Source code for bokeh.models.sources

#-----------------------------------------------------------------------------
# Copyright (c) 2012 - 2021, Anaconda, Inc., and Bokeh Contributors.
# All rights reserved.
#
# The full license is in the file LICENSE.txt, distributed with this software.
#-----------------------------------------------------------------------------

#-----------------------------------------------------------------------------
# Boilerplate
#-----------------------------------------------------------------------------
from __future__ import annotations

import logging # isort:skip
log = logging.getLogger(__name__)

#-----------------------------------------------------------------------------
# Imports
#-----------------------------------------------------------------------------

# Standard library imports
import warnings
from typing import (
    TYPE_CHECKING,
    Any as TAny,
    Dict as TDict,
    List as TList,
    Sequence,
    Set,
    Tuple,
    Union,
    overload,
)

## External imports
if TYPE_CHECKING:
    import pandas as pd
else:
    from ..util.dependencies import import_optional
    pd = import_optional("pandas")

# Bokeh imports
from ..core.has_props import abstract
from ..core.properties import (
    JSON,
    Any,
    Bool,
    ColumnData,
    Dict,
    Enum,
    Instance,
    Int,
    List,
    NonNullable,
    Nullable,
    PandasDataFrame,
    PandasGroupBy,
    Readonly,
    Seq,
    String,
)
from ..model import Model
from ..util.serialization import convert_datetime_array
from ..util.warnings import BokehUserWarning
from .callbacks import CustomJS
from .filters import Filter
from .selections import Selection, SelectionPolicy, UnionRenderers

if TYPE_CHECKING:
    from ..core.has_props import Setter
    from ..core.types import Unknown

#-----------------------------------------------------------------------------
# Globals and constants
#-----------------------------------------------------------------------------

__all__ = (
    'AjaxDataSource',
    'CDSView',
    'ColumnarDataSource',
    'ColumnDataSource',
    'DataSource',
    'GeoJSONDataSource',
    'ServerSentDataSource',
    'WebDataSource',
    'WebSource',
)

#-----------------------------------------------------------------------------
# General API
#-----------------------------------------------------------------------------

if TYPE_CHECKING:
    DataDict = TDict[str, Sequence[Unknown]]

    Index = Union[int, slice, Tuple[Union[int, slice], ...]]

    Patches = TDict[str, TList[Tuple[Index, Unknown]]]

