def DeriveChange(data_frame, params=None):
logger.debug(
"initialized DeriveChange. Use get_params() to see parameter values")
if params == None:
params = {}
params["func_params"] = {
"window_len": ("length of averaging window", 1, False),
"angle_change":
("if the change is between angles, we return the signed smaller angle between the two headings",
False, False)
}
params["measure_rules"] = {
"target_measure":
("name of the target measure", "measure_name", True),
"output_name": ("name of returned measure", "output_name", True)
}
return params
logger.debug("transforming data to %s" %
(params["measure_rules"]["output_name"]))
if "window_len" in params["func_params"]:
window_len = params["func_params"]["window_len"]
else:
window_len = 1
target_array = data_frame[params["measure_rules"]["target_measure"]].values
diffed_data = diff_data(target_array, window_len,
params["func_params"]["angle_change"])
return pd.DataFrame(data=diffed_data,
columns=[params["measure_rules"]["output_name"]],
index=data_frame.index[:-1])
def ButterLowpass(data_frame, params=None):
"""Class to apply a Butterworth lowpass filter to data"""
logger.debug("Calculating ButterLowpass.")
if params == None:
params = {
"order": ("order of the filter", 2, True),
"nyquist": ("Wn parameter from scipy.signal.butter", 0.05, True),
"filter_name":
("name to append to all filtered parameters", "_buttered", False)
}
return params
elif "filter_name" not in params:
params["filter_name"] = "_buttered"
logger.debug("transforming_data")
buttered_data = np.zeros(data_frame.shape)
output_names = []
for col_ind, df_col in enumerate(data_frame):
buttered_data[:, col_ind] = butter_data(data_frame[df_col].values,
params["order"],
params["nyquist"])
output_names.append(df_col + params["filter_name"])
return pd.DataFrame(data=buttered_data,
columns=output_names,
index=data_frame.index)
def compare_threshold(data_array,
comparison_operator,
comparision_val,
absolute_compare=False):
"""
Fucntion for comparing an array to a values with a binary operator
:param data_array: input data
:type data_array: numpy array
:param comparison_operator: string representation of the binary operator for comparison
:type comparison_operator: str
:param comparision_val: The value to be compared against
:type comparision_val: float
:param absolute_compare: specifying whether to compare raw value or absolute value
:type absolute_compare: Boolean
:return: the indices where the binary operator is true
:rtype: numpy array
"""
logger.debug("comparing: %s %d" % (comparison_operator, comparision_val))
if absolute_compare:
data_array = np.abs(data_array)
comparisons = {
"==": np.equal,
"!=": np.not_equal,
">=": np.greater_equal,
"<=": np.less_equal,
">": np.greater,
"<": np.less
}
cur_comp = comparisons[comparison_operator]
match_inds = cur_comp(np.nan_to_num(data_array), comparision_val)
return match_inds
def select_transform(transform_type, transform_name):
"""Method to grab the transform function from the correct module"""
logger.debug("Selecting transform function")
available_transforms = {}
available_transforms["filter_data"] = {
"ButterLowpass": ButterLowpass,
"WindowAverage": WindowAverage,
"dummy": dummy_function
}
available_transforms["derive_param"] = {
"DeriveSlope": DeriveSlope,
"DeriveChange": DeriveChange,
"DeriveCumsum": DeriveCumsum,
"DeriveDistance": DeriveDistance,
"DeriveHeading": DeriveHeading,
"DeriveWindowSum": DeriveWindowSum,
"DeriveScaled": DeriveScaled,
"DeriveInBox": DeriveInBox,
"DeriveThreshold": DeriveThreshold,
"DeriveLogicalCombination": DeriveLogicalCombination,
}
available_transforms["detect_event"] = {
"DetectThreshold": DetectThreshold
}
return available_transforms[transform_type][transform_name]
def __init__(self, template, dstreams):
logger.debug("init BStream")
super().__init__()
self.dstreams = dstreams
self["template_id"] = template["template_id"]
self._load_from_dict(template)
self["stream_token"] = str(template["stream_token"])
def process_data(self, dstream_list, token):
"""
Wrapper method for asynchronously processing data.
