GitLab

Not using it now but quite sure it will be useful eventually.

Reads the bytes of the file and computes the checksum.

Yields the same results as the sha1sum command.

When we train the classifiers, we also save some metadata, to be able
to retrieve what files were used for the training.

I'm suspecting 10 is too little to be able to get accurate measurements.

The function that computed the weight is different in the JBrainy
paper: The samples that yeild less than 5% improvement are just dropped.

Two benchmarks with different method selection strategies are different.

The test uses an old API

We try the same test with many seeds instead of just one.

We added assertions to check that the state is consistent.

Created a fill method inside the BCDataStructureHandler to fill
collections. Uses the methods defined in CollectionFiller (because you
might want to call the same methods "outside" of BCDataStructureHandler.

We would like to use both the ANN and Random Forest classifiers.

Check the benchmark state is null when we call prepareBenchmark.

Checks the data structure is the same as the simulation state.

We remove the fill methods sprinkled everywhere to use
CollectionFiller instead.

The size of benchmark and the size of collection were swapped.

This class fills collection with random data, to be used in benchmarks.

This function is not to be used. Use BCBenchmarkPackage instead.

The benchmark generation method that was used before is supposed to be
deprecated.

Added functions to build BCBenchmarkPackages with a seed assigned to
time(), for sets and maps.

You don't need to provide the seed, it's set to the unix timestamp.

I think I forgot this in an earlier commit...

We use long seeds instead of int seeds. Because if we want tot use
time() for the seed, then I think it's better to use Long values.

The way we generate benchmark for runs with JMH is not the same as the
way we use for gathering the features, so the counters make no sense!

We need to fix that, probably by retiring genBenchmark.

Removed unused imports and unused function

If the markov chain contains information unknown to the jbrainy
(addFirst in our example), we filter it out, so the markov chain
can't generate a benchmark that has this invalid method.

Allows for removing some states that are considered invalid.

When is a state is considered invalid, the probabilities of moving to it
become zero.

I want to do that because functions manipulating streams of doubles are
better for some reason.