this is for holding javascript data
Denes Csala edited Personal reflections.md
over 8 years ago
Commit id: 9e5e8ed5faec95db38b677990a11bfe7592029e3
deletions | additions
diff --git a/Personal reflections.md b/Personal reflections.md
index 96fbc49..a257dc8 100644
--- a/Personal reflections.md
+++ b/Personal reflections.md
...
#Personal reflections
Stochastic The stochastic nature of human behavior makes it difficult to operate with global
patterns
This research patterns, such as the occupancy schedule standards. Reality is much more complex than that and with the rise of cheap electronics integrated into more and more smart buildings, occupancy monitoring and subsequent prediction for energy consumption optimization is a vital component of building energy portfolio analysis. OSL makes it possible to predict occupancy patterns of buildings of similar scope but unfitted with
sensor sensors (oftentimes a certain array of sensors is deployed in a location, then removed - such as the ones at Masdar Institute) or having fewer or lower quality or reliability
sensors
The above step is crucial for the implementation of sensors. I think the
smart grid and renewable energy systems requiring larger energy conservation for expensive storage
Not OSL is in no way meant to replace but to complement agent-based models – opening the way to incorporate feedback processes in the
modeling
C4.5 is modeling. C4.5, the algorithm OSL uses for creating a
decision tree model is suitable
algorithm for learning the occupancy presence in
offices: offices and it leads to a claimed 90.3 % accuracy.
4 occupancy profiles identified using the k-means algorithm with Davies-Bouldin optimization
Day-to-day variation in profiles observed for the same offices
Large variation but recurring patterns activity times lead However this leads to
developing working user profiles some of potential problems of the paper.
I think that MTD OSL is
a powerful concept with a very elegant algorithm. The paper is well-written arguably split into two somewhat disconnected parts: the decision tree model and
it the occupancy profile creation through clustering. It is
easy to follow for anyone with a background not clear in
data mining - for me it represented a bit of a difficulty what ways are the occupancy profiles (the mined or the known ones?) used as
I have a base in statistics and network science but not this kind inputs for the clustering algorithm. Moreover, OSL also provides very little information about the methodology of
technical data mining that the
paper preconditions. It is important to put this paper in clustering - despite the fact the
right context: it time series clustering is
a theory paper, large and potentially very complex area \cite{Denton_2005}. There is no accuracy measure or reproducible, universal metric provided for the
latest in a long succession assessment of
topic modeling papers published by the author group over the past 4-5 years, which were sequentially built upon each other. First, they extensively studied the LDA algorithm, then they extended it with must-link mining the results and
eventually with cannot-link mining. Unless one goes through the
previous bibliography, then mathematical theory could be too fast-paced to follow. The description is quite clear until probable cross-validation process of the
cannot-links. I think decision tree model is not described either.
These flaws, together with a large and watery introduction decrease the
strongest weakness credibility of the
paper comes at this point, while they provide paper, despite the
mathematical model to map fact that the
cannot-links, they only explain authors of OSL have multiple well-received publications in the
method and its plausibility vaguely. building modeling field.