---
license: mit
---

This dataset is created by [DeepSim: deep learning code functional similarity](https://dl.acm.org/doi/10.1145/3236024.3236068).

I downloaded `googlejam4.tar.gz` from [parasol-aser/deepsim](https://github.com/parasol-aser/deepsim/), fixed encoding of `6/googlejam6.p261.Round1B.java` and `1/googlejam1.p815.MushroomMonster.java`, and re-compressed in `zip`.

The `all` split (all 12 problems) is consistent with their paper (I guess...). 

The `test` split (problem 5, 6, 7, 8, 12) is used in experiments of [Language Models are Universal Embedders](https://arxiv.org/pdf/2310.08232.pdf).

| problem | code num |
|---------|----------|
| 1       | 478      |
| 2       | 88       |
| 3       | 242      |
| 4       | 38       |
| 5       | 2        |
| 6       | 435      |
| 7       | 27       |
| 8       | 245      |
| 9       | 68       |
| 10      | 18       |
| 11      | 20       |
| 12      | 4        |