Reproducing results of ICSE 16 paper

[kanghj]

Hi, I am trying to reproduce the results of https://arxiv.org/pdf/1512.06448.pdf.

In my setup, I am using the file-level tokenizer, I've changed sourcerer-cc.properties MIN_TOKENS to 1,

# Ignore all files outside these bounds
MIN_TOKENS=1
MAX_TOKENS=500000

as well as changed runnodes.sh's threshold threshold="${3:-7}".

Using BigCloneEval, I'm using these flags "-st both -mit 50 -mil 6". The default clone matcher is used. I'm getting the following results for type-1 and type-2 clones:

Type-1: 34301 / 35787 = 0.9584765417609746
Type-2: 3334 / 4573 = 0.7290618849770392

According to the ICSE paper, SourcererCC is able to get 1.0 on Type-1, and 0.98 on Type-2.

Is there any step in particular that I missed, or is there another configuration to change, in order to reproduce the ICSE paper's results?

[kanghj]

Ah, perhaps I should be using the block-level tokenizer?

Edit: using the block-level tokenizer seems to result in worse performance

[qw3ry]

If I understand the paper correctly (Table 1), they used a method level tokenizer. Sadly, it doesnt seem to be available :(