Tests and Proofs by Unknown

Author:Unknown
Language: eng
Format: epub
ISBN: 9783030311575
Publisher: Springer International Publishing

Table 2.R-values for different feature sets and learning methods.

Method

Features

R-value

linear regression without preprocessing

Msg, #ActiveMsgs, #Subs

0.6943

linear regression with preprocessing

Msg, #ActiveMsgs, #Subs

0.8567

DNNs with 3 features

Msg, #ActiveMsgs, #Subs

0.8585

DNNs with 8 features

Msg, #ActiveMsgs, #TotalSubs, TopicSize, MsgSize, publish, #Subs, #Receivers

0.9145

Figure 5 shows the predicted compared to the measured latency values of the multiple linear regression and deep learning with and without the feature set reduction. The diagonal dashed line represents a complete match between predicted and true values. It can be seen that for deep learning the data points are much closer to the diagonal line. Note that the values in Fig. 5a do not include all latency values, because of the removed outliers.

Fitting the multiple linear regression requires 285.1357 ms and a prediction needs 0.0176 ms on average, while the neural network requires 862609.8151 ms (14 min) to train and 0.8468 ms on average to predict. These experiments were performed on a MacBook Pro with an Intel Core i5-6360U 2 GHz dual core CPU and 8 GB of RAM running macOS 10.14.1.

Evaluation of our timed model. Another important factor in addition to the quality of our neural network is how well the resulting timed model performs for our predictions with SMC. In order to analyse the accuracy of our model, we compare our predictions to the actual observations at the broker, and we show the difference to our previously applied linear regression method.

Fig. 6.Simulation results with 50, 70 & 90 clients for predictions with a regression, a neural network and the real prob. of the SUT for test-case length 1–3.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.