:param dstream_list: list of dstreams with raw data
:type dstream_list: list of dicts
:param token: stream token
:type token: string
"""
logger.debug("process_data_async")
st = time.time()
# retrieve most recent versioned dstream template
template = dstream_list[0]
# create bstream for dstream list
bstream = self._list_to_bstream(template, dstream_list)
# filter bstream data
bstream.apply_filters()
# apply derived param transforms
bstream.apply_dparam_rules()
# apply event transforms
bstream.find_events()
# post events to server
self._post_parsed_events(bstream)
self._post_dataframe(bstream["stream_token"], bstream["measures"])
print("whoop WHOOOOP", time.time() - st, len(bstream["timestamp"]))
def __init__(self):
"""
Initializes and empty DStream with a unique stream token. All the other expected keys are
initialized as empty data structures of the desired type.
"""
self["stream_name"] = None
self["user_description"] = None
self["version"] = 0
self["stream_token"] = str(uuid.uuid1())
self["source_key"] = None
self["template_id"] = str(uuid.uuid1())
self["storage_rules"] = {}
self["ingest_rules"] = {}
self["engine_rules"] = {}
self["timestamp"] = None
self["measures"] = {}
self["fields"] = {}
self["user_ids"] = {}
self["tags"] = {}
self["foreign_keys"] = []
self["filters"] = []
self["dparam_rules"] = []
self["event_rules"] = {}
self["data_rules"] = {}
logger.debug("DStream initialize")
def DeriveLogicalCombination(data_frame, params=None):
logger.debug("Starting DeriveLogicalCombination")
if params == None:
params = {}
params["func_params"] = {
"combiner":
("Either AND or OR string specifying which operator to use", "AND",
True)
}
params["measure_rules"] = {
"first_measure":
("name of first measure to be combined", "measure_name", True),
"second_measure":
("name of second measure to be combined", "measure_name", True),
"output_name": ("name of returned measure", "output_name", True)
}
logger.debug("Combining {} and {} data to {}".format(
params["measure_rules"]["first_measure"],
params["measure_rules"]["second_measure"],
params["measure_rules"]["output_name"]))
first_array = data_frame[params["measure_rules"]["first_measure"]].values
second_array = data_frame[params["measure_rules"]["second_measure"]].values
print(first_array, second_array)
combined = logical_combine(first_array, second_array,
params["func_params"]["combiner"])
return pd.DataFrame(data=combined,
columns=[params["measure_rules"]["output_name"]],
index=data_frame.index)
def partition_data(self, list_of_partitions, logical_comparison="AND"):
"""This function takes a list of tuples of partition parameters used by partition_rows() and
returns all rows from the measure DataFrame that meet the logical AND or logical OR of those conditions"""
logger.debug("building parition rows")
if logical_comparison == "AND":
start_bools = np.ones((self["measures"].shape[0], ), dtype=bool)
elif logical_comparison == "OR":
start_bools = np.zeros((self["measures"].shape[0], ), dtype=bool)
else:
raise ValueError("{} is not a supported logical comparison".format(
logical_comparison))
for partition in list_of_partitions:
new_inds = self.partition_rows(partition[0], partition[1],
partition[2])
if logical_comparison == "AND":
start_bools = np.logical_and(start_bools, new_inds)
elif logical_comparison == "OR":
start_bools = np.logical_or(start_bools, new_inds)
else:
raise ValueError(
"{} is not a supported logical comparison".format(
logical_comparison))
return self["measures"][start_bools]
def DeriveThreshold(data_frame, params=None):
logger.debug("Starting DeriveThreshold")
if params == None:
params = {}
params["func_params"] = {
"threshold_value": ("value to compare against", 0, True),
"comparison_operator": ("one of == != >= <= > <", "==", True),
"absolute_compare":
("whether to compare against absolute value instead of raw value",
False, False)
}
params["measure_rules"] = {
"target_measure":
("name of the target measure", "measure_name", True),
"output_name": ("name of returned measure", "output_name", True)
}
logger.debug("transforming data to %s" %
(params["measure_rules"]["output_name"]))
target_array = data_frame[params["measure_rules"]["target_measure"]].values
threshold_bool = compare_threshold(
target_array, params["func_params"]["comparison_operator"],
params["func_params"]["threshold_value"],
params["func_params"]["absolute_compare"])
return pd.DataFrame(data=threshold_bool,
columns=[params["measure_rules"]["output_name"]],
index=data_frame.index)
def DeriveWindowSum(data_frame, params=None):
logger.debug("Starting DeriveWindowSum.")