[docs]@abstract class DataSource(Model): ''' A base class for data source types. ''' selected = Readonly(Instance(Selection), default=lambda: Selection(), help=""" An instance of a ``Selection`` that indicates selected indices on this ``DataSource``. This is a read-only property. You may only change the attributes of this object to change the selection (e.g., ``selected.indices``). """)
[docs]@abstract class ColumnarDataSource(DataSource): ''' A base class for data source types, which can be mapped onto a columnar format. ''' selection_policy = Instance(SelectionPolicy, default=lambda: UnionRenderers(), help=""" An instance of a ``SelectionPolicy`` that determines how selections are set. """)
[docs]class ColumnDataSource(ColumnarDataSource): ''' Maps names of columns to sequences or arrays. The ``ColumnDataSource`` is a fundamental data structure of Bokeh. Most plots, data tables, etc. will be driven by a ``ColumnDataSource``. If the ``ColumnDataSource`` initializer is called with a single argument that can be any of the following: * A Python ``dict`` that maps string names to sequences of values, e.g. lists, arrays, etc. .. code-block:: python data = {'x': [1,2,3,4], 'y': np.array([10.0, 20.0, 30.0, 40.0])} source = ColumnDataSource(data) .. note:: ``ColumnDataSource`` only creates a shallow copy of ``data``. Use e.g. ``ColumnDataSource(copy.deepcopy(data))`` if initializing from another ``ColumnDataSource.data`` object that you want to keep independent. * A Pandas ``DataFrame`` object .. code-block:: python source = ColumnDataSource(df) In this case the CDS will have columns corresponding to the columns of the ``DataFrame``. If the ``DataFrame`` columns have multiple levels, they will be flattened using an underscore (e.g. level_0_col_level_1_col). The index of the ``DataFrame`` will be flattened to an ``Index`` of tuples if it's a ``MultiIndex``, and then reset using ``reset_index``. The result will be a column with the same name if the index was named, or level_0_name_level_1_name if it was a named ``MultiIndex``. If the ``Index`` did not have a name or the ``MultiIndex`` name could not be flattened/determined, the ``reset_index`` function will name the index column ``index``, or ``level_0`` if the name ``index`` is not available. * A Pandas ``GroupBy`` object .. code-block:: python group = df.groupby(('colA', 'ColB')) In this case the CDS will have columns corresponding to the result of calling ``group.describe()``. The ``describe`` method generates columns for statistical measures such as ``mean`` and ``count`` for all the non-grouped original columns. The CDS columns are formed by joining original column names with the computed measure. For example, if a ``DataFrame`` has columns ``'year'`` and ``'mpg'``. Then passing ``df.groupby('year')`` to a CDS will result in columns such as ``'mpg_mean'`` If the ``GroupBy.describe`` result has a named index column, then CDS will also have a column with this name. However, if the index name (or any subname of a ``MultiIndex``) is ``None``, then the CDS will have a column generically named ``index`` for the index. Note this capability to adapt ``GroupBy`` objects may only work with Pandas ``>=0.20.0``. .. note:: There is an implicit assumption that all the columns in a given ``ColumnDataSource`` all have the same length at all times. For this reason, it is usually preferable to update the ``.data`` property of a data source "all at once". ''' data: DataDict = ColumnData(String, Seq(Any), help=""" Mapping of column names to sequences of data. The columns can be, e.g, Python lists or tuples, NumPy arrays, etc. The .data attribute can also be set from Pandas DataFrames or GroupBy objects. In these cases, the behaviour is identical to passing the objects to the ``ColumnDataSource`` initializer. """).accepts( PandasDataFrame, lambda x: ColumnDataSource._data_from_df(x) ).accepts( PandasGroupBy, lambda x: ColumnDataSource._data_from_groupby(x) ).asserts(lambda _, data: len({len(x) for x in data.values()}) <= 1, lambda obj, name, data: warnings.warn( "ColumnDataSource's columns must be of the same length. " + "Current lengths: %s" % ", ".join(sorted(str((k, len(v))) for k, v in data.items())), BokehUserWarning)) @overload def __init__(self, data: DataDict | pd.DataFrame | pd.core.groupby.GroupBy, **kwargs: TAny) -> None: ... @overload def __init__(self, **kwargs: TAny) -> None: ...
