5 import torch, torchvision
8 from torch.nn import functional as F
10 ######################################################################
14 def generate_sequences(self, nb):
17 def seq2str(self, seq):
18 return "[NOT IMPLEMENTED]"
24 class ProblemTwoTargets(Problem):
25 def __init__(self, len_total=10, len_targets=3):
26 assert len_targets >= 3
27 assert len_total >= 3 * len_targets - 1
28 self.len_total = len_total
29 self.len_targets = len_targets
31 def generate_sequences(self, nb):
32 k = torch.arange(self.len_total)[None, :]
33 s = torch.randint(10, (nb, self.len_total))
34 l = torch.rand(nb, self.len_total)
35 l = l * (k <= self.len_total - self.len_targets).long()
36 k1 = l.argmax(dim=1, keepdim=True)
37 m = (k != k1).long() * (k != k1 + self.len_targets - 1).long()
38 s = s * m + 10 * (1 - m)
41 - (k + self.len_targets - 1 >= k1).long()
42 * (k < k1 + self.len_targets).long()
44 k2 = l.argmax(dim=1, keepdim=True)
45 m = (k != k2).long() * (k != k2 + self.len_targets - 1).long()
46 s = s * m + 11 * (1 - m)
47 a1 = s.gather(dim=1, index=k1 + 1 + torch.arange(self.len_targets - 2)[None, :])
48 a2 = s.gather(dim=1, index=k2 + 1 + torch.arange(self.len_targets - 2)[None, :])
49 sequences = torch.cat(
52 torch.full((nb, 1), 12),
54 torch.full((nb, 1), 12),
56 torch.full((nb, 1), 12),
60 ar_mask = (sequences == 12).long()
61 ar_mask = (ar_mask.cumsum(1) - ar_mask).clamp(max=1)
62 return sequences, ar_mask
64 def seq2str(self, seq):
65 return "".join("0123456789-+|"[x.item()] for x in seq)
71 class ProblemByHeart(Problem):
72 def __init__(self, nb_sentences=100, len_prompt=8, len_result=8):
73 self.seq = torch.randint(10, (nb_sentences, len_prompt + 1 + len_result))
74 self.seq[:, len_prompt] = 10
76 def generate_sequences(self, nb):
77 sequences = self.seq[torch.randint(self.seq.size(0), (nb,))]
78 ar_mask = (sequences == 10).long()
79 ar_mask = (ar_mask.cumsum(1) - ar_mask).clamp(max=1)
80 return sequences, ar_mask
82 def seq2str(self, seq):
83 return "".join("0123456789|"[x.item()] for x in seq)
89 class ProblemLearnOperator(Problem):
90 def __init__(self, nb_operators=100, len_source=5, len_result=8):
91 self.len_source = len_source
92 self.len_result = len_result
93 self.len_nb_operator = int(math.log(nb_operators) / math.log(10)) + 1
94 self.operators = F.one_hot(
95 torch.rand(nb_operators, len_result, len_source).argmax(-1),
96 num_classes=len_source,
99 def generate_sequences(self, nb):
100 nb_operators = torch.randint(self.operators.size(0), (nb,))
101 operators = self.operators[nb_operators]
103 nb_operators[:, None]
104 // 10 ** torch.arange(self.len_nb_operator - 1, -1, -1)
106 marker1 = torch.full((nb, 1), 10)
107 source = torch.rand(nb, 10).sort(dim=1).indices[:, : self.len_source]
108 marker2 = torch.full((nb, 1), 11)
109 result = operators.bmm(source[:, :, None]).squeeze(-1)
110 sequences = torch.cat((nb_operators, marker1, source, marker2, result), 1)
111 ar_mask = (sequences == 11).long()
112 ar_mask = (ar_mask.cumsum(1) - ar_mask).clamp(max=1)
113 return sequences, ar_mask
115 def seq2str(self, seq):
116 return "".join("0123456789|>"[x.item()] for x in seq)
122 class ProblemGuessOperator(Problem):
123 def __init__(self, len_source=5, len_result=8):
124 self.len_source = len_source
125 self.len_result = len_result
127 def generate_sequences(self, nb):
128 operators = F.one_hot(
129 torch.rand(nb, self.len_result, self.len_source).argmax(-1),
130 num_classes=self.len_source,
132 source1 = torch.rand(nb, 10).sort(dim=1).indices[:, : self.len_source]
133 marker1 = torch.full((nb, 1), 10)
134 result1 = operators.bmm(source1[:, :, None]).squeeze(-1)
135 marker2 = torch.full((nb, 1), 11)
136 source2 = torch.randint(10, (nb, self.len_source))
137 marker3 = torch.full((nb, 1), 12)
138 result2 = operators.bmm(source2[:, :, None]).squeeze(-1)
140 sequences = torch.cat(
141 (source1, marker1, result1, marker2, source2, marker3, result2), 1
143 ar_mask = (sequences == 12).long()
144 ar_mask = (ar_mask.cumsum(1) - ar_mask).clamp(max=1)
145 return sequences, ar_mask
147 def seq2str(self, seq):
148 return "".join("0123456789>|~"[x.item()] for x in seq)
154 class ProblemAddition(Problem):
155 def __init__(self, nb_digits=10, zero_padded=False, inverted_result=False):
156 self.nb_digits = nb_digits
157 self.zero_padded = zero_padded
158 self.inverted_result = inverted_result
159 self.char2id = dict([(c, n) for n, c in enumerate("0123456789+=$")])
160 self.id2char = dict([(n, c) for c, n in self.char2id.items()])
162 def tensorize(self, strings):
163 len_max = max([len(x) for x in strings])
168 [self.char2id[c] for c in s + "$" * (len_max - len(s))]
176 def generate_sequences(self, nb):
179 a, b = torch.randint(10**self.nb_digits, (2,))
181 a, b, c = str(a.item()), str(b.item()), str(c.item())
183 a = "0" * (self.nb_digits - len(a)) + a
184 b = "0" * (self.nb_digits - len(b)) + b
185 c = "0" * (self.nb_digits + 1 - len(c)) + c
186 if self.inverted_result:
188 sequences.append(f"{a}+{b}={c}$")
190 sequences = self.tensorize(sequences)
191 ar_mask = (sequences == self.char2id["="]).long()
192 ar_mask = (ar_mask.cumsum(1) - ar_mask).clamp(max=1)
193 return sequences, ar_mask
195 def seq2str(self, seq):
196 return "".join(self.id2char[x.item()] for x in seq)
199 if __name__ == "__main__":
200 p = ProblemTwoTargets(12, 4)
201 s, m = p.generate_sequences(10)