Repository with the solution from the paper "Robust and Accurate Period Inference using Regression-Based Techniques".

• General info
• Trace generator
• Feature extractor
• Regression algorithms

This project contains three major functionalities:

• A trace generator which creates a dataset of traces.
• A feature extractor which creates a dataset of features from a given set of traces.
• A suite of tree-based regression algorithms which can be trained and then used to predict the periods of the tasks within traces.

Randomly generates a set of traces according to the user's specifications. The result is a csv with two columns: Trace and Periods. A trace is a list of task identifiers denoted as integers from 1 to the total number of tasks. The IDs are assigned to tasks according to their periods (from lowest to highest). The periods are also stored in a list with their order matching the task IDs.

• Python version: 3.7
• simpy version: 2.3.1
• numpy version: 1.19.2
• pandas version: 1.1.3
• tqdm version: 4.50.2

python trace_generation.py [dataset] [dataset size] [no tasks] [utilization] [alpha] [jitter] [preemptive]

• dataset:
  • automotive with periods generated according to [1].
  • loguniform with periods generated according to a log-uniform distribution.
• dataset size: the number of traces to be generated.
• no tasks: the number of tasks within a trace.
• utilization: the total uitlization of the system given as a fractional number.
• alpha: the amount of execution time variation given as a fractional number.
• jitter: the amount with release jitter w.r.t. the period of th task given as a fractional number.
• preemptive: whether or not the execution model is preemptive.
  • yes for preemptive execution.
  • no for non-preemptive execution.

Both the trace and the task periods are saved as strings representing lists such that they can be easily parsed when read by the Feature extractor.

python trace_generation.py automotive 300 4 0.7 0 0 yes

The script will generate a csv file containing 300 traces, each trace with 4 tasks and a total utilization of 70%. Also, there is no execution time variation, no release jitter and the execution is preemptive.

The output file will be named IDEAL_automotive_4_tasks_0.7_utilization_1602426197.csv meaning that the execution is without uncertainties (IDEAL), with other distinguishing parameters mentioned in the name and ending with the timestamp.

Receives as input a set of traces under the format specified in Trace generator and creates a set of features which can be used either for training or for prediction.

• Python version: 3.7
• numpy version: 1.19.2
• pandas version: 1.1.3
• tqdm version: 4.50.2

python feature_extraction.py [path to dataset] [no features] [training]

• path to dataset: the path to the data set of traces.
• no features (recommended 20): the number of features to be extracted from the two signal processing techniques, meaning that the number of resulting features will be double the specified value.
• training: whether or not the data set of traces will be used for training or for prediction.
  • yes for creating a training set of features, meaning that this set will also include the true period for each task. Thus, the list of periods must be specified in the data set of traces.
  • no for creating a set of features for prediction. Here, the true periods may be omitted.

The script will generate a csv file containing the features extracted from the projection of each task from each trace. Also, the resulted file will contain the lower bound and the upper bound required to use the space pruning method.

python feature_extraction.py python feature_extraction.py /Users/myuser/period_inference/IDEAL_automotive_4_tasks_0.7_utilization_1602426197.csv 20 yes

The result will be a file called features_training_1602426752.csv denoting that this data set incorporates the true periods in it. In case the the training parameter would be no, then the result would be features_testing_1602426752.csv.

Receives as input a set of features and generates either a trained model or a set of predictions.

• R version: 4.0.1
• rJava library version: 0.9-12
• CARET library version: 6.0-86
• Cubist library version: 0.2.3
• extraTrees library version: 1.0.5
• bartMachine library version: 1.2.4.2

Rscript run_algorithm.r [algorithm] [path to dataset] [no features regression] [no features period adjustment] [use spm] [path to model]

• algorithm: the regression algorithm to be used.
  • cubist [2]
  • extraTrees [3]
  • gbm [4]
  • bartMachine [5]
• path to dataset: the path to the data set containing the features to be used either for training or for prediction
• no features regression: the number of features from each signal processing technique used for training the regression model.
• no features period adjustment: the number of features from each signal processing technique used for performing the period adjustment step.
• use spm: whether or not to use the space pruning method.
  • yes for using the space pruning method. Here, the input data set must contain the Lower_bound and Upper_bound columns specified in this order after the columns with features.
  • no the script will only perform the period adjustment step.
• path to model: the path to the trained regression model. In case of training this argument should not be introduced.

NOTE: The total numnber of features in the data set must be 2 x no features period adjustment.

When training, the result will be a rds file with the following naming format: model_[algorithm].rds.

When predicting, the result will be a csv file containing the predicted period for each task represented by its set of features. The csv will include 3 columns:

• RPM: the predictions coming directly from the regression algorithms.
• RPMPA: the predictions from the period adjustment step.
• SPM (optional): the predictions from the space pruning method.

• Training: Rscript run_algorithm.r cubist /Users/myuser/period_inference/features_training_1602426752.csv 3 20

The model will be created using 6 features in total: top 3 periods from periodogram and top 3 features from autocorrelation.

• Predicting: Rscript run_algorithm.r cubist /Users/myuser/period_inference/features_testing_1602426752.csv 3 20 yes /Users/myuser/period_inference/model_cubist.rds

The predictions will include the results from space pruning too.

[1] Kramer, S., Ziegenbein, D., & Hamann, A. (2015, July). Real world automotive benchmarks for free. In 6th International Workshop on Analysis Tools and Methodologies for Embedded and Real-time Systems (WATERS).

[2] Max Kuhn and Ross Quinlan (2020). Cubist: Rule- And Instance-Based Regression Modeling. R package version 0.2.3. https://CRAN.R-project.org/package=Cubist.

[3] Simm J, de Abril I, Sugiyama M (2014). Tree-Based Ensemble Multi-Task Learning Method for Classification and Regression, volume 97 number 6. <URL: http://CRAN.R-project.org/package=extraTrees.

[4] Brandon Greenwell, Bradley Boehmke, Jay Cunningham and GBM Developers (2019). gbm: Generalized Boosted Regression Models. R package version 2.1.5. https://CRAN.R-project.org/package=gbm.

[5] Adam Kapelner, Justin Bleich (2016). bartMachine: Machine Learning with Bayesian Additive Regression Trees. Journal of Statistical Software, 70(4), 1-40. doi:10.18637/jss.v070.i04. https://CRAN.R-project.org/package=bartMachine.