Partition training data

[ksuderman]

Set up a system/program to partition the training data (see #1 and #10) into k partitions so we can perform k-fold validation runs.

Inputs

1. k - the number of partitions to generate
2. n - the total size of the generated evaluation set
3. a directory of files containing positive examples
4. a directory of files containing negative examples
5. the output directory to write results.

Outputs

1. k files in the output directory, each with n/k file IDs

Notes If n < |training set| then print a warning and generate the k partitions using the entire training set. else select n documents at random from the entire training set.

[ksuderman]

Part of #5

[nancyide]

No need to do this if using BERT. Should get rid of this task.