if params == None:
params = {}
params["func_params"] = {
"window_len": ("window size for summing", 2, True)
}
params["measure_rules"] = {
"target_measure":
("name of the target measure", "measure_name", True),
"output_name": ("name of returned measure", "output_name", True)
}
return params
logger.debug("transforming data to %s" %
(params["measure_rules"]["output_name"]))
if "window_len" in params["func_params"]:
window_len = params["func_params"]["window_len"]
else:
window_len = 1
target_array = data_frame[params["measure_rules"]["target_measure"]].values
summed_data = window_sum(target_array, window_len)
return pd.DataFrame(data=summed_data,
columns=[params["measure_rules"]["output_name"]],
index=data_frame.index)
def DeriveInBox(data_frame, params=None):
logger.debug("Starting DeriveInBox.")
if params == None:
params = {}
params["func_params"] = {
"upper_left_corner":
("location of upper left corner", (0, 1), True),
"lower_right_corner":
("location of lower right corner", (1, 0), True)
}
params["measure_rules"] = {
"spatial_measure":
("name of geo-spatial measure", "measure_name", True),
"output_name": ("name of returned measure", "measure_name", True)
}
return params
logger.debug("transforming data to %s" %
(params["measure_rules"]["output_name"]))
position_array = pd.DataFrame(
data_frame[params["measure_rules"]["spatial_measure"]].tolist()).values
box_bool = in_box(position_array,
params["func_params"]["upper_left_corner"],
params["func_params"]["lower_right_corner"])
return pd.DataFrame(data=box_bool,
columns=[params["measure_rules"]["output_name"]],
index=data_frame.index)
def DeriveSlope(data_frame, params=None):
logger.debug("Starting DeriveSlope.")
if params == None:
params = {}
params["func_params"] = {
"window_len": ("length of averaging window", 1, False)
}
params["measure_rules"] = {
"rise_measure":
("measure y values (or rise in rise/run calculation of slope)",
"measure_name", True),
"run_measure":
("measure containing x values (or run in rise/run calculation of slope)",
"measure_name", True),
"output_name": ("name of returned measure", "output_name", True)
}
return params
logger.debug("transforming data to %s" %
(params["measure_rules"]["output_name"]))
if "window_len" in params["func_params"]:
window_len = params["func_params"]["window_len"]
else:
window_len = 1
xrun = data_frame[params["measure_rules"]["run_measure"]].values
yrise = data_frame[params["measure_rules"]["rise_measure"]].values
smaller_len = np.min([xrun.shape[0], yrise.shape[0]])
sloped = sloper(yrise[:smaller_len, ], xrun[:smaller_len, ], window_len)
return pd.DataFrame(data=sloped,
columns=[params["measure_rules"]["output_name"]],
index=data_frame.index)
def bearing(position_array, window_len, units="deg"):
"""
Calculates the angle between lat lon points
:param position_array: input vector of lat lon points
:type position_array: N x 2 numpy array
:param window_len: Length of window for averaging
:type window_len: int
:param units: String for output units. Currently 'mi' and 'km' supported
:type units: str
:return: the angle between consecutive latlon points
:rtype: (N - 1) x 1 numpy array
"""
logger.debug("finding bearing of vector")
lat1 = position_array[:-1, 0]
lat2 = position_array[1:, 0]
