 5, "random_seed": 0, "loss_function": "Logloss"})
return local_canonical_file(remove_time_from_json(JSON_LOG_PATH))
def test_cv_logging():
pool = Pool(TRAIN_FILE, column_description=CD_FILE)
cv(pool, {"iterations": 5, "random_seed": 0, "loss_function": "Logloss"})
return local_canonical_file(remove_time_from_json(JSON_LOG_PATH))
