Qubx 0.1.83__cp311-cp311-manylinux_2_35_x86_64.whl → 0.1.85__cp311-cp311-manylinux_2_35_x86_64.whl

qubx/__init__.py

@@ -10,13 +10,20 @@ def formatter(record):
     end = record["extra"].get("end", "\n")
     fmt = "<lvl>{message}</lvl>%s" % end
     if record["level"].name in {"WARNING", "SNAKY"}:
-        fmt = "<cyan>{name}</cyan>:<cyan>{function}</cyan>:<cyan>{line}</cyan> - %s" % fmt
+        fmt = (
+            "<cyan>{name}</cyan>:<cyan>{function}</cyan>:<cyan>{line}</cyan> - %s" % fmt
+        )
 
-    prefix = "<green>{time:YYYY-MM-DD HH:mm:ss.SSS}</green> [ <level>%s</level> ] " % record["level"].icon
+    prefix = (
+        "<green>{time:YYYY-MM-DD HH:mm:ss.SSS}</green> [ <level>%s</level> ] "
+        % record["level"].icon
+    )
 
     if record["exception"] is not None:
         # stackprinter.set_excepthook(style='darkbg2')
-        record["extra"]["stack"] = stackprinter.format(record["exception"], style="darkbg")
+        record["extra"]["stack"] = stackprinter.format(
+            record["exception"], style="darkbg"
+        )
         fmt += "\n{extra[stack]}\n"
 
     if record["level"].name in {"TEXT"}:
@@ -25,24 +32,43 @@ def formatter(record):
     return prefix + fmt
 
 
-config = {
-    "handlers": [ {"sink": sys.stdout, "format": "{time} - {message}"}, ],
-    "extra": {"user": "someone"},
-}
+class QubxLogConfig:
 
+    @staticmethod
+    def get_log_level():
+        return os.getenv("QUBX_LOG_LEVEL", "DEBUG")
+
+    @staticmethod
+    def set_log_level(level: str):
+        os.environ["QUBX_LOG_LEVEL"] = level
+        QubxLogConfig.setup_logger(level)
+
+    @staticmethod
+    def setup_logger(level: str | None = None):
+        global logger
+        config = {
+            "handlers": [
+                {"sink": sys.stdout, "format": "{time} - {message}"},
+            ],
+            "extra": {"user": "someone"},
+        }
+        logger.configure(**config)
+        logger.remove(None)
+        level = level or QubxLogConfig.get_log_level()
+        logger.add(sys.stdout, format=formatter, colorize=True, level=level)
+        logger = logger.opt(colors=True)
+
+
+QubxLogConfig.setup_logger()
 
-logger.configure(**config)
-logger.remove(None)
-logger.add(sys.stdout, format=formatter, colorize=True)
-logger = logger.opt(colors=True)
 
 # - global lookup helper
 lookup = GlobalLookup(InstrumentsLookup(), FeesLookup())
 
 
 # registering magic for jupyter notebook
-if runtime_env() in ['notebook', 'shell']:
-    from IPython.core.magic import (Magics, magics_class, line_magic, line_cell_magic)
+if runtime_env() in ["notebook", "shell"]:
+    from IPython.core.magic import Magics, magics_class, line_magic, line_cell_magic
     from IPython import get_ipython
 
     @magics_class
@@ -52,11 +78,11 @@ if runtime_env() in ['notebook', 'shell']:
 
         @line_magic
         def qubxd(self, line: str):
-            self.qubx_setup('dark' + ' ' + line)
+            self.qubx_setup("dark" + " " + line)
 
         @line_magic
         def qubxl(self, line: str):
-            self.qubx_setup('light' + ' ' + line)
+            self.qubx_setup("light" + " " + line)
 
         @line_magic
         def qubx_setup(self, line: str):
@@ -64,25 +90,26 @@ if runtime_env() in ['notebook', 'shell']:
             QUBX framework initialization
             """
             import os
-            args = [x.strip() for x in line.split(' ')]
-            
+
+            args = [x.strip() for x in line.split(" ")]
+
             # setup cython dev hooks - only if 'dev' is passed as argument
-            if line and 'dev' in args: 
+            if line and "dev" in args:
                 install_pyx_recompiler_for_dev()
 
             tpl_path = os.path.join(os.path.dirname(__file__), "_nb_magic.py")
-            with open(tpl_path, 'r', encoding="utf8") as myfile:
+            with open(tpl_path, "r", encoding="utf8") as myfile:
                 s = myfile.read()
 
             exec(s, self.shell.user_ns)
 