lon1 = position_array[:-1, 1]
lon2 = position_array[1:, 1]
lon1, lat1, lon2, lat2 = map(np.radians, [lon1, lat1, lon2, lat2])
dlon = lon1 - lon2
first_val = np.sin(dlon) * np.cos(lat2)
second_val = np.cos(lat1) * np.sin(lat2) - np.sin(lat1) * np.cos(
lat2) * np.cos(dlon)
cur_bear = np.arctan2(first_val, second_val)
if units == "deg":
cur_bear = (np.rad2deg(cur_bear) + 180.0) % 360 - 180
if window_len > 1:
cur_bear = window_data(cur_bear, window_len)
return cur_bear
def great_circle(position_array, window_len=1, units="mi"):
"""
Function to calculate the great circle distance between consecutive samples in lat lon vector
:param position_array: input vector of lat lon points
:type position_array: N x 2 numpy array
:param window_len: length of window for averaging output
:type window_len: int
:param units: String for output units. Currently 'mi' and 'km' supported
:type units: str
:return: distances between consecutive points
:rtype: (N-1) x 1 numpy array
"""
logger.debug("calculating great circle distance")
lat1 = position_array[:-1, 0]
lat2 = position_array[1:, 0]
lon1 = position_array[:-1, 1]
lon2 = position_array[1:, 1]
lon1, lat1, lon2, lat2 = map(np.radians, [lon1, lat1, lon2, lat2])
dlat = lat1 - lat2
dlon = lon1 - lon2
inner_val = np.sin(
dlat / 2.0)**2 + np.cos(lat1) * np.cos(lat2) * np.sin(dlon / 2.0)**2
outer_val = 2 * np.arcsin(np.sqrt(inner_val))
if units == "mi":
earth_diameter = 3959
elif units == "km":
earth_diameter = 6371
great_dist = outer_val * earth_diameter
if window_len > 1:
great_dist = window_data(great_dist, window_len)
return great_dist
def _aggregate_uids(self):
logger.debug("aggregating uids")
uids = [s["user_ids"] for s in self.dstreams]
self["user_ids"] = {
uidkey: [i[uidkey] for i in uids]
for uidkey, v in self["user_ids"].items()
}
def consume(self):
""" """
tk['Consumer.consume : self.consumer.start'].start()
self.consumer.start() #auto-start
tk['Consumer.consume : self.consumer.start'].stop()
for msg in self.consumer:
if msg is not None:
logger.debug(str(msg.value) + ": {}".format(msg.offset))
def find_events(self):
logger.debug("finding events")
self["events"] = {}
for event_name, event_rule in self["event_rules"].items():
# print("finding event {}".format(event_name))
self["events"][event_name] = self.apply_transform(
event_rule["partition_list"], event_rule["measure_list"],
event_rule["transform_type"], event_rule["transform_name"],
event_rule["param_dict"], event_rule["logical_comparison"])
def aggregate(self):
logger.debug("aggregating everything")
self._aggregate_uids()
self._aggregate_ts()
self._aggregate_fields()
self._aggregate_tags()
self._measure_df()
return self
def get(self, option, section=None, default=None):
try:
if section is None:
option, section = self.__get_option_name(option)
value = self._cfg.get(section, option)
except (NoSectionError, NoOptionError, ConfigParserGeneralError) as err:
logger.info("Configuration parameter didn't exist, returning the default value." % err.message)
return default
logger.debug("Read configuration parameter: (section=%s) %s=%s" % (section, option, value))
return value
def load_from_json(self, json_file):
"""
The standard method for loading data from an existing json dict.
