def get_history_window_series(self,
assets,
end_dt,
bar_count,
field,
data_frequency,
trailing_bar_count=None,
reset_reader=False):
start_dt = get_start_dt(end_dt, bar_count, data_frequency, False)
start_dt, _ = self.get_adj_dates(start_dt, end_dt, assets,
data_frequency)
if trailing_bar_count:
delta = get_delta(trailing_bar_count, data_frequency)
end_dt += delta
# This is an attempt to resolve some caching with the reader
# when auto-ingesting data.
# TODO: needs more work
reader = self.get_reader(data_frequency)
if reset_reader:
del self._readers[reader._rootdir]
reader = self.get_reader(data_frequency)
if reader is None:
symbols = [asset.symbol for asset in assets]
raise PricingDataNotLoadedError(
field=field,
first_trading_day=min([asset.start_date for asset in assets]),
exchange=self.exchange_name,
symbols=symbols,
symbol_list=','.join(symbols),
data_frequency=data_frequency,
start_dt=start_dt,
end_dt=end_dt)
series = dict()
for asset in assets:
asset_start_dt, _ = self.get_adj_dates(start_dt, end_dt, assets,
data_frequency)
in_bundle = range_in_bundle(asset, asset_start_dt, end_dt, reader)
if not in_bundle:
raise PricingDataNotLoadedError(
field=field,
first_trading_day=asset.start_date,
exchange=self.exchange_name,
symbols=asset.symbol,
symbol_list=asset.symbol,
data_frequency=data_frequency,
start_dt=asset_start_dt,
end_dt=end_dt)
periods = self.get_calendar_periods_range(asset_start_dt, end_dt,
data_frequency)
# This does not behave well when requesting multiple assets
# when the start or end date of one asset is outside of the range
# looking at the logic in load_raw_arrays(), we are not achieving
# any performance gain by requesting multiple sids at once. It's
# looping through the sids and making separate requests anyway.
arrays = reader.load_raw_arrays(sids=[asset.sid],
fields=[field],
start_dt=start_dt,
end_dt=end_dt)
if len(arrays) == 0:
raise DataCorruptionError(exchange=self.exchange_name,
symbols=asset.symbol,
start_dt=asset_start_dt,
end_dt=end_dt)
field_values = arrays[0][:, 0]
try:
value_series = pd.Series(field_values, index=periods)
series[asset] = value_series
except ValueError as e:
raise PricingDataValueError(exchange=asset.exchange,
symbol=asset.symbol,
start_dt=asset_start_dt,
end_dt=end_dt,
error=e)
return series
def get_history_window(self,
assets,
end_dt,
bar_count,
frequency,
field,
data_frequency=None,
ffill=True):
"""
Public API method that returns a dataframe containing the requested
history window. Data is fully adjusted.
Parameters
----------
assets : list of catalyst.data.Asset objects
The assets whose data is desired.
end_dt: not applicable to cryptocurrencies
bar_count: int
The number of bars desired.
frequency: string
"1d" or "1m"
field: string
The desired field of the asset.
data_frequency: string
The frequency of the data to query; i.e. whether the data is
'daily' or 'minute' bars.
# TODO: fill how?
ffill: boolean
Forward-fill missing values. Only has effect if field
is 'price'.
Returns
-------
A dataframe containing the requested data.
"""
freq_match = re.match(r'([0-9].*)(m|M|d|D)', frequency, re.M | re.I)
if freq_match:
candle_size = int(freq_match.group(1))
unit = freq_match.group(2)
else:
raise InvalidHistoryFrequencyError(frequency)
if unit.lower() == 'd':
if data_frequency == 'minute':
data_frequency = 'daily'
elif unit.lower() == 'm':
if data_frequency == 'daily':
data_frequency = 'minute'
else:
raise InvalidHistoryFrequencyError(frequency)
adj_bar_count = candle_size * bar_count
try:
series = self.bundle.get_history_window_series_and_load(
assets=assets,
end_dt=end_dt,
bar_count=adj_bar_count,
field=field,
data_frequency=data_frequency)
except PricingDataNotLoadedError:
series = dict()
for asset in assets:
if asset not in series or series[asset].index[-1] < end_dt:
