data_portal.py

#
# Copyright 2016 Quantopian, Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from operator import mul

from logbook import Logger

import numpy as np
import pandas as pd
from pandas.tslib import normalize_date
from six import iteritems
from six.moves import reduce

from zipline.assets import Asset, Future, Equity
from zipline.data.dispatch_bar_reader import (
    AssetDispatchMinuteBarReader,
    AssetDispatchSessionBarReader
)
from zipline.data.resample import (
    DailyHistoryAggregator,
    ReindexMinuteBarReader,
    ReindexSessionBarReader,
)
from zipline.data.history_loader import (
    DailyHistoryLoader,
    MinuteHistoryLoader,
)
from zipline.data.us_equity_pricing import NoDataOnDate

from zipline.utils.math_utils import (
    nansum,
    nanmean,
    nanstd
)
from zipline.utils.memoize import remember_last, weak_lru_cache
from zipline.utils.pandas_utils import timedelta_to_integral_minutes
from zipline.errors import (
    NoTradeDataAvailableTooEarly,
    NoTradeDataAvailableTooLate,
    HistoryWindowStartsBeforeData,
)

log = Logger('DataPortal')

BASE_FIELDS = frozenset([
    "open", "high", "low", "close", "volume", "price", "last_traded"
])

OHLCV_FIELDS = frozenset([
    "open", "high", "low", "close", "volume"
])

OHLCVP_FIELDS = frozenset([
    "open", "high", "low", "close", "volume", "price"
])

HISTORY_FREQUENCIES = set(["1m", "1d"])


class DataPortal(object):
    """Interface to all of the data that a zipline simulation needs.

    This is used by the simulation runner to answer questions about the data,
    like getting the prices of assets on a given day or to service history
    calls.

    Parameters
    ----------
    asset_finder : zipline.assets.assets.AssetFinder
        The AssetFinder instance used to resolve assets.
    trading_calendar: zipline.utils.calendar.exchange_calendar.TradingCalendar
        The calendar instance used to provide minute->session information.
    first_trading_day : pd.Timestamp
        The first trading day for the simulation.
    equity_daily_reader : BcolzDailyBarReader, optional
        The daily bar reader for equities. This will be used to service
        daily data backtests or daily history calls in a minute backetest.
        If a daily bar reader is not provided but a minute bar reader is,
        the minutes will be rolled up to serve the daily requests.
    equity_minute_reader : BcolzMinuteBarReader, optional
        The minute bar reader for equities. This will be used to service
        minute data backtests or minute history calls. This can be used
        to serve daily calls if no daily bar reader is provided.
    future_daily_reader : BcolzDailyBarReader, optional
        The daily bar ready for futures. This will be used to service
        daily data backtests or daily history calls in a minute backetest.
        If a daily bar reader is not provided but a minute bar reader is,
        the minutes will be rolled up to serve the daily requests.
    future_minute_reader : BcolzFutureMinuteBarReader, optional
        The minute bar reader for futures. This will be used to service
        minute data backtests or minute history calls. This can be used
        to serve daily calls if no daily bar reader is provided.
    adjustment_reader : SQLiteAdjustmentWriter, optional
        The adjustment reader. This is used to apply splits, dividends, and
        other adjustment data to the raw data from the readers.
    last_available_session : pd.Timestamp, optional
        The last session to make available in session-level data.
    last_available_minute : pd.Timestamp, optional
        The last minute to make available in minute-level data.
    """
    def __init__(self,
                 asset_finder,
                 trading_calendar,
                 first_trading_day,
                 equity_daily_reader=None,
                 equity_minute_reader=None,
                 future_daily_reader=None,
                 future_minute_reader=None,
                 adjustment_reader=None,
                 last_available_session=None,
                 last_available_minute=None):

        self.trading_calendar = trading_calendar
        self.asset_finder = asset_finder

        self._adjustment_reader = adjustment_reader

        # caches of sid -> adjustment list
        self._splits_dict = {}
        self._mergers_dict = {}
        self._dividends_dict = {}

        # Cache of sid -> the first trading day of an asset.
        self._asset_start_dates = {}
        self._asset_end_dates = {}

