王半仙的学习分享

pip install tsfresh # 安装

{'x':DataFrame(x),'y':DataFrame(y)}

def extract_features(timeseries_container, default_fc_parameters=None,
                     kind_to_fc_parameters=None,
                     column_id=None, column_sort=None, column_kind=None, column_value=None,
                     chunksize=defaults.CHUNKSIZE,
                     n_jobs=defaults.N_PROCESSES, show_warnings=defaults.SHOW_WARNINGS,
                     disable_progressbar=defaults.DISABLE_PROGRESSBAR,
                     impute_function=defaults.IMPUTE_FUNCTION,
                     profile=defaults.PROFILING,
                     profiling_filename=defaults.PROFILING_FILENAME,
                     profiling_sorting=defaults.PROFILING_SORTING,
                     distributor=None, pivot=True):
"""


Examples

========

:param timeseries_container: 输入数据，数据框或数据框组成的字典
:param default_fc_parameters: 用于全局配置（对不同类型数据都计算）所需计算的特征，dict
:param kind_to_fc_parameters: 用于单独配置每一类数据所需计算的特征，dict
:param column_id/column_sort/column_kind/column_value: str
:param n_jobs: 并行数，默认为可用CPU数的一半. 设置为0则表示不进行并行计算，int
:param chunksize: 设置每个任务计算chunksize个时序的所有特征，默认为自适应设定，None or int
:param show_warnings: 是否显示特征计算过程中的警告，默认为False，bool
:param disable_progressbar: 是否不显示特征计算过程中的进度条，默认为False，bool

:return: The (maybe imputed) DataFrame containing extracted features.

"""

fc_parameters = {
    "length": None,
    "large_standard_deviation": [{"r": 0.05}, {"r": 0.1}]
}

kind_to_fc_parameters = {
    "temperature": {"mean": None},
    "pressure": {"maximum": None, "minimum": None}
}

def calculate_relevance_table(X, y, ml_task='auto', n_jobs=1, show_warnings=False, chunksize=None, test_for_binary_target_binary_feature='fisher', test_for_binary_target_real_feature='mann', test_for_real_target_binary_feature='mann', test_for_real_target_real_feature='kendall', fdr_level=0.05, hypotheses_independent=False):
    """
    针对特定模型类型，借助假设检验与p值评估特征和预测目标间的关联性
    :param X: 包含所有特征的pandas.DataFrame
    :param y: 包含预测目标值的pandas.Series或numpy.ndarray
    :param ml_task: 建模类型，主要包括'classification', 'regression' or 'auto（默认）'，str
    :param test_for_binary_target_binary_feature / test_for_binary_target_real_featur / test_for_real_target_binary_feature / test_for_real_target_real_feature: 指定不同数据类型的假设检验方法（fisher检验、），str
    :param fdr_level: 预设假阳率（错误发现率）水平
    :param hypotheses_independent: 特征间是否是独立的，基本上都是False，bool
    :param n_jobs: 假设检验计算的进程数，int
    :param show_warnings: 是否显示警告信息，bool

    :return: 类型为pandas.DataFrame，主要包括特征、特征类型、假设检验p值、相关性等信息
    """

from tsfresh.feature_extraction import feature_calculators


@feature_calculators.set_property("fctype", "simple")
def mode(x):
    """Return the mode of the parameter (i.e. most common value)

    :param x: the time series to calculate the feature of
    :type x: np.ndarray
    :return: the different feature values
    :return type: tuple
    """
    c = Counter(x)
    return tuple(x for x, count in c.items() if count == c.most_common(1)[0][1])

@feature_calculators.set_property("fctype", "combiner")
def value_count_all(x, param):
    """
    Returns the number of values in x 

    :param x: the time series on which to calculate the feature.
    :type x: pandas.Series
    :param param: None
    :return: the value of this feature
    :return type: list
    """
    values, counts = np.unique(x, return_counts=True)

    return [("value_count__value_\"{}\"".format(value), \
    		 feature_calculators.value_count(x, value)) for value in values]

@feature_calculators.set_property("input", "pd.Series") # 输入格式为pd.Series（默认是numpy）
@feature_calculators.set_property("fctype", "combiner")
@feature_calculators.set_property("minimal", True)
def last(x, param):
    """Return the last index and value of x.

    :param x: the time series on which to calculate the feature.
    :type x: pandas.Series
    :return: the value of this feature
    :return type: list
    """
    return [("value",x.values[-1]), ("index",x.index[-1])]

from tsfresh.feature_extraction import feature_calculators
custom_functions = [mode, value_count_all, last]
for func in custom_functions:
    setattr(feature_calculators, func.__name__, func)