[docs] def __init__(self, *args: TAny, **kwargs: TAny) -> None: ''' If called with a single argument that is a dict or ``pandas.DataFrame``, treat that implicitly as the "data" attribute. ''' if len(args) == 1 and "data" not in kwargs: kwargs["data"] = args[0] # TODO (bev) invalid to pass args and "data", check and raise exception raw_data: DataDict = kwargs.pop("data", {}) if not isinstance(raw_data, dict): if pd and isinstance(raw_data, pd.DataFrame): raw_data = self._data_from_df(raw_data) elif pd and isinstance(raw_data, pd.core.groupby.GroupBy): raw_data = self._data_from_groupby(raw_data) else: raise ValueError(f"expected a dict or pandas.DataFrame, got {raw_data}") super().__init__(**kwargs) self.data.update(raw_data)
@property def column_names(self) -> TList[str]: ''' A list of the column names in this data source. ''' return list(self.data) @staticmethod def _data_from_df(df: pd.DataFrame) -> DataDict: ''' Create a ``dict`` of columns from a Pandas ``DataFrame``, suitable for creating a ColumnDataSource. Args: df (DataFrame) : data to convert Returns: dict[str, np.array] ''' _df = df.copy() # Flatten columns if isinstance(df.columns, pd.MultiIndex): try: _df.columns = ['_'.join(col) for col in _df.columns.values] except TypeError: raise TypeError('Could not flatten MultiIndex columns. ' 'use string column names or flatten manually') # Transform columns CategoricalIndex in list if isinstance(df.columns, pd.CategoricalIndex): _df.columns = df.columns.tolist() # Flatten index index_name = ColumnDataSource._df_index_name(df) if index_name == 'index': _df.index = pd.Index(_df.index.values) else: _df.index = pd.Index(_df.index.values, name=index_name) _df.reset_index(inplace=True) tmp_data = {c: v.values for c, v in _df.items()} new_data: DataDict = {} for k, v in tmp_data.items(): new_data[k] = v return new_data @staticmethod def _data_from_groupby(group: pd.core.groupby.GroupBy) -> DataDict: ''' Create a ``dict`` of columns from a Pandas ``GroupBy``, suitable for creating a ``ColumnDataSource``. The data generated is the result of running ``describe`` on the group. Args: group (GroupBy) : data to convert Returns: dict[str, np.array] ''' return ColumnDataSource._data_from_df(group.describe()) @staticmethod def _df_index_name(df: pd.DataFrame) -> str: ''' Return the Bokeh-appropriate column name for a ``DataFrame`` index If there is no named index, then `"index" is returned. If there is a single named index, then ``df.index.name`` is returned. If there is a multi-index, and the index names are all strings, then the names are joined with '_' and the result is returned, e.g. for a multi-index ``['ind1', 'ind2']`` the result will be "ind1_ind2". Otherwise if any index name is not a string, the fallback name "index" is returned. Args: df (DataFrame) : the ``DataFrame`` to find an index name for Returns: str ''' if df.index.name: return df.index.name elif df.index.names: try: return "_".join(df.index.names) except TypeError: return "index" else: return "index"
[docs] @classmethod def from_df(cls, data: pd.DataFrame) -> DataDict: ''' Create a ``dict`` of columns from a Pandas ``DataFrame``, suitable for creating a ``ColumnDataSource``. Args: data (DataFrame) : data to convert Returns: dict[str, np.array] ''' return cls._data_from_df(data)
[docs] @classmethod def from_groupby(cls, data: pd.core.groupby.GroupBy) -> DataDict: ''' Create a ``dict`` of columns from a Pandas ``GroupBy``, suitable for creating a ``ColumnDataSource``. The data generated is the result of running ``describe`` on the group. Args: data (Groupby) : data to convert Returns: dict[str, np.array] ''' return cls._data_from_df(data.describe())
[docs] def to_df(self) -> pd.DataFrame: ''' Convert this data source to pandas ``DataFrame``. Returns: DataFrame ''' if not pd: raise RuntimeError('Pandas must be installed to convert to a Pandas Dataframe') return pd.DataFrame(self.data)
[docs] def add(self, data: Sequence[Unknown], name: str | None = None) -> str: ''' Appends a new column of data to the data source. Args: data (seq) : new data to add name (str, optional) : column name to use. If not supplied, generate a name of the form "Series ####" Returns: str: the column name used ''' if name is None: n = len(self.data) while f"Series {n}" in self.data: n += 1 name = f"Series {n}" self.data[name] = data return name
[docs] def remove(self, name: str) -> None: ''' Remove a column of data. Args: name (str) : name of the column to remove Returns: None .. note:: If the column name does not exist, a warning is issued. ''' try: del self.data[name] except (ValueError, KeyError): warnings.warn("Unable to find column '%s' in data source" % name)