        # Handle extra sources, like Fetcher.
        self._augmented_sources_map = {}
        self._extra_source_df = None

        self._first_trading_session = first_trading_day

        _last_sessions = [r.last_available_dt
                          for r in [equity_daily_reader, future_daily_reader]
                          if r is not None]
        if _last_sessions:
            self._last_trading_session = min(_last_sessions)
        else:
            self._last_trading_session = None

        aligned_equity_minute_reader = self._ensure_reader_aligned(
            equity_minute_reader)
        aligned_equity_session_reader = self._ensure_reader_aligned(
            equity_daily_reader)
        aligned_future_minute_reader = self._ensure_reader_aligned(
            future_minute_reader)
        aligned_future_session_reader = self._ensure_reader_aligned(
            future_daily_reader)

        aligned_minute_readers = {}
        aligned_session_readers = {}

        if aligned_equity_minute_reader is not None:
            aligned_minute_readers[Equity] = aligned_equity_minute_reader
        if aligned_equity_session_reader is not None:
            aligned_session_readers[Equity] = aligned_equity_session_reader

        if aligned_future_minute_reader is not None:
            aligned_minute_readers[Future] = aligned_future_minute_reader
        if aligned_future_session_reader is not None:
            aligned_session_readers[Future] = aligned_future_session_reader

        _dispatch_minute_reader = AssetDispatchMinuteBarReader(
            self.trading_calendar,
            self.asset_finder,
            aligned_minute_readers,
            last_available_minute,
        )

        _dispatch_session_reader = AssetDispatchSessionBarReader(
            self.trading_calendar,
            self.asset_finder,
            aligned_session_readers,
            last_available_session,
        )

        self._pricing_readers = {
            'minute': _dispatch_minute_reader,
            'daily': _dispatch_session_reader,
        }

        self._daily_aggregator = DailyHistoryAggregator(
            self.trading_calendar.schedule.market_open,
            _dispatch_minute_reader,
            self.trading_calendar
        )
        self._history_loader = DailyHistoryLoader(
            self.trading_calendar,
            _dispatch_session_reader,
            self._adjustment_reader
        )
        self._minute_history_loader = MinuteHistoryLoader(
            self.trading_calendar,
            _dispatch_minute_reader,
            self._adjustment_reader
        )

        self._first_trading_day = first_trading_day

        # Get the first trading minute
        self._first_trading_minute, _ = (
            self.trading_calendar.open_and_close_for_session(
                self._first_trading_day
            )
            if self._first_trading_day is not None else (None, None)
        )

        # Store the locs of the first day and first minute
        self._first_trading_day_loc = (
            self.trading_calendar.all_sessions.get_loc(self._first_trading_day)
            if self._first_trading_day is not None else None
        )

    def _ensure_reader_aligned(self, reader):
        if reader is None:
            return

        if reader.trading_calendar.name == self.trading_calendar.name:
            return reader
        elif reader.data_frequency == 'minute':
            return ReindexMinuteBarReader(
                self.trading_calendar,
                reader,
                self._first_trading_session,
                self._last_trading_session
            )
        elif reader.data_frequency == 'session':
            return ReindexSessionBarReader(
                self.trading_calendar,
                reader,
                self._first_trading_session,
                self._last_trading_session
            )

    def _reindex_extra_source(self, df, source_date_index):
        return df.reindex(index=source_date_index, method='ffill')

    def handle_extra_source(self, source_df, sim_params):
        """
        Extra sources always have a sid column.

        We expand the given data (by forward filling) to the full range of
        the simulation dates, so that lookup is fast during simulation.
        """
        if source_df is None:
            return

        # Normalize all the dates in the df
        source_df.index = source_df.index.normalize()

        # source_df's sid column can either consist of assets we know about
        # (such as sid(24)) or of assets we don't know about (such as
        # palladium).
        #
        # In both cases, we break up the dataframe into individual dfs
        # that only contain a single asset's information.  ie, if source_df
        # has data for PALLADIUM and GOLD, we split source_df into two
        # dataframes, one for each. (same applies if source_df has data for
        # AAPL and IBM).
        #
        # We then take each child df and reindex it to the simulation's date
        # range by forward-filling missing values. this makes reads simpler.
        #
        # Finally, we store the data. For each column, we store a mapping in
        # self.augmented_sources_map from the column to a dictionary of
        # asset -> df.  In other words,
        # self.augmented_sources_map['days_to_cover']['AAPL'] gives us the df
        # holding that data.
        source_date_index = self.trading_calendar.sessions_in_range(
            sim_params.start_session,
            sim_params.end_session
        )

        # Break the source_df up into one dataframe per sid.  This lets
        # us (more easily) calculate accurate start/end dates for each sid,
        # de-dup data, and expand the data to fit the backtest start/end date.
        grouped_by_sid = source_df.groupby(["sid"])
        group_names = grouped_by_sid.groups.keys()
        group_dict = {}
        for group_name in group_names:
            group_dict[group_name] = grouped_by_sid.get_group(group_name)

