0
$\begingroup$

This website contains numerous datasets that can be used for machine learning projects. I wanted to experiment with one of the datasets, so I took a look at the "Poker Hand Data Set".

However, something seems odd to me when looking at the description:

Training set - Total of 25010 instances in a domain of 311,875,200.

Testing set - Total of one million instances in a domain of 311,875,200.

Why is the training set so small? And why is the testing set so big? I thought that normally these sets should be distributed 80%/20%, but this is 2.4%/97.6%.

$\endgroup$
0
$\begingroup$

There could be alot of things atrributed to this.

Firstly, the data set is quite old (2007). Since then advances in the management and ability to parallelise training of such larger sets has been huge. Batch learning, GPU optimisation etc. A problem you dont tend to have when testing - no such iterative process.

It of course shouldnt make a difference if you swap the data sets. Assuming the distribution of each "hand" is the same in each (it seems to be approximately) you will be able to train/ pool all data and divide how you wish.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.