This is a fairly extensive set of changes - I've removed some of the functionality that was duplicated in depth_analysis.py and the preprocessing.py.
The results are now shown in a format much more similar to the tables in the paper, which makes comparing the original and reproduced results easier. The content of Tables 3, 4, and 5 are consistent pre- and post-changes apart from the dev row of Table 3 - this is due to the "best trial ID" being hardcoded in the original version. This ID varies between tests on different machines, so isn't always 60 as originally specified. Further suggestions are welcome!
This is a fairly extensive set of changes - I've removed some of the functionality that was duplicated in depth_analysis.py and the preprocessing.py.
The results are now shown in a format much more similar to the tables in the paper, which makes comparing the original and reproduced results easier. The content of Tables 3, 4, and 5 are consistent pre- and post-changes apart from the dev row of Table 3 - this is due to the "best trial ID" being hardcoded in the original version. This ID varies between tests on different machines, so isn't always 60 as originally specified. Further suggestions are welcome!