        # This will be the dataframe which we query to get fetcher assets at
        # any given time. Get's overwritten every time there's a new fetcher
        # call
        extra_source_df = pd.DataFrame()

        for identifier, df in iteritems(group_dict):
            # Before reindexing, save the earliest and latest dates
            earliest_date = df.index[0]
            latest_date = df.index[-1]

            # Since we know this df only contains a single sid, we can safely
            # de-dupe by the index (dt). If minute granularity, will take the
            # last data point on any given day
            df = df.groupby(level=0).last()

            # Reindex the dataframe based on the backtest start/end date.
            # This makes reads easier during the backtest.
            df = self._reindex_extra_source(df, source_date_index)

            if not isinstance(identifier, Asset):
                # for fake assets we need to store a start/end date
                self._asset_start_dates[identifier] = earliest_date
                self._asset_end_dates[identifier] = latest_date

            for col_name in df.columns.difference(['sid']):
                if col_name not in self._augmented_sources_map:
                    self._augmented_sources_map[col_name] = {}

                self._augmented_sources_map[col_name][identifier] = df

            # Append to extra_source_df the reindexed dataframe for the single
            # sid
            extra_source_df = extra_source_df.append(df)

        self._extra_source_df = extra_source_df

    def _get_pricing_reader(self, data_frequency):
        return self._pricing_readers[data_frequency]

    def get_last_traded_dt(self, asset, dt, data_frequency):
        """
        Given an asset and dt, returns the last traded dt from the viewpoint
        of the given dt.

        If there is a trade on the dt, the answer is dt provided.
        """
        return self._get_pricing_reader(data_frequency).get_last_traded_dt(
            asset, dt)

    @staticmethod
    def _is_extra_source(asset, field, map):
        """
        Internal method that determines if this asset/field combination
        represents a fetcher value or a regular OHLCVP lookup.
        """
        # If we have an extra source with a column called "price", only look
        # at it if it's on something like palladium and not AAPL (since our
        # own price data always wins when dealing with assets).

        return not (field in BASE_FIELDS and isinstance(asset, Asset))

    def _get_fetcher_value(self, asset, field, dt):
        day = normalize_date(dt)

        try:
            return \
                self._augmented_sources_map[field][asset].loc[day, field]
        except KeyError:
            return np.NaN

    def get_spot_value(self, asset, field, dt, data_frequency):
        """
        Public API method that returns a scalar value representing the value
        of the desired asset's field at either the given dt.

        Parameters
        ----------
        asset : Asset
            The asset whose data is desired.
        field : {'open', 'high', 'low', 'close', 'volume',
                 'price', 'last_traded'}
            The desired field of the asset.
        dt : pd.Timestamp
            The timestamp for the desired value.
        data_frequency : str
            The frequency of the data to query; i.e. whether the data is
            'daily' or 'minute' bars

        Returns
        -------
        value : float, int, or pd.Timestamp
            The spot value of ``field`` for ``asset`` The return type is based
            on the ``field`` requested. If the field is one of 'open', 'high',
            'low', 'close', or 'price', the value will be a float. If the
            ``field`` is 'volume' the value will be a int. If the ``field`` is
            'last_traded' the value will be a Timestamp.
        """
        if self._is_extra_source(asset, field, self._augmented_sources_map):
            return self._get_fetcher_value(asset, field, dt)

        if field not in BASE_FIELDS:
            raise KeyError("Invalid column: " + str(field))

        session_label = self.trading_calendar.minute_to_session_label(dt)

        if dt < asset.start_date or \
                (data_frequency == "daily" and
                    session_label > asset.end_date) or \
                (data_frequency == "minute" and
                 session_label > asset.end_date):
            if field == "volume":
                return 0
            elif field != "last_traded":
                return np.NaN

        if data_frequency == "daily":
            return self._get_daily_data(asset, field, session_label)
        else:
            if field == "last_traded":
                return self.get_last_traded_dt(asset, dt, 'minute')
            elif field == "price":
                return self._get_minute_spot_value(asset, "close", dt,
                                                   ffill=True)
            else:
                return self._get_minute_spot_value(asset, field, dt)

    def get_adjustments(self, assets, field, dt, perspective_dt):
        """
        Returns a list of adjustments between the dt and perspective_dt for the
        given field and list of assets

        Parameters
        ----------
        assets : list of type Asset, or Asset
            The asset, or assets whose adjustments are desired.
        field : {'open', 'high', 'low', 'close', 'volume', \
                 'price', 'last_traded'}
            The desired field of the asset.
        dt : pd.Timestamp
            The timestamp for the desired value.
        perspective_dt : pd.Timestamp
            The timestamp from which the data is being viewed back from.
        data_frequency : str
            The frequency of the data to query; i.e. whether the data is
            'daily' or 'minute' bars

