Module reference

Benchmark class

A class that manages the database entries for the measured metrics which are logged into the database.

Parameters

db_filestr

The path of the database file

descriptionstr, optional

The description of the whole pipeline use case. Even though the description is optional, it should be set so the database entries are distinguishable without evaluating the uuid’s. This parameter is ignored for Benchmark objects initialized in mode ‘r’.

modestr, default=’a’

One of [‘w’, ‘a’, ‘r’]. The mode corresponds to conventional python file handling modes. Modes ‘a’ and ‘w’ are used for storing metrics in a database during a pipeline run and ‘r’ is used for querying metrics from the database.

Attributes

db_filestr

path to the database file

modestr

mode of the Benchmark object. One of [‘w’, ‘a’, ‘r’].

descriptionstr

description of the pipeline run. Not relevant if mode is ‘r’.

sessionsqlalchemy.orm.session.Session

SQLalchemy session

Metrics

Supervised metrics

class umlaut.BenchmarkSupervisor(metrics, benchmark)

A supervisor object managing all supervised metrics

This object should be used as a decorator.

metrics: list of Metric

A list of metrics to be collected while running the decorated function

benchmark: Benchmark

The central benchmark object used in the pipeline

class umlaut.CPUMetric(description, interval=1)

The metric object to measure CPU usage of the running Python instance in percent

description: str

The description of this metric and function which is added to the database

interval: int, default=1

The number of seconds between CPU usage measurements

class umlaut.ThroughputMetric(description)

The metric object to measure throughput of the pipeline function

To pass the number of data points processed, use method track()

description: str

The description of this metric and function which is added to the database

class umlaut.LatencyMetric(description)

The metric object to measure latency of the pipeline function

To pass the number of data points processed, use method track()

description: str

The description of this metric and function which is added to the database

class umlaut.PowerMetric(description, interval=1)

The metric object to measure power used in the execution

description: str

The description of this metric and function which is added to the database

interval: int, default=1

The number of seconds between memory measurements

class umlaut.EnergyMetric(description)

The metric object to measure energy used in the execution

description: str

The description of this metric and function which is added to the database

class umlaut.MemoryMetric(description, interval=1)

The metric object to measure memory used in the execution

description: str

The description of this metric and function which is added to the database

interval: int, default=1

The number of seconds between memory measurements

class umlaut.TimeMetric(description)

The metric object to measure the time taken for the execution

description: str

The description of this metric and function which is added to the database

Tracked metrics

class umlaut.LossTracker(benchmark)
serialize(loss_values)
loss_valueslist of ints

List of tracked loss values of the run.

pickle object

Serialized data.

track(loss_values, description)
loss_valueslist of ints

List of tracked loss values of the run.

descriptionstr

Description of tracked loss.

class umlaut.TTATracker(benchmark)
serialize(accuracies)
accuracieslist of ints

List of tracked accuracies of the run.

pickle object

Serialized data.

track(accuracies, description)
accuracieslist of ints

List of tracked accuracies of the run.

descriptionstr

Description of tracked TTA.

class umlaut.HyperparameterTracker(benchmark, description, hyperparameters, target, low_means_good=True)
serialize()
pickle object

Serialized data.

track(measurement)
measurementlist of str

List of hyperparameters

class umlaut.ConfusionMatrixTracker(benchmark)
serialize(matrix, labels)
matrixlist of list of ints

Tracker matrix values.

labelslist of str

Class labels of confustion matrix.

pickle object

Serialized data.

track(matrix, labels, description)

Pass an ndarray where axis 0 is predicted and axis 1 is actual.

matrixlist of list of ints

Tracker matrix values.

labelslist of str

Class labels of confusion matrix.

description:

Description of tracked confusion matrix.