def seq2str(self, seq):
return "[NOT IMPLEMENTED]"
+ def compute_nb_correct(self, input, ar_mask, result):
+ nb_total = ar_mask.sum().item()
+ nb_correct = ((result == input).long() * ar_mask).sum().item()
+ return nb_total, nb_correct
+
+####################
+
+
+class ProblemDegradation(Problem):
+ def __init__(self, nb_state_tokens=5, nb_time_steps=5, value_max=25, hard=False):
+ self.nb_state_tokens = nb_state_tokens
+ self.nb_time_steps = nb_time_steps
+ self.value_max = value_max
+ self.hard = hard
+
+ def generate_sequences(self,nb):
+
+ x = (torch.rand(nb,self.nb_state_tokens).sort(dim=-1).indices == 0).long() * self.value_max
+ seq = [x]
+
+ for t in range(self.nb_time_steps-1):
+ v = torch.rand(x.size()) * (x > 0).float()
+ u = (v.max(dim=-1,keepdim=True).values == v).long()
+ n = (u*x*torch.rand(x.size())).long().sum(dim=-1,keepdim=True) // 2
+ x = x + n * (u.roll(shifts=-1,dims=-1) - 2 * u + u.roll(shifts=1,dims=-1))
+ seq.append(x)
+
+ if self.hard: seq.reverse()
+
+ seq = torch.cat(seq,dim=1)
+ return seq,seq.new_full(seq.size(), 1, dtype=torch.int64)
+
+ def compute_nb_correct(self, input, ar_mask, result):
+ nb_total = result.size(0)
+ nb_correct = 0
+
+ for seq in result:
+ states = list(seq.split(self.nb_state_tokens))
+ if self.hard:
+ states.reverse()
+
+ d = states[0]
+ j=d.sort(descending=True).indices[0]
+ e=d.new_zeros(d.size())
+ e[j]=self.value_max
+ if (d-e).abs().sum() == 0:
+ nb_errors = 0
+ for k in range(len(states)-1):
+ d=states[k]-states[k+1]
+ j=d.sort(descending=True).indices[0]
+ e=d.new_zeros(d.size())
+ e[j]=d[j]
+ e[(j+1)%e.size(0)]=-d[j]//2
+ e[(j-1)%e.size(0)]=-d[j]//2
+ if (d-e).abs().sum() > 0:
+ nb_errors += 1
+ if nb_errors == 0:
+ nb_correct += 1
+
+ return nb_total, nb_correct
+
+ def seq2str(self, seq):
+ return " | ".join( [ " ".join([f"{x:02d}" for x in s ]) for s in seq.split(self.nb_state_tokens) ] )
####################
class ProblemLearnOperator(Problem):
- def __init__(self, nb_operators=100, len_source=5, len_result=8):
+ def __init__(self, nb_operators=100, len_source=6, len_result=9):
self.len_source = len_source
self.len_result = len_result
self.len_nb_operator = int(math.log(nb_operators) / math.log(10)) + 1
if __name__ == "__main__":
- p = ProblemTwoTargets(12, 4)
- s, m = p.generate_sequences(10)
- for x in s:
- print(p.seq2str(x))
+ p = ProblemDegradation(hard=False)
+ s, m = p.generate_sequences(10000)
+ print(p.seq2str(s[0]))
+ print(p.compute_nb_correct(None, None, s))