--train_tasks 1.# always generates few stories from task 1.3

[NicolasAG]

Hi,

I noticed that when running python main.py with --train_tasks 1.2,1.4,1.6 it also generated a few stories from task 1.3.

This is the detailed command I run to reproduce the bug: python main.py --train_tasks 1.2,1.4,1.6 --test_tasks 1.2,1.3,1.4,1.5,1.6,1.7,1.8,1.9,1.10 --train_rows 100000 --test_rows 10000 --equal --template_split which generated the following training lines:

• 96,134 lines for relations of length 2
• 96,326 lines for relations of length 4
• 94,374 lines for relations of length 6
• 7,231 lines for relations of length 3 These numbers can be found with the following command: cat data/.../1.2,1.4,1.6_train.csv | grep task_1.3 | wc -l

A similar behavior is observed with --train_tasks 1.2,1.4,1.8.

I didn't try with other tasks than 1.#

Thanks :)

[koustuvsinha]

Should be fixed with GLC integration #9. @NicolasAG, can you retry this on the develop branch? Closing this for now, feel free to re-open if issue persists.