             # setup more funcy mpl theme instead of ugly default
             if line:
-                if 'dark' in line.lower():
-                    set_mpl_theme('dark')
+                if "dark" in line.lower():
+                    set_mpl_theme("dark")
 
-                elif 'light' in line.lower():
-                    set_mpl_theme('light')
+                elif "light" in line.lower():
+                    set_mpl_theme("light")
 
             # install additional plotly helpers
             # from qube.charting.plot_helpers import install_plotly_helpers
@@ -91,6 +118,7 @@ if runtime_env() in ['notebook', 'shell']:
         def _get_manager(self):
             if self.__manager is None:
                 import multiprocessing as m
+
                 self.__manager = m.Manager()
             return self.__manager
 
@@ -102,7 +130,7 @@ if runtime_env() in ['notebook', 'shell']:
             >>> %%proc x, y as MyProc1
             >>> x.set('Hello')
             >>> y.set([1,2,3,4])
-            
+
             """
             import multiprocessing as m
             import time, re
@@ -111,8 +139,8 @@ if runtime_env() in ['notebook', 'shell']:
             name = None
             if line:
                 # check if custom process name was provided
-                if ' as ' in line:
-                    line, name = line.split('as')
+                if " as " in line:
+                    line, name = line.split("as")
                     if not name.isspace():
                         name = name.strip()
                     else:
@@ -120,11 +148,11 @@ if runtime_env() in ['notebook', 'shell']:
                         return
 
                 ipy = get_ipython()
-                for a in [x for x in re.split('[\ ,;]', line.strip()) if x]:
+                for a in [x for x in re.split("[\ ,;]", line.strip()) if x]:
                     ipy.push({a: self._get_manager().Value(None, None)})
 
             # code to run
-            lines = '\n'.join(['    %s' % x for x in cell.split('\n')])
+            lines = "\n".join(["    %s" % x for x in cell.split("\n")])
 
             def fn():
                 result = get_ipython().run_cell(lines)
@@ -136,17 +164,18 @@ if runtime_env() in ['notebook', 'shell']:
                 if result.error_in_exec:
                     raise result.error_in_exec
 
-            t_start = str(time.time()).replace('.', '_')
-            f_id = f'proc_{t_start}' if name is None else name
+            t_start = str(time.time()).replace(".", "_")
+            f_id = f"proc_{t_start}" if name is None else name
             if self._is_task_name_already_used(f_id):
                 f_id = f"{f_id}_{t_start}"
 
             task = m.Process(target=fn, name=f_id)
             task.start()
-            print(' -> Task %s is started' % f_id)
+            print(" -> Task %s is started" % f_id)
 
         def _is_task_name_already_used(self, name):
             import multiprocessing as m
+
             for p in m.active_children():
                 if p.name == name:
                     return True
@@ -155,16 +184,17 @@ if runtime_env() in ['notebook', 'shell']:
         @line_magic
         def list_proc(self, line):
             import multiprocessing as m
+
             for p in m.active_children():
                 print(p.name)
 
         @line_magic
         def kill_proc(self, line):
             import multiprocessing as m
+
             for p in m.active_children():
                 if line and p.name.startswith(line):
                     p.terminate()
 
-
     # - registering magic here
     get_ipython().register_magics(QubxMagics)

qubx/_nb_magic.py

@@ -1,7 +1,6 @@
 """"
 Here stuff we want to have in every Jupyter notebook after calling %qube magic
 """
-import importlib_metadata
 
 import qubx
 from qubx.utils import runtime_env
@@ -15,11 +14,19 @@ def np_fmt_short():
 
 def np_fmt_reset():
     # reset default np printing options
-    np.set_printoptions(edgeitems=3, infstr='inf', linewidth=75, nanstr='nan', precision=8,
-                        suppress=False, threshold=1000, formatter=None)
+    np.set_printoptions(
+        edgeitems=3,
+        infstr="inf",
+        linewidth=75,
+        nanstr="nan",
+        precision=8,
+        suppress=False,
+        threshold=1000,
+        formatter=None,
+    )
 
 
-if runtime_env() in ['notebook', 'shell']:
+if runtime_env() in ["notebook", "shell"]:
 
     # - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
     # -- all imports below will appear in notebook after calling %%alphalab magic ---
@@ -39,19 +46,18 @@ if runtime_env() in ['notebook', 'shell']:
     # - - - - Learn stuff - - - -
     # - - - - Charting stuff - - - -
     from matplotlib import pyplot as plt
-    from qubx.utils.charting.mpl_helpers import fig, subplot, sbp
+    from qubx.utils.charting.mpl_helpers import fig, subplot, sbp, plot_trends, ohlc_plot
 
     # - - - - Utils - - - -
     from qubx.pandaz.utils import scols, srows, ohlc_resample, continuous_periods, generate_equal_date_ranges
 
     # - setup short numpy output format
     np_fmt_short()
-    
+
     # - add project home to system path
     add_project_to_system_path()
 
     # show logo first time
-    if not hasattr(qubx.QubxMagics, '__already_initialized__'):
+    if not hasattr(qubx.QubxMagics, "__already_initialized__"):
         setattr(qubx.QubxMagics, "__already_initialized__", True)
         logo()
-

qubx/core/series.pxd

@@ -15,7 +15,7 @@ cdef class TimeSeries:
     cdef public long long timeframe
     cdef public Indexed times
     cdef public Indexed values
-    cdef float max_series_length
+    cdef public float max_series_length
     cdef unsigned short _is_new_item
     cdef public str name
     cdef dict indicators        # it's used for indicators caching
@@ -28,8 +28,12 @@ cdef class TimeSeries:
 
 
 cdef class Indicator(TimeSeries):
-    cdef TimeSeries series
-    cdef TimeSeries parent
+    cdef public TimeSeries series
+    cdef public TimeSeries parent
+
+
+cdef class IndicatorOHLC(Indicator):
+    pass
 
 
 cdef class RollingSum:

qubx/core/series.pyx

@@ -300,6 +300,9 @@ def _wrap_indicator(series: TimeSeries, clz, *args, **kwargs):
 
 
 cdef class Indicator(TimeSeries):
+    """
+    Basic class for indicator that can be attached to TimeSeries
+    """
 
     def __init__(self, str name, TimeSeries series):
         if not name:
@@ -309,7 +312,7 @@ cdef class Indicator(TimeSeries):
         self.name = name
 
         # - we need to make a empty copy and fill it 
-        self.series = TimeSeries(series.name, series.timeframe, series.max_series_length)
+        self.series = self._instantiate_base_series(series.name, series.timeframe, series.max_series_length)
         self.parent = series 
         
         # - notify the parent series that indicator has been attached
@@ -318,6 +321,9 @@ cdef class Indicator(TimeSeries):
         # - recalculate indicator on data as if it would being streamed
         self._initial_data_recalculate(series)
 
+    def _instantiate_base_series(self, str name, long long timeframe, float max_series_length):
+        return TimeSeries(name, timeframe, max_series_length)
+
     def _on_attach_indicator(self, indicator: Indicator, indicator_input: TimeSeries):
         self.parent._on_attach_indicator(indicator, indicator_input)
 
@@ -345,6 +351,17 @@ cdef class Indicator(TimeSeries):
         return _wrap_indicator(series, clz, *args, **kwargs)
 
 
+cdef class IndicatorOHLC(Indicator):
+    """
+    Extension of indicator class to be used for OHLCV series
+    """
+    def _instantiate_base_series(self, str name, long long timeframe, float max_series_length):
+        return OHLCV(name, timeframe, max_series_length)
+
+    def calculate(self, long long time, Bar value, short new_item_started) -> object:
+        raise ValueError("Indicator must implement calculate() method")
+
+
 cdef class Lag(Indicator):
     cdef int period
 

qubx/core/utils.pyx

@@ -50,5 +50,5 @@ cpdef recognize_timeframe(timeframe):
         tf = np.int64(timeframe.item().total_seconds() * NS) 
 
     else:
-        raise ValueError('Unknown timeframe type !')
+        raise ValueError(f'Unknown timeframe type: {timeframe} !')
     return tf

qubx/data/readers.py

@@ -11,6 +11,7 @@ from functools import wraps
 from qubx import logger
 from qubx.core.series import TimeSeries, OHLCV, time_as_nsec, Quote, Trade
 from qubx.utils.time import infer_series_frequency, handle_start_stop
+from psycopg.types.datetime import TimestampLoader
 