        Returns
        -------
        adjustments : list[Adjustment]
            The adjustments to that field.
        """
        if isinstance(assets, Asset):
            assets = [assets]

        adjustment_ratios_per_asset = []
        split_adj_factor = lambda x: x if field != 'volume' else 1.0 / x

        for asset in assets:
            adjustments_for_asset = []
            split_adjustments = self._get_adjustment_list(
                asset, self._splits_dict, "SPLITS"
            )
            for adj_dt, adj in split_adjustments:
                if dt <= adj_dt <= perspective_dt:
                    adjustments_for_asset.append(split_adj_factor(adj))
                elif adj_dt > perspective_dt:
                    break

            if field != 'volume':
                merger_adjustments = self._get_adjustment_list(
                    asset, self._mergers_dict, "MERGERS"
                )
                for adj_dt, adj in merger_adjustments:
                    if dt <= adj_dt <= perspective_dt:
                        adjustments_for_asset.append(adj)
                    elif adj_dt > perspective_dt:
                        break

                dividend_adjustments = self._get_adjustment_list(
                    asset, self._dividends_dict, "DIVIDENDS",
                )
                for adj_dt, adj in dividend_adjustments:
                    if dt <= adj_dt <= perspective_dt:
                        adjustments_for_asset.append(adj)
                    elif adj_dt > perspective_dt:
                        break

            ratio = reduce(mul, adjustments_for_asset, 1.0)
            adjustment_ratios_per_asset.append(ratio)

        return adjustment_ratios_per_asset

    def get_adjusted_value(self, asset, field, dt,
                           perspective_dt,
                           data_frequency,
                           spot_value=None):
        """
        Returns a scalar value representing the value
        of the desired asset's field at the given dt with adjustments applied.

        Parameters
        ----------
        asset : Asset
            The asset whose data is desired.
        field : {'open', 'high', 'low', 'close', 'volume', \
                 'price', 'last_traded'}
            The desired field of the asset.
        dt : pd.Timestamp
            The timestamp for the desired value.
        perspective_dt : pd.Timestamp
            The timestamp from which the data is being viewed back from.
        data_frequency : str
            The frequency of the data to query; i.e. whether the data is
            'daily' or 'minute' bars

        Returns
        -------
        value : float, int, or pd.Timestamp
            The value of the given ``field`` for ``asset`` at ``dt`` with any
            adjustments known by ``perspective_dt`` applied. The return type is
            based on the ``field`` requested. If the field is one of 'open',
            'high', 'low', 'close', or 'price', the value will be a float. If
            the ``field`` is 'volume' the value will be a int. If the ``field``
            is 'last_traded' the value will be a Timestamp.
        """
        if spot_value is None:
            # if this a fetcher field, we want to use perspective_dt (not dt)
            # because we want the new value as of midnight (fetcher only works
            # on a daily basis, all timestamps are on midnight)
            if self._is_extra_source(asset, field,
                                     self._augmented_sources_map):
                spot_value = self.get_spot_value(asset, field, perspective_dt,
                                                 data_frequency)
            else:
                spot_value = self.get_spot_value(asset, field, dt,
                                                 data_frequency)

        if isinstance(asset, Equity):
            ratio = self.get_adjustments(asset, field, dt, perspective_dt)[0]
            spot_value *= ratio

        return spot_value

    def _get_minute_spot_value(self, asset, column, dt, ffill=False):
        reader = self._get_pricing_reader('minute')
        try:
            result = reader.get_value(
                asset.sid, dt, column
            )
        except NoDataOnDate:
            if not ffill:
                if column == 'volume':
                    return 0
                else:
                    return np.nan

        if not ffill:
            return result

        # we are looking for price, and didn't find one. have to go hunting.
        last_traded_dt = reader.get_last_traded_dt(asset, dt)

        if last_traded_dt is pd.NaT:
            # no last traded dt, bail
            if column == 'volume':
                return 0
            else:
                return np.nan

        # get the value as of the last traded dt
        result = reader.get_value(
            asset.sid,
            last_traded_dt,
            column
        )

        if np.isnan(result):
            return np.nan

        if dt == last_traded_dt or dt.date() == last_traded_dt.date():
            return result