# Adding bars too recent to be contained in the consolidated
# exchanges bundles. We go directly against the exchange
# to retrieve the candles.
start_dt = get_start_dt(end_dt, adj_bar_count, data_frequency)
trailing_dt = \
series[asset].index[-1] + get_delta(1, data_frequency) \
if asset in series else start_dt
trailing_bar_count = \
get_periods(trailing_dt, end_dt, data_frequency)
# The get_history method supports multiple asset
candles = self.get_candles(data_frequency=data_frequency,
assets=asset,
bar_count=trailing_bar_count,
end_dt=end_dt)
last_value = series[asset].iloc(0) if asset in series \
else np.nan
candle_series = self.get_series_from_candles(
candles=candles,
start_dt=trailing_dt,
end_dt=end_dt,
field=field,
previous_value=last_value)
if asset in series:
series[asset].append(candle_series)
else:
series[asset] = candle_series
df = pd.DataFrame(series)
if candle_size > 1:
if field == 'open':
agg = 'first'
elif field == 'high':
agg = 'max'
elif field == 'low':
agg = 'min'
elif field == 'close':
agg = 'last'
elif field == 'volume':
agg = 'sum'
else:
raise ValueError('Invalid field.')
df = df.resample('{}T'.format(candle_size)).agg(agg)
return df
def get_history_window_with_bundle(self,
assets,
end_dt,
bar_count,
frequency,
field,
data_frequency=None,
ffill=True,
force_auto_ingest=False):
"""
Public API method that returns a dataframe containing the requested
history window. Data is fully adjusted.
Parameters
----------
assets : list[TradingPair]
The assets whose data is desired.
end_dt: datetime
The date of the last bar.
bar_count: int
The number of bars desired.
frequency: string
"1d" or "1m"
field: string
The desired field of the asset.
data_frequency: string
The frequency of the data to query; i.e. whether the data is
'daily' or 'minute' bars.
# TODO: fill how?
ffill: boolean
Forward-fill missing values. Only has effect if field
is 'price'.
Returns
-------
DataFrame
A dataframe containing the requested data.
"""
freq, candle_size, unit, data_frequency = get_frequency(
frequency, data_frequency)
adj_bar_count = candle_size * bar_count
try:
series = self.bundle.get_history_window_series_and_load(
assets=assets,
end_dt=end_dt,
bar_count=adj_bar_count,
field=field,
data_frequency=data_frequency,
force_auto_ingest=force_auto_ingest)
except (PricingDataNotLoadedError, NoDataAvailableOnExchange):
series = dict()
for asset in assets:
if asset not in series or series[asset].index[-1] < end_dt:
# Adding bars too recent to be contained in the consolidated
# exchanges bundles. We go directly against the exchange
# to retrieve the candles.
start_dt = get_start_dt(end_dt, adj_bar_count, data_frequency)
trailing_dt = \
series[asset].index[-1] + get_delta(1, data_frequency) \
if asset in series else start_dt
# The get_history method supports multiple asset
# Use the original frequency to let each api optimize
# the size of result sets
trailing_bar_count = get_periods(trailing_dt, end_dt, freq)
candles = self.get_candles(freq=freq,
assets=asset,
bar_count=trailing_bar_count,
start_dt=start_dt,
end_dt=end_dt)
last_value = series[asset].iloc(0) if asset in series \
else np.nan
# Create a series with the common data_frequency, ffill
# missing values
candle_series = self.get_series_from_candles(
candles=candles,
start_dt=trailing_dt,
end_dt=end_dt,
data_frequency=data_frequency,
field=field,
previous_value=last_value)
if asset in series:
series[asset].append(candle_series)
else:
series[asset] = candle_series
df = resample_history_df(pd.DataFrame(series), freq, field)
# TODO: consider this more carefully
df.dropna(inplace=True)
return df
def get_history_window(self,
assets,
end_dt,
bar_count,
frequency,
field,
data_frequency=None,
is_current=False):
"""
Public API method that returns a dataframe containing the requested
history window. Data is fully adjusted.
Parameters
----------
assets : list[TradingPair]
The assets whose data is desired.
end_dt: datetime
The date of the last bar
bar_count: int
The number of bars desired.
frequency: string
"1d" or "1m"
field: string
The desired field of the asset.
data_frequency: string
The frequency of the data to query; i.e. whether the data is
'daily' or 'minute' bars.
is_current: bool
Skip date filters when current data is requested (last few bars
until now).
Notes
-----
Catalysts requires an end data with bar count both CCXT wants a
start data with bar count. Since we have to make calculations here,
we ensure that the last candle match the end_dt parameter.
Returns
-------
DataFrame
A dataframe containing the requested data.
