# We take length itself half of the time, and uniform between 1
# and length otherwise. The actual length can be slightly greater
- length = min(length, 1+torch.randint(length*2, (1,)).item())
+ length = min(length, 1 + torch.randint(length * 2, (1,)).item())
while len(s) < length:
v = random_var(nb_variables=nb_variables)
s += v + "=" + random_expr(variables, budget=20) + ";"
def generate_sequences(nb, nb_variables=5, length=20):
assert nb_variables <= 26
sequences = []
- result_max=99
+ result_max = 99
for n in range(nb):
result = None
while result == None or max(result.values()) > result_max: