catboost的CatBoostRegressor函数源代码简介、解读
class CatBoostRegressor Found at: catboost.core
class CatBoostRegressor(CatBoost):
_estimator_type = 'regressor'
"""
Implementation of the scikit-learn API for CatBoost regression.
Parameters
----------
Like in CatBoostClassifier, except loss_function, classes_count, class_names and class_weights
loss_function : string, [default='RMSE']
'RMSE'
'MAE'
'Quantile:alpha=value'
'LogLinQuantile:alpha=value'
'Poisson'
'MAPE'
'Lq:q=value'
"""
实现scikit-learn API的CatBoost回归。
参数
----------
像CatBoostClassifier，除了loss_function, classes_count, class_names和class_weights
def __init__(
self,
iterations=None,
learning_rate=None,
depth=None,
l2_leaf_reg=None,
model_size_reg=None,
rsm=None,
loss_function='RMSE',
border_count=None,
feature_border_type=None,
per_float_feature_quantization=None,
input_borders=None,
output_borders=None,
fold_permutation_block=None,
od_pval=None,
od_wait=None,
od_type=None,
nan_mode=None,
counter_calc_method=None,
leaf_estimation_iterations=None,
leaf_estimation_method=None,
thread_count=None,
random_seed=None,
use_best_model=None,
best_model_min_trees=None,
verbose=None,
silent=None,
logging_level=None,
metric_period=None,
ctr_leaf_count_limit=None,
store_all_simple_ctr=None,
max_ctr_complexity=None,
has_time=None,
allow_const_label=None,
target_border=None,
one_hot_max_size=None,
random_strength=None,
name=None,
ignored_features=None,
train_dir=None,
custom_metric=None,
eval_metric=None,
bagging_temperature=None,
save_snapshot=None,
snapshot_file=None,
snapshot_interval=None,
fold_len_multiplier=None,
used_ram_limit=None,
gpu_ram_part=None,
pinned_memory_size=None,
allow_writing_files=None,
final_ctr_computation_mode=None,
approx_on_full_history=None,
boosting_type=None,
simple_ctr=None,
combinations_ctr=None,
per_feature_ctr=None,
ctr_description=None,
ctr_target_border_count=None,
task_type=None,
device_config=None,
devices=None,
bootstrap_type=None,
subsample=None,
mvs_reg=None,
sampling_frequency=None,
sampling_unit=None,
dev_score_calc_obj_block_size=None,
dev_efb_max_buckets=None,
sparse_features_conflict_fraction=None,
max_depth=None,
n_estimators=None,
num_boost_round=None,
num_trees=None,
colsample_bylevel=None,
random_state=None,
reg_lambda=None,
objective=None,
eta=None,
max_bin=None,
gpu_cat_features_storage=None,
data_partition=None,
metadata=None,
early_stopping_rounds=None,
cat_features=None,
grow_policy=None,
min_data_in_leaf=None,
min_child_samples=None,
max_leaves=None,
num_leaves=None,
score_function=None,
leaf_estimation_backtracking=None,
ctr_history_unit=None,
monotone_constraints=None,
feature_weights=None,
penalties_coefficient=None,
first_feature_use_penalties=None,
per_object_feature_penalties=None,
model_shrink_rate=None,
model_shrink_mode=None,
langevin=None,
diffusion_temperature=None,
posterior_sampling=None,
boost_from_average=None):
params = {}
not_params = ["not_params", "self", "params", "__class__"]
for key, value in iteritems(locals().copy()):
if key not in not_params and value is not None:
params[key] = value
super(CatBoostRegressor, self).__init__(params)
def fit(self, X, y=None, cat_features=None, sample_weight=None, baseline=None,
use_best_model=None,
eval_set=None, verbose=None, logging_level=None, plot=False,
column_description=None,
verbose_eval=None, metric_period=None, silent=None, early_stopping_rounds=None,
save_snapshot=None, snapshot_file=None, snapshot_interval=None, init_model=None):
"""
Fit the CatBoost model.