[docs] def stream(self, new_data: DataDict, rollover: int | None = None) -> None: ''' Efficiently update data source columns with new append-only data. In cases where it is necessary to update data columns in, this method can efficiently send only the new data, instead of requiring the entire data set to be re-sent. Args: new_data (dict[str, seq]) : a mapping of column names to sequences of new data to append to each column. All columns of the data source must be present in ``new_data``, with identical-length append data. rollover (int, optional) : A maximum column size, above which data from the start of the column begins to be discarded. If None, then columns will continue to grow unbounded (default: None) Returns: None Raises: ValueError Example: .. code-block:: python source = ColumnDataSource(data=dict(foo=[], bar=[])) # has new, identical-length updates for all columns in source new_data = { 'foo' : [10, 20], 'bar' : [100, 200], } source.stream(new_data) ''' # calls internal implementation self._stream(new_data, rollover)
def _stream(self, new_data: DataDict | pd.Series | pd.DataFrame, rollover: int | None = None, setter: Setter | None = None) -> None: ''' Internal implementation to efficiently update data source columns with new append-only data. The internal implementation adds the setter attribute. [https://github.com/bokeh/bokeh/issues/6577] In cases where it is necessary to update data columns in, this method can efficiently send only the new data, instead of requiring the entire data set to be re-sent. Args: new_data (dict[str, seq] or DataFrame or Series) : a mapping of column names to sequences of new data to append to each column, a pandas DataFrame, or a pandas Series in case of a single row - in this case the Series index is used as column names All columns of the data source must be present in ``new_data``, with identical-length append data. rollover (int, optional) : A maximum column size, above which data from the start of the column begins to be discarded. If None, then columns will continue to grow unbounded (default: None) setter (ClientSession or ServerSession or None, optional) : This is used to prevent "boomerang" updates to Bokeh apps. (default: None) In the context of a Bokeh server application, incoming updates to properties will be annotated with the session that is doing the updating. This value is propagated through any subsequent change notifications that the update triggers. The session can compare the event setter to itself, and suppress any updates that originate from itself. Returns: None Raises: ValueError Example: .. code-block:: python source = ColumnDataSource(data=dict(foo=[], bar=[])) # has new, identical-length updates for all columns in source new_data = { 'foo' : [10, 20], 'bar' : [100, 200], } source.stream(new_data) ''' needs_length_check = True if pd and isinstance(new_data, (pd.Series, pd.DataFrame)): if isinstance(new_data, pd.Series): new_data = new_data.to_frame().T needs_length_check = False # DataFrame lengths equal by definition _df = new_data newkeys = set(_df.columns) index_name = ColumnDataSource._df_index_name(_df) newkeys.add(index_name) new_data = dict(_df.items()) new_data[index_name] = _df.index.values else: newkeys = set(new_data.keys()) oldkeys = set(self.data.keys()) if newkeys != oldkeys: missing = oldkeys - newkeys extra = newkeys - oldkeys if missing and extra: raise ValueError( "Must stream updates to all existing columns (missing: %s, extra: %s)" % (", ".join(sorted(missing)), ", ".join(sorted(extra))) ) elif missing: raise ValueError("Must stream updates to all existing columns (missing: %s)" % ", ".join(sorted(missing))) else: raise ValueError("Must stream updates to all existing columns (extra: %s)" % ", ".join(sorted(extra))) import numpy as np if needs_length_check: lengths: Set[int] = set() arr_types = (np.ndarray, pd.Series) if pd else np.ndarray for _, x in new_data.items(): if isinstance(x, arr_types): if len(x.shape) != 1: raise ValueError("stream(...) only supports 1d sequences, got ndarray with size %r" % (x.shape,)) lengths.add(x.shape[0]) else: lengths.add(len(x)) if len(lengths) > 1: raise ValueError("All streaming column updates must be the same length") # slightly awkward that we have to call convert_datetime_array here ourselves # but the downstream code expects things to already be ms-since-epoch for key, values in new_data.items(): if pd and isinstance(values, (pd.Series, pd.Index)): values = values.values old_values = self.data[key] # Apply the transformation if the new data contains datetimes # but the current data has already been transformed if (isinstance(values, np.ndarray) and values.dtype.kind.lower() == 'm' and isinstance(old_values, np.ndarray) and old_values.dtype.kind.lower() != 'm'): new_data[key] = convert_datetime_array(values) else: new_data[key] = values self.data._stream(self.document, self, new_data, rollover, setter)