        # the value we found came from a different day, so we have to adjust
        # the data if there are any adjustments on that day barrier
        return self.get_adjusted_value(
            asset, column, last_traded_dt,
            dt, "minute", spot_value=result
        )

    def _get_daily_data(self, asset, column, dt):
        reader = self._get_pricing_reader('daily')
        if column == "last_traded":
            last_traded_dt = reader.get_last_traded_dt(asset, dt)

            if pd.isnull(last_traded_dt):
                return pd.NaT
            else:
                return last_traded_dt
        elif column in OHLCV_FIELDS:
            # don't forward fill
            try:
                val = reader.get_value(asset, dt, column)
                if val == -1:
                    if column == "volume":
                        return 0
                    else:
                        return np.nan
                else:
                    return val
            except NoDataOnDate:
                return np.nan
        elif column == "price":
            found_dt = dt
            while True:
                try:
                    value = reader.get_value(
                        asset, found_dt, "close"
                    )
                    if value != -1:
                        if dt == found_dt:
                            return value
                        else:
                            # adjust if needed
                            return self.get_adjusted_value(
                                asset, column, found_dt, dt, "minute",
                                spot_value=value
                            )
                    else:
                        found_dt -= self.trading_calendar.day
                except NoDataOnDate:
                    return np.nan

    @remember_last
    def _get_days_for_window(self, end_date, bar_count):
        tds = self.trading_calendar.all_sessions
        end_loc = tds.get_loc(end_date)
        start_loc = end_loc - bar_count + 1
        if start_loc < self._first_trading_day_loc:
            raise HistoryWindowStartsBeforeData(
                first_trading_day=self._first_trading_day.date(),
                bar_count=bar_count,
                suggested_start_day=tds[
                    self._first_trading_day_loc + bar_count
                ].date(),
            )
        return tds[start_loc:end_loc + 1]

    def _get_history_daily_window(self, assets, end_dt, bar_count,
                                  field_to_use):
        """
        Internal method that returns a dataframe containing history bars
        of daily frequency for the given sids.
        """
        session = self.trading_calendar.minute_to_session_label(end_dt)
        days_for_window = self._get_days_for_window(session, bar_count)

        if len(assets) == 0:
            return pd.DataFrame(None,
                                index=days_for_window,
                                columns=None)

        data = self._get_history_daily_window_data(
            assets, days_for_window, end_dt, field_to_use
        )
        return pd.DataFrame(
            data,
            index=days_for_window,
            columns=assets
        )

    def _get_history_daily_window_data(
            self, assets, days_for_window, end_dt, field_to_use):
        ends_at_midnight = end_dt.hour == 0 and end_dt.minute == 0

        if ends_at_midnight:
            # two cases where we use daily data for the whole range:
            # 1) the history window ends at midnight utc.
            # 2) the last desired day of the window is after the
            # last trading day, use daily data for the whole range.
            return self._get_daily_window_for_sids(
                assets,
                field_to_use,
                days_for_window,
                extra_slot=False
            )
        else:
            # minute mode, requesting '1d'
            daily_data = self._get_daily_window_for_sids(
                assets,
                field_to_use,
                days_for_window[0:-1]
            )

            if field_to_use == 'open':
                minute_value = self._daily_aggregator.opens(
                    assets, end_dt)
            elif field_to_use == 'high':
                minute_value = self._daily_aggregator.highs(
                    assets, end_dt)
            elif field_to_use == 'low':
                minute_value = self._daily_aggregator.lows(
                    assets, end_dt)
            elif field_to_use == 'close':
                minute_value = self._daily_aggregator.closes(
                    assets, end_dt)
            elif field_to_use == 'volume':
                minute_value = self._daily_aggregator.volumes(
                    assets, end_dt)

            # append the partial day.
            daily_data[-1] = minute_value

            return daily_data

    def _handle_minute_history_out_of_bounds(self, bar_count):
        first_trading_minute_loc = (
            self.trading_calendar.all_minutes.get_loc(
                self._first_trading_minute
            )
            if self._first_trading_minute is not None else None
        )

        suggested_start_day = (
            self.trading_calendar.all_minutes[
                first_trading_minute_loc + bar_count
            ] + self.trading_calendar.day
        ).date()

        raise HistoryWindowStartsBeforeData(
            first_trading_day=self._first_trading_day.date(),
            bar_count=bar_count,
            suggested_start_day=suggested_start_day,
        )

    def _get_history_minute_window(self, assets, end_dt, bar_count,
                                   field_to_use):
        """
        Internal method that returns a dataframe containing history bars
        of minute frequency for the given sids.
        """
        # get all the minutes for this window
        try:
            minutes_for_window = self.trading_calendar.minutes_window(
                end_dt, -bar_count
            )
        except KeyError:
            self._handle_minute_history_out_of_bounds(bar_count)

        if minutes_for_window[0] < self._first_trading_minute:
            self._handle_minute_history_out_of_bounds(bar_count)

        asset_minute_data = self._get_minute_window_for_assets(
            assets,
            field_to_use,
            minutes_for_window,
        )

        return pd.DataFrame(
            asset_minute_data,
            index=minutes_for_window,
            columns=assets
        )

    def get_history_window(self, assets, end_dt, bar_count, frequency, field,
                           ffill=True):
        """
        Public API method that returns a dataframe containing the requested
        history window.  Data is fully adjusted.