Parameters
----------
X : catboost.Pool or list or numpy.ndarray or pandas.DataFrame or pandas.Series. If not catboost.Pool, 2 dimensional Feature matrix or string - file with dataset.
y : list or numpy.ndarray or pandas.DataFrame or pandas.Series, optional (default=None). Labels, 1 dimensional array like. Use only if X is not catboost.Pool.
cat_features : list or numpy.ndarray, optional (default=None). If not None, giving the list of Categ columns indices.Use only if X is not catboost.Pool.
sample_weight : list or numpy.ndarray or pandas.DataFrame or pandas.Series, optional (default=None). Instance weights, 1 dimensional array like.
baseline : list or numpy.ndarray, optional (default=None). If not None, giving 2 dimensional array like data. Use only if X is not catboost.Pool.
use_best_model : bool, optional (default=None). Flag to use best model
eval_set : catboost.Pool or list, optional (default=None). A list of (X, y) tuple pairs to use as a validation set for early-stopping
metric_period : int. Frequency of evaluating metrics.
verbose : bool or int. If verbose is bool, then if set to True, logging_level is set to Verbose, if set to False, logging_level is set to Silent. If verbose is int, it determines the frequency of writing metrics to output and logging_level is set to Verbose.
silent : bool. If silent is True, logging_level is set to Silent. If silent is False, logging_level is set to Verbose.
logging_level : string, optional (default=None). Possible values:
- 'Silent'
- 'Verbose'
- 'Info'
- 'Debug'
plot : bool, optional (default=False). If True, draw train and eval error in Jupyter notebook
verbose_eval : bool or int. Synonym for verbose. Only one of these parameters should be set.
early_stopping_rounds : int. Activates Iter overfitting detector with od_wait set to early_stopping_rounds.
save_snapshot : bool, [default=None]. Enable progress snapshotting for restoring progress after crashes or interruptions
snapshot_file : string, [default=None]. Learn progress snapshot file path, if None will use default filename snapshot_interval: int, [default=600]. Interval between saving snapshots (seconds)
init_model : CatBoost class or string, [default=None]. Continue training starting from the existing model. If this parameter is a string, load initial model from the path specified by this string.
Returns
-------
model : CatBoost
"""
params = deepcopy(self._init_params)
_process_synonyms(params)
if 'loss_function' in params:
X: catboost。pool或list或numpy。ndarray或pandas.DataFrame或pandas.Series。如果不是catboost。Pool，二维特征矩阵或字符串文件与数据集。
y: list或numpy。ndarray或pandas.DataFrame或pandas.Series。可选(默认= None)。标签，类似于一维数组。仅当X不是catboost.Pool时使用。
cat_features: list或numpy.ndarray,可选(默认= None)。如果不是None，则给出类别列索引的列表。仅当X不是catboost.Pool时使用。
sample_weight:列表或numpy。ndarray或pandas.DataFrame或pandas.Series,可选(默认= None)。实例权重，类似于一维数组。
baseline:列表或numpy。ndarray,可选(默认= None)。如果不是None，则给出像data这样的二维数组。仅当X不是catboost.Pool时使用。
use_best_model: bool，可选(默认为None)。标记使用最佳模型
eval_set: catboost。Pool或列表，可选(默认为None)。(X, y)元组对的列表，用作早期停止的验证集。
metric_period: int。评估指标的频率。
verbose: bool或int。如果verbose是bool，那么如果设置为True, logging_level将设置为verbose，如果设置为False, logging_level将设置为Silent。如果verbose为int，则它确定向输出写入指标的频率，并将logging_level设置为verbose。
silent : bool。如果silent为True, loging_level设置为silent。如果silent为False, loging_level设置为Verbose。
logging_level:字符串，可选(默认为None)。可能的值:
——“沉默”
——“详细”
——“信息”
——“调试”
plot: bool，可选(默认=False)。如果为真，在Jupyter中绘制训练集和测试集的error
verbose_eval: bool或int。详细的同义词。应该只设置这些参数中的一个。
early_stopping_rounds: int。激活Iter过拟合检测器，od_wait设置为early_stopping_rounds。
save_snapshot: bool， [default=None]。启用进度快照，以便在崩溃或中断后恢复进度
snapshot_file: string， [default=None]。学习进度快照文件路径，如果没有将使用默认文件名snapshot_interval: int，[默认=600]。保存快照的时间间隔(秒)
init_model: CatBoost类或字符串，[default=None]。从现有的模式开始继续培训。如果该参数为字符串，则从该字符串指定的路径加载初始模型。
self._check_is_regressor_loss(params['loss_function'])
return self._fit(X, y, cat_features, None, None, None, sample_weight, None, None, None,
None, baseline,
use_best_model, eval_set, verbose, logging_level, plot, column_description,
verbose_eval, metric_period, silent, early_stopping_rounds,
save_snapshot, snapshot_file, snapshot_interval, init_model)
def predict(self, data, prediction_type=None, ntree_start=0, ntree_end=0, thread_count=-
1, verbose=None):
"""
Predict with data.