 _DT = lambda x: pd.Timedelta(x).to_numpy().item()
 D1, H1 = _DT("1D"), _DT("1h")
@@ -20,6 +21,12 @@ STOCK_DAILY_SESSION = (_DT("9:30:00.100"), _DT("15:59:59.900"))
 CME_FUTURES_DAILY_SESSION = (_DT("8:30:00.100"), _DT("15:14:59.900"))
 
 
+class NpTimestampLoader(TimestampLoader):
+    def load(self, data) -> np.datetime64:
+        dt = super().load(data)
+        return np.datetime64(dt)
+
+
 def _recognize_t(t: Union[int, str], defaultvalue, timeunit) -> int:
     if isinstance(t, (str, pd.Timestamp)):
         try:
@@ -46,7 +53,7 @@ def _find_column_index_in_list(xs, *args):
 
 
 _FIND_TIME_COL_IDX = lambda column_names: _find_column_index_in_list(
-    column_names, "time", "timestamp", "datetime", "date", "open_time"
+    column_names, "time", "timestamp", "datetime", "date", "open_time", "ts"
 )
 
 
@@ -56,7 +63,13 @@ class DataTransformer:
         self.buffer = []
         self._column_names = []
 
-    def start_transform(self, name: str, column_names: List[str]):
+    def start_transform(
+        self,
+        name: str,
+        column_names: List[str],
+        start: str | None = None,
+        stop: str | None = None,
+    ):
         self._column_names = column_names
         self.buffer = []
 
@@ -181,7 +194,9 @@ class CsvStorageDataReader(DataReader):
 
             def _iter_chunks():
                 for n in range(0, length // chunksize + 1):
-                    transform.start_transform(data_id, fieldnames)
+                    transform.start_transform(
+                        data_id, fieldnames, start=start, stop=stop
+                    )
                     raw_data = (
                         selected_table[n * chunksize : min((n + 1) * chunksize, length)]
                         .to_pandas()
@@ -192,7 +207,7 @@ class CsvStorageDataReader(DataReader):
 
             return _iter_chunks()
 
-        transform.start_transform(data_id, fieldnames)
+        transform.start_transform(data_id, fieldnames, start=start, stop=stop)
         raw_data = selected_table.to_pandas().to_numpy()
         transform.process_data(raw_data)
         return transform.collect()
@@ -213,7 +228,7 @@ class AsPandasFrame(DataTransformer):
     def __init__(self, timestamp_units=None) -> None:
         self.timestamp_units = timestamp_units
 
-    def start_transform(self, name: str, column_names: List[str]):
+    def start_transform(self, name: str, column_names: List[str], **kwargs):
         self._time_idx = _FIND_TIME_COL_IDX(column_names)
         self._column_names = column_names
         self._frame = pd.DataFrame()
@@ -256,7 +271,7 @@ class AsOhlcvSeries(DataTransformer):
         self._data_type = None
         self.timestamp_units = timestamp_units
 
-    def start_transform(self, name: str, column_names: List[str]):
+    def start_transform(self, name: str, column_names: List[str], **kwargs):
         self._time_idx = _FIND_TIME_COL_IDX(column_names)
         self._volume_idx = None
         self._b_volume_idx = None
@@ -376,7 +391,7 @@ class AsQuotes(DataTransformer):
     Data must have appropriate structure: bid, ask, bidsize, asksize and time
     """
 
-    def start_transform(self, name: str, column_names: List[str]):
+    def start_transform(self, name: str, column_names: List[str], **kwargs):
         self.buffer = list()
         self._time_idx = _FIND_TIME_COL_IDX(column_names)
         self._bid_idx = _find_column_index_in_list(column_names, "bid")
@@ -422,7 +437,7 @@ class AsTimestampedRecords(DataTransformer):
     def __init__(self, timestamp_units: str | None = None) -> None:
         self.timestamp_units = timestamp_units
 
-    def start_transform(self, name: str, column_names: List[str]):
+    def start_transform(self, name: str, column_names: List[str], **kwargs):
         self.buffer = list()
         self._time_idx = _FIND_TIME_COL_IDX(column_names)
         self._column_names = column_names
@@ -465,7 +480,7 @@ class RestoreTicksFromOHLC(DataTransformer):
         self._d_session_start = daily_session_start_end[0]
         self._d_session_end = daily_session_start_end[1]
 