        Parameters
        ----------
        assets : list of zipline.data.Asset objects
            The assets whose data is desired.

        bar_count: int
            The number of bars desired.

        frequency: string
            "1d" or "1m"

        field: string
            The desired field of the asset.

        ffill: boolean
            Forward-fill missing values. Only has effect if field
            is 'price'.

        Returns
        -------
        A dataframe containing the requested data.
        """
        if field not in OHLCVP_FIELDS:
            raise ValueError("Invalid field: {0}".format(field))

        if frequency == "1d":
            if field == "price":
                df = self._get_history_daily_window(assets, end_dt, bar_count,
                                                    "close")
            else:
                df = self._get_history_daily_window(assets, end_dt, bar_count,
                                                    field)
        elif frequency == "1m":
            if field == "price":
                df = self._get_history_minute_window(assets, end_dt, bar_count,
                                                     "close")
            else:
                df = self._get_history_minute_window(assets, end_dt, bar_count,
                                                     field)
        else:
            raise ValueError("Invalid frequency: {0}".format(frequency))

        # forward-fill price
        if field == "price":
            if frequency == "1m":
                data_frequency = 'minute'
            elif frequency == "1d":
                data_frequency = 'daily'
            else:
                raise Exception(
                    "Only 1d and 1m are supported for forward-filling.")

            dt_to_fill = df.index[0]

            perspective_dt = df.index[-1]
            assets_with_leading_nan = np.where(pd.isnull(df.iloc[0]))[0]
            for missing_loc in assets_with_leading_nan:
                asset = assets[missing_loc]
                previous_dt = self.get_last_traded_dt(
                    asset, dt_to_fill, data_frequency)
                if pd.isnull(previous_dt):
                    continue
                previous_value = self.get_adjusted_value(
                    asset,
                    field,
                    previous_dt,
                    perspective_dt,
                    data_frequency,
                )
                df.iloc[0, missing_loc] = previous_value

            df.fillna(method='ffill', inplace=True)

            for asset in df.columns:
                if df.index[-1] >= asset.end_date:
                    # if the window extends past the asset's end date, set
                    # all post-end-date values to NaN in that asset's series
                    series = df[asset]
                    series[series.index.normalize() > asset.end_date] = np.NaN

        return df

    def _get_minute_window_for_assets(self, assets, field, minutes_for_window):
        """
        Internal method that gets a window of adjusted minute data for an asset
        and specified date range.  Used to support the history API method for
        minute bars.

        Missing bars are filled with NaN.

        Parameters
        ----------
        asset : Asset
            The asset whose data is desired.

        field: string
            The specific field to return.  "open", "high", "close_price", etc.

        minutes_for_window: pd.DateTimeIndex
            The list of minutes representing the desired window.  Each minute
            is a pd.Timestamp.

        Returns
        -------
        A numpy array with requested values.
        """
        return self._get_minute_window_data(assets, field, minutes_for_window)

    def _get_minute_window_data(
            self, assets, field, minutes_for_window):
        return self._minute_history_loader.history(assets,
                                                   minutes_for_window,
                                                   field,
                                                   False)

    def _get_daily_window_for_sids(
            self, assets, field, days_in_window, extra_slot=True):
        """
        Internal method that gets a window of adjusted daily data for a sid
        and specified date range.  Used to support the history API method for
        daily bars.

        Parameters
        ----------
        asset : Asset
            The asset whose data is desired.

        start_dt: pandas.Timestamp
            The start of the desired window of data.

        bar_count: int
            The number of days of data to return.

        field: string
            The specific field to return.  "open", "high", "close_price", etc.

        extra_slot: boolean
            Whether to allocate an extra slot in the returned numpy array.
            This extra slot will hold the data for the last partial day.  It's
            much better to create it here than to create a copy of the array
            later just to add a slot.

        Returns
        -------
        A numpy array with requested values.  Any missing slots filled with
        nan.

