def plot_forecast(model_output, queue):
"""
Plots output from R forecast. Uses R functions extracted via rpy2
(see global vars above).
Parameters
----------
model_output
R object returned by leptoid.forecasting._forecast_utilization(). Contains
forecast information (confidence intervals, etc.).
queue
leptoid.ServiceQueue with information about the service
Returns nothing, but saves a plot to disk with a timestamp.
"""
forecast_method = get_forecast_attribute(model_output, attr='method')
n = datetime.datetime.now()
RBMP('%s%s-%s-%i-%i-%i-%i:%i.jpg' %
(PLOT_DIRECTORY, queue.service, queue.instance_id, n.year,
n.month, n.day, n.hour, n.minute),
width=1400, height=800, units='px', type='jpeg')
RPLOT(model_output, main="Util forecast for %s:%s (using %s)" %
(queue.service, queue.instance_id, forecast_method),
xlab="Minutes elapsed since %s" % queue.get_first_timestamp())
RDEVOFF()
def plot_forecast(model_output, queue):
"""
Plots output from R forecast. Uses R functions extracted via rpy2
(see global vars above).
Parameters
----------
model_output
R object returned by leptoid.forecasting._forecast_utilization(). Contains
forecast information (confidence intervals, etc.).
queue
leptoid.ServiceQueue with information about the service
Returns nothing, but saves a plot to disk with a timestamp.
"""
forecast_method = get_forecast_attribute(model_output, attr='method')
n = datetime.datetime.now()
RBMP('%s%s-%s-%i-%i-%i-%i:%i.jpg' %
(PLOT_DIRECTORY, queue.service, queue.instance_id, n.year, n.month,
n.day, n.hour, n.minute),
width=1400,
height=800,
units='px',
type='jpeg')
RPLOT(model_output,
main="Util forecast for %s:%s (using %s)" %
(queue.service, queue.instance_id, forecast_method),
xlab="Minutes elapsed since %s" % queue.get_first_timestamp())
RDEVOFF()
def forecast(queue):
"""
Forecasting values using R's forecast package. Series reporting empty data
over the last ${recent_data_window} minutes will return None.
Parameters
----------
queue
pandas.TimeSeries with utilization data. Data is extracted from the
underlying buffer using np.frombuffer.
service, instance_id
information for instance, used when plotting
TODO: Currently Leptoid uses R via rpy2. This will change since rpy2 is
poorly maintained.
"""
LOG.info("Generating forecast for %s:%s" %
(queue.service, queue.instance_id))
# Retrieving forecast from rpy2, then extracting the attributes
# (forecasted utilization, one-step ahead forecast) we want.
model_output = _forecast_utilization(np.frombuffer(queue.utilization.data))
# Handling case where output == None, indicating a dormant instance.
if model_output == None:
in_sample_forecast = None
util_estimate = None
else:
in_sample_forecast = get_forecast_attribute(model_output, "fitted")
util_estimate = get_forecast_attribute(model_output, "mean")
# Let's avoid empty plots
if np.max(util_estimate) > PLOT_SIGNIFICANCE_THRESHOLD:
plot_forecast(model_output, queue)
return (in_sample_forecast, util_estimate)
def forecast(queue):
"""
Forecasting values using R's forecast package. Series reporting empty data
over the last ${recent_data_window} minutes will return None.
Parameters
----------
queue
pandas.TimeSeries with utilization data. Data is extracted from the
underlying buffer using np.frombuffer.
service, instance_id
information for instance, used when plotting
TODO: Currently Leptoid uses R via rpy2. This will change since rpy2 is
poorly maintained.
"""
LOG.info("Generating forecast for %s:%s" %
(queue.service, queue.instance_id))
# Retrieving forecast from rpy2, then extracting the attributes
# (forecasted utilization, one-step ahead forecast) we want.
model_output = _forecast_utilization(np.frombuffer(queue.utilization.data))
# Handling case where output == None, indicating a dormant instance.
if model_output == None:
in_sample_forecast = None
util_estimate = None
else:
in_sample_forecast = get_forecast_attribute(model_output, "fitted")
util_estimate = get_forecast_attribute(model_output, "mean")
# Let's avoid empty plots
if np.max(util_estimate) > PLOT_SIGNIFICANCE_THRESHOLD:
plot_forecast(model_output, queue)
return (in_sample_forecast, util_estimate)