:param json_file: the json dict containing data to be loaded into our DStream
:type json_file: dict
"""
for key in json_file.keys():
if key != 'stream_token' and key != 'template_id':
self[key] = json_file[key]
logger.debug("added key %s" % (key))
def apply_dparam_rules(self):
logger.debug("deriving parameters")
self["derived_measures"] = {}
for dparam_rule in self["dparam_rules"]:
# print("deriving {}".format(dparam_rule["param_dict"]["measure_rules"]["output_name"]))
self.apply_transform(dparam_rule["partition_list"],
dparam_rule["measure_list"],
dparam_rule["transform_type"],
dparam_rule["transform_name"],
dparam_rule["param_dict"],
dparam_rule["logical_comparison"])
def apply_filters(self):
logger.debug("applying filters")
self["filter_measures"] = {}
for filter_rule in self["filters"]:
# logger.debug("applying filter {}".format(filter_rule["param_dict"]["filter_name"]))
self.apply_transform(filter_rule["partition_list"],
filter_rule["measure_list"],
filter_rule["transform_type"],
filter_rule["transform_name"],
filter_rule["param_dict"],
filter_rule["logical_comparison"])
def cumsum(data_array, offset=0):
"""
Calculate the cumulative sum of a vector
:param data_array: data to be summed
:type data_array: numpy array
:param offset: starting value for sum
:type offset: float
:return: the cumulative sum of the data_array
:rtype: numpy array
"""
logger.debug("cumsum")
return np.cumsum(data_array) + offset
def produce(self, dmsg):
""" Produce to given topic w/ partition_key and log e. 1k msg.
:param dmsg: Message to produce
"""
tk['Producer.produce'].start()
bcount = str(self.count).encode()
tk['Producer.produce : self.producer.produce'].start()
self.producer.produce(dmsg, partition_key=bcount)
tk['Producer.produce : self.producer.produce'].stop()
logger.debug("Just produced a message")
self.count += 1
tk['Producer.produce'].stop()
def _post_events(event_data):
"""
Sends post request containing event data to API
:param event_data: event data (individual event)
:type event_data: dict
:return: request status
:rtype: string
"""
endpoint = 'http://{}:{}/new_event'.format(config['server_host'],
config['server_port'])
logger.debug(event_data)
r = requests.post(endpoint, json=event_data)
return {'request_status': r.status_code}
def _measure_df(self):
logger.debug("aggregating into DataFrame")
all_measures = [s["measures"] for s in self.dstreams]
self["measures"] = {
m: [i[m]['val'] for i in all_measures]
for m, v in self["measures"].items()
}
self["measures"]["timestamp"] = self["timestamp"]
for user_id, value in self["user_ids"].items():
self["measures"][user_id] = value
self["measures"]["tags"] = self["tags"]
self["measures"]["fields"] = self["fields"]
self["measures"] = pd.DataFrame.from_dict(self["measures"])
def euclidean_dist(position_array, window_len=1):
"""
Function to calculate euclidean distance between consecutive samples in a positional vector
:param position_array: input vector of positions
:type position_array: N x 2 numpy array
:param window_len: length of window for averaging output
:type window_len: int
:return: distances between consecutive points
:rtype: (N-1) x 1 numpy array
"""
logger.debug("calculating euclidean distance")
euclid_array = np.sqrt(np.sum(np.diff(position_array, axis=0)**2, axis=1))
if window_len > 1:
euclid_array = window_data(euclid_array, window_len)
return euclid_array
def logical_combine(array1, array2, combiner):
"""
Function for creating elementwise AND or OR of two vectors
:param array1: First array for combination
:type array1: (n,1) numpy array
:param array2: Second array for combination
:type array2: (n,1) numpy array
:param combiner: "AND" or "OR" specifying which method to combine
:type combiner: str
:return: Boolean array combining the two inputs
:rtype: (n,1) boolean numpy array
"""
logger.debug(f"Combining with{combiner}")
combining_func = {"AND": np.logical_and, "OR": np.logical_or}
return combining_func[combiner](array1, array2)
def _post_dataframe(stream_token, dataframe):
"""
Sends post request containing event data to API
:param event_data: event data (individual event)
:type event_data: dict
:return: request status
:rtype: string
"""
endpoint = 'http://{}:{}/data_storage'.format(config['server_host'],
config['server_port'])
logger.debug(dataframe)
r = requests.post(endpoint,
data=pickle.dumps((stream_token, dataframe)))
return {'request_status': r.status_code}