        """
        bar_count = len(days_in_window)
        # create an np.array of size bar_count
        if extra_slot:
            return_array = np.zeros((bar_count + 1, len(assets)))
        else:
            return_array = np.zeros((bar_count, len(assets)))

        if field != "volume":
            # volumes default to 0, so we don't need to put NaNs in the array
            return_array[:] = np.NAN

        if bar_count != 0:
            data = self._history_loader.history(assets,
                                                days_in_window,
                                                field,
                                                extra_slot)
            if extra_slot:
                return_array[:len(return_array) - 1, :] = data
            else:
                return_array[:len(data)] = data
        return return_array

    def _get_adjustment_list(self, asset, adjustments_dict, table_name):
        """
        Internal method that returns a list of adjustments for the given sid.

        Parameters
        ----------
        asset : Asset
            The asset for which to return adjustments.

        adjustments_dict: dict
            A dictionary of sid -> list that is used as a cache.

        table_name: string
            The table that contains this data in the adjustments db.

        Returns
        -------
        adjustments: list
            A list of [multiplier, pd.Timestamp], earliest first

        """
        if self._adjustment_reader is None:
            return []

        sid = int(asset)

        try:
            adjustments = adjustments_dict[sid]
        except KeyError:
            adjustments = adjustments_dict[sid] = self._adjustment_reader.\
                get_adjustments_for_sid(table_name, sid)

        return adjustments

    def _check_is_currently_alive(self, asset, dt):
        sid = int(asset)

        if sid not in self._asset_start_dates:
            self._get_asset_start_date(asset)

        start_date = self._asset_start_dates[sid]
        if self._asset_start_dates[sid] > dt:
            raise NoTradeDataAvailableTooEarly(
                sid=sid,
                dt=normalize_date(dt),
                start_dt=start_date
            )

        end_date = self._asset_end_dates[sid]
        if self._asset_end_dates[sid] < dt:
            raise NoTradeDataAvailableTooLate(
                sid=sid,
                dt=normalize_date(dt),
                end_dt=end_date
            )

    def _get_asset_start_date(self, asset):
        self._ensure_asset_dates(asset)
        return self._asset_start_dates[asset]

    def _get_asset_end_date(self, asset):
        self._ensure_asset_dates(asset)
        return self._asset_end_dates[asset]

    def _ensure_asset_dates(self, asset):
        sid = int(asset)

        if sid not in self._asset_start_dates:
            if self._first_trading_day is not None:
                self._asset_start_dates[sid] = \
                    max(asset.start_date, self._first_trading_day)
            else:
                self._asset_start_dates[sid] = asset.start_date

            self._asset_end_dates[sid] = asset.end_date

    def get_splits(self, sids, dt):
        """
        Returns any splits for the given sids and the given dt.

        Parameters
        ----------
        sids : container
            Sids for which we want splits.
        dt : pd.Timestamp
            The date for which we are checking for splits. Note: this is
            expected to be midnight UTC.

        Returns
        -------
        splits : list[(int, float)]
            List of splits, where each split is a (sid, ratio) tuple.
        """
        if self._adjustment_reader is None or not sids:
            return {}

        # convert dt to # of seconds since epoch, because that's what we use
        # in the adjustments db
        seconds = int(dt.value / 1e9)

        splits = self._adjustment_reader.conn.execute(
            "SELECT sid, ratio FROM SPLITS WHERE effective_date = ?",
            (seconds,)).fetchall()

        splits = [split for split in splits if split[0] in sids]

        return splits

    def get_stock_dividends(self, sid, trading_days):
        """
        Returns all the stock dividends for a specific sid that occur
        in the given trading range.

        Parameters
        ----------
        sid: int
            The asset whose stock dividends should be returned.

        trading_days: pd.DatetimeIndex
            The trading range.

        Returns
        -------
        list: A list of objects with all relevant attributes populated.
        All timestamp fields are converted to pd.Timestamps.
        """

        if self._adjustment_reader is None:
            return []

        if len(trading_days) == 0:
            return []

        start_dt = trading_days[0].value / 1e9
        end_dt = trading_days[-1].value / 1e9

        dividends = self._adjustment_reader.conn.execute(
            "SELECT * FROM stock_dividend_payouts WHERE sid = ? AND "
            "ex_date > ? AND pay_date < ?", (int(sid), start_dt, end_dt,)).\
            fetchall()

        dividend_info = []
        for dividend_tuple in dividends:
            dividend_info.append({
                "declared_date": dividend_tuple[1],
                "ex_date": pd.Timestamp(dividend_tuple[2], unit="s"),
                "pay_date": pd.Timestamp(dividend_tuple[3], unit="s"),
                "payment_sid": dividend_tuple[4],
                "ratio": dividend_tuple[5],
                "record_date": pd.Timestamp(dividend_tuple[6], unit="s"),
                "sid": dividend_tuple[7]
            })

        return dividend_info

    def contains(self, asset, field):
        return field in BASE_FIELDS or \
            (field in self._augmented_sources_map and
             asset in self._augmented_sources_map[field])

    def get_fetcher_assets(self, dt):
        """
        Returns a list of assets for the current date, as defined by the
        fetcher data.

