Fequently asked questions
Q: Why should I use ELF?
A: ELF is a high-performance framework mixed with state-of-the art algorithms. It is suited for a quick start on a supervised learning problem with non-sparse features.
Q: Can I apply the ELF to large scale data?
A: Yes, it is designed to handle large data sets. For example one million samples and a few hundred features should work for a few build-in algorithms.
Q: How fast is ELF?
A: We have training and prediction time measures for several datasets, see here, here and here
Q: How do I install ELF?
A: See here
Q: Which OS is supported?
A: Tested on 64bit Linux (Ubuntu), should run on other platforms as well.
Q: Can I run ELF on 32bit machines?
A: Yes, we have tested it on an Intel Atom N270 platform.
Q: Why is the programming language C++?
A: Offers structural advantages to C, easy integration of Intel's performance and math libraries MKL and IPP.
Q: What are the hardware requirements of ELF?
A: It depends on the size of the dataset, more is better.
Q: Can I use multi-core machines to speedup the training?
A: Yes, ELF supports multicore machines (via OpenMP) in the cross-validation or bagging process for parameter selection.
Q: Can I use a cluster of PCs to speedup the training?
A: No support for cluster architecture now.
Q: I will read my own dataset, what are the required steps?
A: See here for the structure of the internal data representation.
Q: When I execute ELF the console output is: ELF: DatasetReader.cpp:2239: ..[skipped].. Assertion `features.size() == targets.size()' failed. ?
A: Here, an internal error occured, maybe the datafile has the wrong format.
Q: When I execute ELF the console output is finalized with "Aborted." ?
A: Maybe a hardware problem, or an unknown software bug.