"""
freq, candle_size, unit, data_frequency = get_frequency(
frequency, data_frequency)
adj_bar_count = candle_size * bar_count
start_dt = get_start_dt(end_dt, adj_bar_count, data_frequency)
# The get_history method supports multiple asset
candles = self.get_candles(
freq=freq,
assets=assets,
bar_count=bar_count,
start_dt=start_dt if not is_current else None,
end_dt=end_dt if not is_current else None,
)
series = dict()
for asset in candles:
asset_series = self.get_series_from_candles(
candles=candles[asset],
start_dt=start_dt,
end_dt=end_dt,
data_frequency=frequency,
field=field,
)
if end_dt is not None:
delta = get_delta(candle_size, data_frequency)
adj_end_dt = end_dt - delta
last_traded = asset_series.index[-1]
if last_traded < adj_end_dt:
raise LastCandleTooEarlyError(
last_traded=last_traded,
end_dt=adj_end_dt,
exchange=self.name,
)
series[asset] = asset_series
df = pd.DataFrame(series)
df.dropna(inplace=True)
return df
def ingest_df(self,
ohlcv_df,
data_frequency,
asset,
writer,
empty_rows_behavior='strip'):
"""
Ingest a DataFrame of OHLCV data for a given market.
Parameters
----------
ohlcv_df: DataFrame
data_frequency: str
asset: TradingPair
writer:
empty_rows_behavior: str
"""
if empty_rows_behavior is not 'ignore':
nan_rows = ohlcv_df[ohlcv_df.isnull().T.any().T].index
if len(nan_rows) > 0:
dates = []
previous_date = None
for row_date in nan_rows.values:
row_date = pd.to_datetime(row_date)
if previous_date is None:
dates.append(row_date)
else:
seq_date = previous_date + get_delta(1, data_frequency)
if row_date > seq_date:
dates.append(previous_date)
dates.append(row_date)
previous_date = row_date
dates.append(pd.to_datetime(nan_rows.values[-1]))
name = '{} from {} to {}'.format(asset.symbol,
ohlcv_df.index[0],
ohlcv_df.index[-1])
if empty_rows_behavior == 'warn':
log.warn(
'\n{name} with end minute {end_minute} has empty rows '
'in ranges: {dates}'.format(
name=name,
end_minute=asset.end_minute,
dates=dates))
elif empty_rows_behavior == 'raise':
raise EmptyValuesInBundleError(name=name,
end_minute=asset.end_minute,
dates=dates)
else:
ohlcv_df.dropna(inplace=True)
data = []
if not ohlcv_df.empty:
ohlcv_df.sort_index(inplace=True)
data.append((asset.sid, ohlcv_df))
self._write(data, writer, data_frequency)
def ingest_ctable(self,
asset,
data_frequency,
period,
start_dt,
end_dt,
writer,
empty_rows_behavior='strip',
cleanup=False):
"""
Merge a ctable bundle chunk into the main bundle for the exchange.
:param asset: TradingPair
:param data_frequency: str
:param period: str
:param writer:
:param empty_rows_behavior: str
Ensure that the bundle does not have any missing data.
:param cleanup: bool
Remove the temp bundle directory after ingestion.
:return:
"""
path = get_bcolz_chunk(exchange_name=self.exchange.name,
symbol=asset.symbol,
data_frequency=data_frequency,
period=period)
reader = self.get_reader(data_frequency, path=path)
if reader is None:
raise TempBundleNotFoundError(path=path)
arrays = reader.load_raw_arrays(
sids=[asset.sid],
fields=['open', 'high', 'low', 'close', 'volume'],
start_dt=start_dt,
end_dt=end_dt)
if not arrays:
return path
periods = self.get_calendar_periods_range(start_dt, end_dt,
data_frequency)
df = get_df_from_arrays(arrays, periods)
if empty_rows_behavior is not 'ignore':
nan_rows = df[df.isnull().T.any().T].index
if len(nan_rows) > 0:
dates = []
previous_date = None
for row_date in nan_rows.values:
row_date = pd.to_datetime(row_date)
if previous_date is None:
dates.append(row_date)
else:
seq_date = previous_date + get_delta(1, data_frequency)
if row_date > seq_date:
dates.append(previous_date)
dates.append(row_date)
previous_date = row_date
dates.append(pd.to_datetime(nan_rows.values[-1]))
name = path.split('/')[-1]
if empty_rows_behavior == 'warn':
log.warn(
'\n{name} with end minute {end_minute} has empty rows '
'in ranges: {dates}'.format(
name=name,
end_minute=asset.end_minute,
dates=dates))
elif empty_rows_behavior == 'raise':
raise EmptyValuesInBundleError(name=name,
end_minute=asset.end_minute,
dates=dates)
else:
df.dropna(inplace=True)
data = []
if not df.empty:
df.sort_index(inplace=True)
data.append((asset.sid, df))
self._write(data, writer, data_frequency)
if cleanup:
log.debug('removing bundle folder following '
'ingestion: {}'.format(path))
shutil.rmtree(path)
return path