        Returns
        -------
        list: a list of Asset objects.
        """
        # return a list of assets for the current date, as defined by the
        # fetcher source
        if self._extra_source_df is None:
            return []

        day = normalize_date(dt)

        if day in self._extra_source_df.index:
            assets = self._extra_source_df.loc[day]['sid']
        else:
            return []

        if isinstance(assets, pd.Series):
            return [x for x in assets if isinstance(x, Asset)]
        else:
            return [assets] if isinstance(assets, Asset) else []

    # cache size picked somewhat loosely.  this code exists purely to
    # handle deprecated API.
    @weak_lru_cache(20)
    def _get_minute_count_for_transform(self, ending_minute, days_count):
        # This function works in three steps.
        # Step 1. Count the minutes from ``ending_minute`` to the start of its
        #         session.
        # Step 2. Count the minutes from the prior ``days_count - 1`` sessions.
        # Step 3. Return the sum of the results from steps (1) and (2).

        # Example (NYSE Calendar)
        #     ending_minute = 2016-12-28 9:40 AM US/Eastern
        #     days_count = 3
        # Step 1. Calculate that there are 10 minutes in the ending session.
        # Step 2. Calculate that there are 390 + 210 = 600 minutes in the prior
        #         two sessions. (Prior sessions are 2015-12-23 and 2015-12-24.)
        #         2015-12-24 is a half day.
        # Step 3. Return 600 + 10 = 610.

        cal = self.trading_calendar

        ending_session = cal.minute_to_session_label(
            ending_minute,
            direction="none",  # It's an error to pass a non-trading minute.
        )

        # Assume that calendar days are always full of contiguous minutes,
        # which means we can just take 1 + (number of minutes between the last
        # minute and the start of the session). We add one so that we include
        # the ending minute in the total.
        ending_session_minute_count = timedelta_to_integral_minutes(
            ending_minute - cal.open_and_close_for_session(ending_session)[0]
        ) + 1

        if days_count == 1:
            # We just need sessions for the active day.
            return ending_session_minute_count

        # XXX: We're subtracting 2 here to account for two offsets:
        # 1. We only want ``days_count - 1`` sessions, since we've already
        #    accounted for the ending session above.
        # 2. The API of ``sessions_window`` is to return one more session than
        #    the requested number.  I don't think any consumers actually want
        #    that behavior, but it's the tested and documented behavior right
        #    now, so we have to request one less session than we actually want.
        completed_sessions = cal.sessions_window(
            cal.previous_session_label(ending_session),
            2 - days_count,
        )

        completed_sessions_minute_count = (
            self.trading_calendar.minutes_count_for_sessions_in_range(
                completed_sessions[0],
                completed_sessions[-1]
            )
        )
        return ending_session_minute_count + completed_sessions_minute_count

    def get_simple_transform(self, asset, transform_name, dt, data_frequency,
                             bars=None):
        if transform_name == "returns":
            # returns is always calculated over the last 2 days, regardless
            # of the simulation's data frequency.
            hst = self.get_history_window(
                [asset], dt, 2, "1d", "price", ffill=True
            )[asset]

            return (hst.iloc[-1] - hst.iloc[0]) / hst.iloc[0]

        if bars is None:
            raise ValueError("bars cannot be None!")

        if data_frequency == "minute":
            freq_str = "1m"
            calculated_bar_count = int(self._get_minute_count_for_transform(
                dt, bars
            ))
        else:
            freq_str = "1d"
            calculated_bar_count = bars

        price_arr = self.get_history_window(
            [asset], dt, calculated_bar_count, freq_str, "price", ffill=True
        )[asset]

        if transform_name == "mavg":
            return nanmean(price_arr)
        elif transform_name == "stddev":
            return nanstd(price_arr, ddof=1)
        elif transform_name == "vwap":
            volume_arr = self.get_history_window(
                [asset], dt, calculated_bar_count, freq_str, "volume",
                ffill=True
            )[asset]

            vol_sum = nansum(volume_arr)

            try:
                ret = nansum(price_arr * volume_arr) / vol_sum
            except ZeroDivisionError:
                ret = np.nan

            return ret