Echoes From Space - Grouping Commands with Large Scale Telemetry Data

Published at ICSE 2018 at the Software Engineering in Practice track, in Gothenburg Sweden (ACM Paper Link)

This repository contains the source code, poster, and presentation accompanying the paper.

Alexandros Lattas¹, Diomedis Spinellis^2
1Imperial College London, ²Athens University of Economics and Business

Source Code

The source code for all optimization algorithms used in paper can be found in the notebook:

code/cmd-optimize.ipynb

To pre-process anonymized telemetry such as the ones used in the experiments use:

python code/preprocess_data.py path/to/file.txt

Summary

The purpose of this study is to create a simple User Interaction Model to evaluate alternative approaches to optimizing a user interface.

Model

$T=M(S,T_C,T_M,C_A,C_G)$ Where $S$ is a command execution scenario expressed as a vector of commands, $T_C$ is the time required to switch between commands of the same group, $T_M$ is the time required to switch between commands of different groups, $C_A$ is a set of commands always available at the desktop, $C_G$ is a set of command groups, and T is the time required to execute.

All approaches can be expressed by varying $C_A$ and $C_G$. A user's expertise level is expressed through $T_C$ and $T_M$.

Calculating the Time Needed

In this simplified version of the User Interaction Model, we assert that only 2 weights are needed:

Same group: $T_s$ = command1 to command2 Different group: $T_d$ = command1 change-group command2

Approaches to be evaluated

• ALL Put all commands on the screen (actually possible with new extra wide LCD screens). This establishes the theoretical optimum.
• NAIVE Naive: make the most frequently executed commands always available.
• GROUP Group heuristics: group together the most used common commands (e.g. delete and move) by doing away the class-verb method for selecting them.
• MRU-B User optimized, based on the most recently used (MRU) commands of a particular user executes more frequently during a training period (batch).
• MRU-O User optimized, based on the most recently used (MRU) commands of a particular user executes more frequently during continuously adjusted (online).
• OPT Optimized: run a (stochastic) optimization algorithm to select those commands whose combination yields the fastest user interaction time. This differs from NAIVE in that it takes into account the actual switching between entities.
• CLUSTER Clustered, based on 2-3 clusters of users. The clustering is performed based on the commands they execute. A stochastic optimization similar to OPT is ran to select the commands executed in each cluster.

Data Format

The file format is: init_time, click-on-group_name/click-on-command_group-name_command-name. For example:

1497807650.013000 en_objects
1497807651.223000 vm_object_add
1497807655.943000 vm_object_get_attributes

The data used in these experiments are proprietary and are not made public in this repository.

Poster Overview