The alkanes data set. 

Alessio Micheli.
E-mail: micheli@di.unipi.it
http://pages.di.unipi.it/micheli/dataset/index.html 


- Input domain: collection of 150 alkanes CnH2n+2
with up to 10 carbon atoms (1331 vertexes)
The alkanes are trees (satured acyclic hydrocarbons molecules).

- Task: Prediction of the boiling point temperature in Celsius degree scale
(regression).

- Target values: Celsius degree / 100


There are 150 trees.
The training and test data were obtained by  10-fold-CV
(e.g the first fold has 15 compounds: number 1,11,21,31, ... in the test
set,  the second fold: 2,12,22,32,...).

The representation format of these trees is based on strings.
The format is then  transformed by a parser into an input graph.


'max arity' is the maximum fan-out of tree vertexes.
The 'symbol table' shows for each item:
the symbol, the fan-out, [the vectorial label]
(always the constant 1 for alkanes 
with a hydrogens suppressed representations).

The data set has a raw for each structure:
The tree. The target value. (The ID of the compound).

The tree is in  the parentheses (symbolic) representation.
Given a tree T(r) rooted in the node r, its  parentheses
representation is:

- rep[empty] = null string
- rep[T(r)] =  label(r) (rep[T(ch_1(r)]},rep[T(ch_2(r)]},...,rep[T(ch_k(r)]})

where ch_j(r) is the  j-th child of r.



For details (data, task, machine learning model, results) 
and references:


- [2009]: A. Micheli.
Neural Network for Graphs: A Contextual Constructive Approach.
IEEE Transactions on Neural Networks.
Vol. 20, n. 3, Pages 498-511, March 2009. IEEE Inc. ISSN 1045-9227.


Origins:


- A.M. Bianucci, A. Micheli, A. Sperduti, A. Starita.
Application of Cascade Correlation Networks for Structures to Chemistry, 
Applied Intelligence Journal (Kluwer Academic Publishers), 
Special Issue on "Neural Networks and Structured Knowledge" 
Vol. 12 (1/2): 117-146, 2000

- A. Micheli. 
Recursive Processing of Structured Domains in Machine Learning.
PhD Thesis, Department of Computer Science, University of Pisa, TD-13/03,
December 2003. Servizio Editoriale Universitario di Pisa.

or the other updated publications for improved results.


See also:


- D. Cherqaoui and D. Villemin. 
Use of neural network to determine the boiling point of alkanes. 
J. Chem. Soc. Faraday Trans., 1994. 90(1):97-102

- A. Micheli, F. Portera, A. Sperduti.
A preliminary empirical comparison of recursive neural networks and tree
kernel methods on regression tasks for tree structured domains.
Neurocomputing. Volume 64, March 2005, Pages 73-92.  (c) 2005 Elsevier B.V


- A. Micheli, D. Sona, A. Sperduti,
Contextual Processing of Structured Data by Recursive Cascade Correlation.
IEEE Transactions on Neural Networks. Vol. 15, n. 6, Pages 1396- 1410,
November 2004.




For questions regarding the database content, the origins of the
learning dataset, the updating on the results for this data and
collaborations, 
please contact Alessio Micheli (E-mail: micheli@di.unipi.it).