def get_merged_dataframes(self, feature_requests, start_dt, end_dt, repopulate=False):
"""
Return a TimeseriesFeature by sticker for start_dt <= t < end_dt
:param feature_requests:
:return: {sticker: DataFrame} where Dataframe has all the features merged
"""
start_dt = parse_date_if_necessary(start_dt, to_utc=True)
end_dt = parse_date_if_necessary(end_dt, to_utc=True)
if not isinstance(feature_requests, list):
feature_requests = [feature_requests]
self._instantiate_features(feature_requests)
for request in feature_requests:
feature_id = request.feature_id
self.feature_map[feature_id].initialize(start_dt, end_dt)
features_df = list()
for request in feature_requests:
feature_id = request.feature_id
feature = self.feature_map[feature_id]
if not isinstance(feature, InfiniteTimeSeriesFeature):
raise ValueError("Not a InfiniteTimeSeriesFeature")
logging.info("Requesting {} {} {}".format(feature_id, start_dt, end_dt))
df = feature.get_df(start_dt, end_dt, repopulate=repopulate)
df.columns = [request.prefix + c for c in df.columns]
features_df.append(df)
merged = merge_timeseries_features(features_df)
return merged
def get_dataframes_by_stickers(self,
feature_requests,
stickers,
repopulate=False):
"""
Return a TimeserieFeature by sticker
:param feature_requests:
:param stickers: with or without bookmakers
:return: {sticker: DataFrame} where Dataframe has all the features merged
"""
self._instantiate_features(feature_requests)
for request in feature_requests:
feature_id = request.feature_id
self.feature_map[feature_id].initialize_stickers(stickers)
ret = {}
# TODO parallelize
for sticker in stickers:
features_df = list()
for request in feature_requests:
feature_id = request.feature_id
feature = self.feature_map[feature_id]
if not isinstance(feature, TimeSeriesFeature):
raise ValueError("Not a TimeSeriesFeature")
logging.info("Requesting {} {}".format(feature_id, sticker))
df = feature.get_df_from_sticker(sticker,
repopulate=repopulate)
df.columns = [request.prefix + c for c in df.columns]
features_df.append(df)
ret[sticker] = merge_timeseries_features(features_df)
return ret
def compute_for_sticker_q(feature_requests, stickers, feature_map,
result_queue):
ret = dict()
for sticker in stickers:
features_df = list()
for request in feature_requests:
feature_id = request.feature_id
# logging.info("Requesting {} {}".format(feature_id, sticker))
features_df.append(
feature_map[feature_id].get_df_from_sticker(sticker))
ret[sticker] = merge_timeseries_features(features_df)
result_queue.put(ret)
def get_dataframes_by_event_ids(self,
feature_requests,
event_ids,
repopulate=False,
recompute_missing=True):
"""
Return a TimeserieFeature by event_ids
:param feature_requests:
:param event_ids: list of strings like ['GSM55554', 'GSM55555'] or ['ENP7777', 'ENP7778']
:return: {event_id: DataFrame} where Dataframe has all the features merged
"""
self._instantiate_features(feature_requests)
for request in feature_requests:
feature_id = request.feature_id
self.feature_map[feature_id].initialize_events(event_ids)
ret = {}
for event_id in event_ids:
features_df = list()
for request in feature_requests:
feature_id = request.feature_id
feature = self.feature_map[feature_id]
if not isinstance(feature, TimeSeriesFeature):
raise ValueError("Not a TimeSeriesFeature")
logging.info("Requesting {} {}".format(feature_id, event_id))
request_df = feature.get_df_from_event_id(
event_id,
repopulate=repopulate,
recompute_missing=recompute_missing)
features_df.append(request_df)
ret[event_id] = merge_timeseries_features(features_df)
return ret
def merge_timeseries_features(cls, dfs):
return merge_timeseries_features(dfs)