Update documentation

docs/elephants.rst

@@ -9,17 +9,17 @@ The `elephants` subpackage provides functionalities related to elephant flows mo
 
 The `simulate` tool performs simulations at the packet level. It reads flow statistics from a histogram file or generates flows based on a provided JSON mixture, and generates sample packets for these flows. Simulations can be repeated to obtain confidence intervals. On the other hand, the `calculate`}` tool can analytically calculate flow table occupancy reduction curves based on provided flow records, histograms, or mixtures.
 
-Tools
-=====
+Programs
+========
 
 .. toctree::
    :glob:
    :maxdepth: 1
 
    elephants/*
 
-Examples for scikit-learn
-=========================
+Example scripts for scikit-learn
+================================
 
 Additionally, we provide the `elephants.skl` subpackage, which offers examples on how to use the scikit-learn library to train machine learning algorithms for detecting elephant flows based on the first packet. None of previous works analyze metrics such as flow table reduction or the amount of traffic transmitted after flow classification, which we believe are crucial from the perspective of traffic engineering and QoS. These studies primarily focus on classification accuracy, measured by parameters like true positive rate, true negative rate, and accuracy of flow size and duration prediction. They provide limited insight into the effectiveness of the analyzed algorithms in our specific application. For example, misclassifying the largest flow in the network has a much greater impact on the change in traffic coverage than misclassifying a small flow. The metrics presented so far do not account for this difference. Our proposal is to use novel metrics for evaluating ML algorithms in the context of elephant flow detection, specifically flow table occupancy reduction and fraction of traffic covered. There is a tradeoff between these two metrics: increasing the elephant detection threshold leads to greater flow table reduction but decreases the fraction of covered traffic.
 

docs/first_mirror.rst

@@ -5,8 +5,8 @@ The first_mirror subpackage enables simulations and analytical calculations of t
 
 This capability is also valuable for early detection of elephant flows. By classifying flows based on their first packets, the need for mid-flow rerouting is eliminated. Furthermore, it ensures that for the majority of a flow's lifespan, it will be subject to traffic engineering mechanisms specifically designed for elephant flows, such as individual routing paths. Additionally, the controller can continuously learn and refine its detection models based on the stream of first packets from flows.
 
-Tools
-=====
+Programs
+========
 
 .. toctree::
    :glob:

docs/index.rst

@@ -68,9 +68,9 @@ Contents
 
    workflow
    tools
-   tutorial
    first_mirror
    elephants
+   tutorial
    elephants_tutorial
 
 Reference

docs/tutorial.rst

@@ -1,5 +1,5 @@
-Tutorial: Distribution fitting
-******************************
+Tutorial: Model creation and distribution fitting
+*************************************************
 
 .. toctree::
    :maxdepth: 2