You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The DeepCoder specializes in functional programs that manipulate lists.
4
+
Each problem is written as a set of input-output examples.
5
+
6
+
The DeepCoder benchmark is derived from Balog et al. (2016) using the setup from Neo (Feng et al., 2018), as the evaluation benchmarks are not publicly available.
7
+
Neo thus generated 100 benchmarks following this workflow:
8
+
9
+
> We enumerate DSL programs with
10
+
> at least 5 components and randomly generate inputs and the
11
+
> corresponding output. This procedure is repeated for a fixed
12
+
> number of times until we either obtain 5 valid input-output
13
+
> examples or no examples have been found within the iter-
14
+
> ation limit. In the latter case, we restart this process and
15
+
> randomly search for a different program.
16
+
17
+
See
18
+
> Balog, M., Gaunt, A. L., Brockschmidt, M., Nowozin, S., & Tarlow, D. (2016). Deepcoder: Learning to write programs. arXiv preprint arXiv:1611.01989.
19
+
and
20
+
> Feng, Y., Martins, R., Bastani, O., & Dillig, I. (2018). Program synthesis using conflict-driven learning. ACM SIGPLAN Notices, 53(4), 420-435.
0 commit comments