Parameters
----------
data : catboost.Pool or list of features or list of lists or numpy.ndarray or pandas. DataFrame or pandas.Series or catboost.FeaturesData. Data to apply model on. If data is a simple list (not list of lists) or a one-dimensional numpy.ndarray it is interpreted as a list of features for a single object.
prediction_type : string, optional (default='RawFormulaVal'). Can be:
- 'RawFormulaVal' : return raw formula value.
- 'Exponent' : return Exponent of raw formula value.
ntree_start: int, optional (default=0)
Model is applied on the interval [ntree_start, ntree_end) (zero-based indexing).
ntree_end: int, optional (default=0)
Model is applied on the interval [ntree_start, ntree_end) (zero-based indexing). If value equals to 0 this parameter is ignored and ntree_end equal to tree_count_.
thread_count : int (default=-1). The number of threads to use when applying the model. Allows you to optimize the speed of execution. This parameter doesn't affect results. If -1, then the number of threads is set to the number of CPU cores.
verbose : bool. If True, writes the evaluation metric measured set to stderr.
Returns
-------
prediction : If data is for a single object, the return value is single float formula return value otherwise one-dimensional numpy.ndarray of formula return values for each object.
"""
if prediction_type is None:
prediction_type = self._get_default_prediction_type()
return self._predict(data, prediction_type, ntree_start, ntree_end, thread_count, verbose,
'predict')
参数
---------
data : catboost。池或特性列表或列表的列表或numpy。ndarray或熊猫。DataFrame或熊猫。系列或catboost.FeaturesData。应用模型的数据。如果data是一个简单的列表(不是列表的列表)或一维numpy。ndarray它被解释为一个对象的特性列表。
prediction_type :字符串，可选(默认为'RawFormulaVal')。可以是:
- 'RawFormulaVal':返回原始公式值。
- 'Exponent':返回原始公式值的指数。
ntree_start: int，可选(默认为0)
模型应用于区间[ntree_start, ntree_end)(从零开始索引)。
ntree_end: int，可选(默认为0)
模型应用于区间[ntree_start, ntree_end)(从零开始索引)。如果value等于0，则忽略该参数，ntree_end等于tree_count_。
thread_count :int(默认=-1)。应用模型时要使用的线程数。允许您优化执行速度。此参数不影响结果。如果-1，则线程数设置为CPU核数。
verbose :bool。如果为真，则将评估度量值写入stderr。
返回
-------
prediction：如果数据是针对单个对象的，则返回值为单个float公式返回值，否则为一维numpy。ndarray的公式返回每个对象的值。
def staged_predict(self, data, prediction_type='RawFormulaVal', ntree_start=0,
ntree_end=0, eval_period=1, thread_count=-1, verbose=None):
"""
Predict target at each stage for data.
Parameters
----------
data : catboost.Pool or list of features or list of lists or numpy.ndarray or pandas. DataFrame or pandas.Series or catboost.FeaturesData. Data to apply model on. If data is a simple list (not list of lists) or a one-dimensional numpy.ndarray it is interpreted as a list of features for a single object.
ntree_start: int, optional (default=0). Model is applied on the interval [ntree_start, ntree_end) with the step eval_period (zero-based indexing).
ntree_end: int, optional (default=0).Model is applied on the interval [ntree_start, ntree_end) with the step eval_period (zero-based indexing). If value equals to 0 this parameter is ignored and ntree_end equal to tree_count_.
eval_period: int, optional (default=1). Model is applied on the interval [ntree_start, ntree_end) with the step eval_period (zero-based indexing).
thread_count : int (default=-1). The number of threads to use when applying the model. Allows you to optimize the speed of execution. This parameter doesn't affect results. If -1, then the number of threads is set to the number of CPU cores.
verbose : bool. If True, writes the evaluation metric measured set to stderr.