-    def start_transform(self, name: str, column_names: List[str]):
+    def start_transform(self, name: str, column_names: List[str], **kwargs):
         self.buffer = []
         # - it will fail if receive data doesn't look as ohlcv
         self._time_idx = _FIND_TIME_COL_IDX(column_names)
@@ -606,10 +621,8 @@ def _retry(fn):
             # print(x, cls._reconnect_tries)
             try:
                 return fn(*args, **kw)
-            except (pg.InterfaceError, pg.OperationalError) as e:
-                logger.warning(
-                    "Database Connection [InterfaceError or OperationalError]"
-                )
+            except (pg.InterfaceError, pg.OperationalError, AttributeError) as e:
+                logger.debug("Database Connection [InterfaceError or OperationalError]")
                 # print ("Idle for %s seconds" % (cls._reconnect_idle))
                 # time.sleep(cls._reconnect_idle)
                 cls._connect()
@@ -700,7 +713,7 @@ class QuestDBSqlCandlesBuilder(QuestDBSqlBuilder):
                 resample
             )
             if resample
-            else resample
+            else "1m"  # if resample is empty let's use 1 minute timeframe
         )
         _rsmpl = f"SAMPLE by {resample}" if resample else ""
 
@@ -749,6 +762,16 @@ class QuestDBConnector(DataReader):
         self._builder = builder
         self._connect()
 
+    def __getstate__(self):
+        if self._connection:
+            self._connection.close()
+            self._connection = None
+        if self._cursor:
+            self._cursor.close()
+            self._cursor = None
+        state = self.__dict__.copy()
+        return state
+
     def _connect(self):
         self._connection = pg.connect(self.connection_url, autocommit=True)
         self._cursor = self._connection.cursor()
@@ -761,7 +784,7 @@ class QuestDBConnector(DataReader):
         stop: str | None = None,
         transform: DataTransformer = DataTransformer(),
         chunksize=0,  # TODO: use self._cursor.fetchmany in this case !!!!
-        timeframe: str = "1m",
+        timeframe: str | None = "1m",
         data_type="candles_1m",
     ) -> Any:
         return self._read(
@@ -786,7 +809,7 @@ class QuestDBConnector(DataReader):
         stop: str | None,
         transform: DataTransformer,
         chunksize: int,  # TODO: use self._cursor.fetchmany in this case !!!!
-        timeframe: str,
+        timeframe: str | None,
         data_type: str,
         builder: QuestDBSqlBuilder,
     ) -> Any:
@@ -795,9 +818,11 @@ class QuestDBConnector(DataReader):
 
         self._cursor.execute(_req)  # type: ignore
         records = self._cursor.fetchall()  # TODO: for chunksize > 0 use fetchmany etc
+        if not records:
+            return None
 
         names = [d.name for d in self._cursor.description]  # type: ignore
-        transform.start_transform(data_id, names)
+        transform.start_transform(data_id, names, start=start, stop=stop)
 
         transform.process_data(records)
         return transform.collect()
@@ -811,54 +836,20 @@ class QuestDBConnector(DataReader):
     def __del__(self):
         for c in (self._cursor, self._connection):
             try:
-                logger.info("Closing connection")
+                logger.debug("Closing connection")
                 c.close()
             except:
                 pass
 
 
-class SnapshotsBuilder(DataTransformer):
-    """
-    Snapshots assembler from OB updates
-    """
-
-    def __init__(
-        self,
-        levels: int = -1,  # how many levels restore, 1 - TOB, -1 - all
-        as_frame=False,  # result is dataframe
-    ):
-        self.buffer = []
-        self.levels = levels
-        self.as_frame = as_frame
-
-    def start_transform(self, name: str, column_names: List[str]):
-        # initialize buffer / series etc
-        # let's keep restored snapshots into some buffer etc
-        self.buffer = []
-
-        # do additional init stuff here
-
-    def process_data(self, rows_data: List[List]) -> Any:
-        for r in rows_data:
-            # restore snapshots and put into buffer or series
-            pass
-
-    def collect(self) -> Any:
-        # - may be convert it to pandas DataFrame ?
-        if self.as_frame:
-            return pd.DataFrame.from_records(self.buffer)  # or custom transform
-
-        # - or just returns as plain list
-        return self.buffer
-
-
-class QuestDBSqlOrderBookBilder(QuestDBSqlBuilder):
+class QuestDBSqlOrderBookBuilder(QuestDBSqlCandlesBuilder):
     """
     Sql builder for snapshot data
     """
 
