Jupyter notebook of experiments

[dkamm]

• 2d embedding visualization
• neural network prediction of example program
• training summary (not provided in paper)
• search speedups (dfs + sort-and-add):
  • train on T = 3, test on T = 3, P = 500
  • train on T = 4, test on T = 5, P = 100
• generalization: T_train = [1...4], T_test = [1,...,5]

Test programs are semantically disjoint from train programs where T_train <= T_test.

Each training set is composed of programs of length T (not <= T).