[docs] def patch(self, patches: Patches, setter: Setter | None = None) -> None: ''' Efficiently update data source columns at specific locations If it is only necessary to update a small subset of data in a ``ColumnDataSource``, this method can be used to efficiently update only the subset, instead of requiring the entire data set to be sent. This method should be passed a dictionary that maps column names to lists of tuples that describe a patch change to apply. To replace individual items in columns entirely, the tuples should be of the form: .. code-block:: python (index, new_value) # replace a single column value # or (slice, new_values) # replace several column values Values at an index or slice will be replaced with the corresponding new values. In the case of columns whose values are other arrays or lists, (e.g. image or patches glyphs), it is also possible to patch "subregions". In this case the first item of the tuple should be a whose first element is the index of the array item in the CDS patch, and whose subsequent elements are integer indices or slices into the array item: .. code-block:: python # replace the entire 10th column of the 2nd array: +----------------- index of item in column data source | | +--------- row subindex into array item | | | | +- column subindex into array item V V V ([2, slice(None), 10], new_values) Imagining a list of 2d NumPy arrays, the patch above is roughly equivalent to: .. code-block:: python data = [arr1, arr2, ...] # list of 2d arrays data[2][:, 10] = new_data There are some limitations to the kinds of slices and data that can be accepted. * Negative ``start``, ``stop``, or ``step`` values for slices will result in a ``ValueError``. * In a slice, ``start > stop`` will result in a ``ValueError`` * When patching 1d or 2d subitems, the subitems must be NumPy arrays. * New values must be supplied as a **flattened one-dimensional array** of the appropriate size. Args: patches (dict[str, list[tuple]]) : lists of patches for each column Returns: None Raises: ValueError Example: The following example shows how to patch entire column elements. In this case, .. code-block:: python source = ColumnDataSource(data=dict(foo=[10, 20, 30], bar=[100, 200, 300])) patches = { 'foo' : [ (slice(2), [11, 12]) ], 'bar' : [ (0, 101), (2, 301) ], } source.patch(patches) After this operation, the value of the ``source.data`` will be: .. code-block:: python dict(foo=[11, 12, 30], bar=[101, 200, 301]) For a more comprehensive complete example, see :bokeh-tree:`examples/howto/patch_app.py`. ''' import numpy as np extra = set(patches.keys()) - set(self.data.keys()) if extra: raise ValueError("Can only patch existing columns (extra: %s)" % ", ".join(sorted(extra))) for name, patch in patches.items(): col_len = len(self.data[name]) for ind, _ in patch: # integer index, patch single value of 1d column if isinstance(ind, int): if ind > col_len or ind < 0: raise ValueError("Out-of bounds index (%d) in patch for column: %s" % (ind, name)) # slice index, patch multiple values of 1d column elif isinstance(ind, slice): _check_slice(ind) if ind.stop is not None and ind.stop > col_len: raise ValueError("Out-of bounds slice index stop (%d) in patch for column: %s" % (ind.stop, name)) # multi-index, patch sub-regions of "n-d" column elif isinstance(ind, (list, tuple)): if len(ind) == 0: raise ValueError("Empty (length zero) patch multi-index") if len(ind) == 1: raise ValueError("Patch multi-index must contain more than one subindex") ind_0 = ind[0] if not isinstance(ind_0, int): raise ValueError("Initial patch sub-index may only be integer, got: %s" % ind_0) if ind_0 > col_len or ind_0 < 0: raise ValueError("Out-of bounds initial sub-index (%d) in patch for column: %s" % (ind, name)) if not isinstance(self.data[name][ind_0], np.ndarray): raise ValueError("Can only sub-patch into columns with NumPy array items") if len(self.data[name][ind_0].shape) != (len(ind)-1): raise ValueError("Shape mismatch between patch slice and sliced data") elif isinstance(ind_0, slice): _check_slice(ind_0) if ind_0.stop is not None and ind_0.stop > col_len: raise ValueError("Out-of bounds initial slice sub-index stop (%d) in patch for column: %s" % (ind.stop, name)) # Note: bounds of sub-indices after the first are not checked! for subind in ind[1:]: if not isinstance(subind, (int, slice)): raise ValueError("Invalid patch sub-index: %s" % subind) if isinstance(subind, slice): _check_slice(subind) else: raise ValueError("Invalid patch index: %s" % ind) self.data._patch(self.document, self, patches, setter)