-    def get_table_name(self, data_id: str, sfx: str = "") -> str:
-        return ""
+    MAX_TIME_DELTA = pd.Timedelta("5h")
+    SNAPSHOT_DELTA = pd.Timedelta("1h")
+    MIN_DELTA = pd.Timedelta("1s")
 
     def prepare_data_sql(
         self,
@@ -868,7 +859,23 @@ class QuestDBSqlOrderBookBilder(QuestDBSqlBuilder):
         resample: str,
         data_type: str,
     ) -> str:
-        return ""
+        if not start or not end:
+            raise ValueError("Start and end dates must be provided for orderbook data!")
+        start_dt, end_dt = pd.Timestamp(start), pd.Timestamp(end)
+        delta = end_dt - start_dt
+        if delta > self.MAX_TIME_DELTA:
+            raise ValueError(
+                f"Time range is too big for orderbook data: {delta}, max allowed: {self.MAX_TIME_DELTA}"
+            )
+
+        raw_start_dt = start_dt.floor(self.SNAPSHOT_DELTA) - self.MIN_DELTA
+
+        table_name = self.get_table_name(data_id, data_type)
+        query = f"""
+SELECT * FROM {table_name}
+WHERE timestamp BETWEEN '{raw_start_dt}' AND '{end_dt}'
+"""
+        return query
 
 
 class TradeSql(QuestDBSqlCandlesBuilder):
@@ -931,7 +938,8 @@ class MultiQdbConnector(QuestDBConnector):
     _TYPE_TO_BUILDER = {
         "candles_1m": QuestDBSqlCandlesBuilder(),
         "trade": TradeSql(),
-        "orderbook": QuestDBSqlOrderBookBilder(),
+        "agg_trade": TradeSql(),
+        "orderbook": QuestDBSqlOrderBookBuilder(),
     }
 
     _TYPE_MAPPINGS = {
@@ -940,6 +948,9 @@ class MultiQdbConnector(QuestDBConnector):
         "ob": "orderbook",
         "trd": "trade",
         "td": "trade",
+        "aggTrade": "agg_trade",
+        "agg_trades": "agg_trade",
+        "aggTrades": "agg_trade",
     }
 
     def __init__(
@@ -974,9 +985,9 @@ class MultiQdbConnector(QuestDBConnector):
         start: str | None = None,
         stop: str | None = None,
         transform: DataTransformer = DataTransformer(),
-        chunksize=0,  # TODO: use self._cursor.fetchmany in this case !!!!
+        chunksize: int = 0,  # TODO: use self._cursor.fetchmany in this case !!!!
         timeframe: str | None = None,
-        data_type="candles",
+        data_type: str = "candles",
     ) -> Any:
         _mapped_data_type = self._TYPE_MAPPINGS.get(data_type, data_type)
         return self._read(

qubx/math/__init__.py

@@ -1 +1 @@
-from .math import percentile_rank
+from .stats import compare_to_norm, percentile_rank, kde

qubx/math/stats.py

@@ -30,13 +30,30 @@ def compare_to_norm(xs, xranges=None):
     fit = stats.norm.pdf(sorted(xs), _m, _s)
 
     sbp(12, 1)
-    plt.plot(sorted(xs), fit, 'r--', lw=2, label='N(%.2f, %.2f)' % (_m, _s))
-    plt.legend(loc='upper right')
+    plt.plot(sorted(xs), fit, "r--", lw=2, label="N(%.2f, %.2f)" % (_m, _s))
+    plt.legend(loc="upper right")
 
-    sns.kdeplot(xs, color='g', label='Data', shade=True)
+    sns.kdeplot(xs, color="g", label="Data", fill=True)
     if xranges is not None and len(xranges) > 1:
         plt.xlim(xranges)
-    plt.legend(loc='upper right')
+    plt.legend(loc="upper right")
 
     sbp(12, 2)
     stats.probplot(xs, dist="norm", sparams=(_m, _s), plot=plt)
+
+
+def kde(array, cut_down=True, bw_method="scott"):
+    """
+    Kernel dense estimation
+    """
+    from scipy.stats import gaussian_kde
+
+    if cut_down:
+        bins, counts = np.unique(array, return_counts=True)
+        f_mean = counts.mean()
+        f_above_mean = bins[counts > f_mean]
+        if len(f_above_mean) > 0:
+            bounds = [f_above_mean.min(), f_above_mean.max()]
+            array = array[np.bitwise_and(bounds[0] < array, array < bounds[1])]
+
+    return gaussian_kde(array, bw_method=bw_method)