Returns
-------
prediction : generator for each iteration that generates:If data is for a single object, the return value is single float formula return value otherwise one-dimensional numpy.ndarray of formula return values for each object.
"""
return self._staged_predict(data, prediction_type, ntree_start, ntree_end, eval_period,
thread_count, verbose, 'staged_predict')
data : catboost。池或特性列表或列表的列表或numpy。ndarray或DataFrame 或pandas.Series or catboost.FeaturesData。应用模型的数据。如果data是一个简单的列表(不是列表的列表)或一维numpy。ndarray它被解释为一个对象的特性列表。
ntree_start: int，可选(默认为0)。模型应用于间隔[ntree_start, ntree_end)，步长为eval_period(从零开始索引)。
ntree_end:int，可选(默认为0)。模型应用于间隔[ntree_start, ntree_end)，步长为eval_period(从零开始索引)。如果value等于0，则忽略该参数，ntree_end等于tree_count_。
eval_period: int，可选(默认为1)。模型应用于间隔[ntree_start, ntree_end)，步长为eval_period(从零开始索引)。
thread_count : int(默认=-1)。应用模型时要使用的线程数。允许您优化执行速度。此参数不影响结果。如果-1，则线程数设置为CPU核数。
verbose :bool。如果为真，则将评估度量值写入stderr。
返回
-------
prediction :为每个迭代生成的生成器:如果数据是针对单个对象的，则返回值为单个float公式返回值，否则为一维numpy。ndarray的公式返回每个对象的值。
def score(self, X, y=None):
"""
Calculate R^2.
Parameters
----------
X : catboost.Pool or list or numpy.ndarray or pandas.DataFrame or pandas.Series.Data to apply model on.
y : list or numpy.ndarray.True labels.
Returns
-------
R^2 : float
"""
if isinstance(X, Pool):
if y is not None:
raise CatBoostError("Wrong initializing y: X is catboost.Pool object, y must be
initialized inside catboost.Pool.")
y = X.get_label()
if y is None:
raise CatBoostError("Label in X has not initialized.")
elif y is None:
raise CatBoostError("y should be specified.")
y = np.array(y, dtype=np.float64)
predictions = self._predict(X,
prediction_type=self._get_default_prediction_type(),
ntree_start=0,
ntree_end=0,
thread_count=-1,
verbose=None,
parent_method_name='score')
loss = self._object._get_loss_function_name()
if loss == 'RMSEWithUncertainty':
predictions = predictions[:0]
total_sum_of_squares = np.sum((y - y.mean(axis=0)) ** 2)
residual_sum_of_squares = np.sum((y - predictions) ** 2)
return 1 - residual_sum_of_squares / total_sum_of_squares
def _check_is_regressor_loss(self, loss_function):
is_regression = self._is_regression_objective(loss_function) or self.
_is_multiregression_objective(loss_function)
if isinstance(loss_function, str) and not is_regression:
raise CatBoostError("Invalid loss_function='{}': for regressor use "
"RMSE, MultiRMSE, MAE, Quantile, LogLinQuantile, Poisson, MAPE, Lq or custom
objective object".format(loss_function))
def _get_default_prediction_type(self):
# TODO(ilyzhin) change on get_all_params after MLTOOLS-4758
params = deepcopy(self._init_params)
_process_synonyms(params)
loss_function = params.get('loss_function')
if loss_function and isinstance(loss_function, str):
if loss_function.startswith('Poisson') or loss_function.startswith('Tweedie'):
return 'Exponent'
if loss_function == 'RMSEWithUncertainty':
return 'RMSEWithUncertainty'
return 'RawFormulaVal'