[docs]class CDSView(Model): ''' A view into a ``ColumnDataSource`` that represents a row-wise subset. ''' filters = List(Instance(Filter), default=[], help=""" List of filters that the view comprises. """) source = Instance(ColumnarDataSource, help=""" The ``ColumnDataSource`` associated with this view. Used to determine the length of the columns. """)
[docs]class GeoJSONDataSource(ColumnarDataSource): ''' ''' geojson = NonNullable(JSON, help=""" GeoJSON that contains features for plotting. Currently ``GeoJSONDataSource`` can only process a ``FeatureCollection`` or ``GeometryCollection``. """)
[docs]@abstract class WebDataSource(ColumnDataSource): ''' Base class for web column data sources that can update from data URLs. .. note:: This base class is typically not useful to instantiate on its own. ''' adapter = Nullable(Instance(CustomJS), help=""" A JavaScript callback to adapt raw JSON responses to Bokeh ``ColumnDataSource`` format. If provided, this callback is executes immediately after the JSON data is received, but before appending or replacing data in the data source. The ``CustomJS`` callback will receive the ``AjaxDataSource`` as ``cb_obj`` and will receive the raw JSON response as ``cb_data.response``. The callback code should return a ``data`` object suitable for a Bokeh ``ColumnDataSource`` (i.e. a mapping of string column names to arrays of data). """) max_size = Nullable(Int, help=""" Maximum size of the data columns. If a new fetch would result in columns larger than ``max_size``, then earlier data is dropped to make room. """) mode = Enum("replace", "append", help=""" Whether to append new data to existing data (up to ``max_size``), or to replace existing data entirely. """) data_url = NonNullable(String, help=""" A URL to to fetch data from. """)
# TODO: deprecated, remove at bokeh 3.0 WebSource = WebDataSource
[docs]class ServerSentDataSource(WebDataSource): ''' A data source that can populate columns by receiving server sent events endpoints. '''
[docs]class AjaxDataSource(WebDataSource): ''' A data source that can populate columns by making Ajax calls to REST endpoints. The ``AjaxDataSource`` can be especially useful if you want to make a standalone document (i.e. not backed by the Bokeh server) that can still dynamically update using an existing REST API. The response from the REST API should match the ``.data`` property of a standard ``ColumnDataSource``, i.e. a JSON dict that maps names to arrays of values: .. code-block:: python { 'x' : [1, 2, 3, ...], 'y' : [9, 3, 2, ...] } Alternatively, if the REST API returns a different format, a ``CustomJS`` callback can be provided to convert the REST response into Bokeh format, via the ``adapter`` property of this data source. Initial data can be set by specifying the ``data`` property directly. This is necessary when used in conjunction with a ``FactorRange``, even if the columns in `data`` are empty. A full example can be seen at :bokeh-tree:`examples/howto/ajax_source.py` ''' polling_interval = Nullable(Int, help=""" A polling interval (in milliseconds) for updating data source. """) method = Enum('POST', 'GET', help=""" Specify the HTTP method to use for the Ajax request (GET or POST) """) if_modified = Bool(False, help=""" Whether to include an ``If-Modified-Since`` header in Ajax requests to the server. If this header is supported by the server, then only new data since the last request will be returned. """) content_type = String(default='application/json', help=""" Set the "contentType" parameter for the Ajax request. """) http_headers = Dict(String, String, help=""" Specify HTTP headers to set for the Ajax request. Example: .. code-block:: python ajax_source.headers = { 'x-my-custom-header': 'some value' } """)
#----------------------------------------------------------------------------- # Dev API #----------------------------------------------------------------------------- #----------------------------------------------------------------------------- # Private API #----------------------------------------------------------------------------- def _check_slice(s: slice) -> None: if (s.start is not None and s.stop is not None and s.start > s.stop): raise ValueError("Patch slices must have start < end, got %s" % s) if (s.start is not None and s.start < 0) or \ (s.stop is not None and s.stop < 0) or \ (s.step is not None and s.step < 0): raise ValueError("Patch slices must have non-negative (start, stop, step) values, got %s" % s) #----------------------------------------------------------------------------- # Code #-----------------------------------------------------------------------------