add AdderSampler
Browse files- automata.py +68 -2
automata.py
CHANGED
@@ -150,6 +150,8 @@ class BinaryInputSampler(AutomatonSampler):
|
|
150 |
"""
|
151 |
This is a parent class that must be inherited.
|
152 |
Subclasses: ParitySampler, GridworldSampler, ABABSampler
|
|
|
|
|
153 |
"""
|
154 |
def __init__(self, data_config):
|
155 |
super().__init__(data_config)
|
@@ -287,6 +289,65 @@ class ABABSampler(BinaryInputSampler):
|
|
287 |
return super().sample()
|
288 |
|
289 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
290 |
|
291 |
|
292 |
class FlipFlopSampler(AutomatonSampler):
|
@@ -303,7 +364,7 @@ class FlipFlopSampler(AutomatonSampler):
|
|
303 |
|
304 |
self.__info__ = f"Flipflop with n={self.n_states} states:\n" \
|
305 |
+f"- Inputs: tokens are either 0 (read) or 1:{self.n} (write).\n" \
|
306 |
-
+ "- Labels:
|
307 |
+ "- Config:\n" \
|
308 |
+ " - n (int): number of write states; i.e. the states are 1,2,...,n, plus a default start state 0.\n" \
|
309 |
+ self.__info__
|
@@ -638,8 +699,13 @@ class QuaternionSampler(AutomatonSampler):
|
|
638 |
return x, self.f(x)
|
639 |
|
640 |
|
|
|
|
|
|
|
|
|
641 |
dataset_map = {
|
642 |
'abab': ABABSampler,
|
|
|
643 |
'alternating': AlternatingSampler,
|
644 |
'cyclic': CyclicSampler,
|
645 |
'dihedral': DihedralSampler,
|
@@ -648,6 +714,6 @@ dataset_map = {
|
|
648 |
'parity': ParitySampler,
|
649 |
'quaternion': QuaternionSampler,
|
650 |
'symmetric': SymmetricSampler,
|
651 |
-
# TODO:
|
652 |
}
|
653 |
|
|
|
150 |
"""
|
151 |
This is a parent class that must be inherited.
|
152 |
Subclasses: ParitySampler, GridworldSampler, ABABSampler
|
153 |
+
|
154 |
+
TODO: sample sequences with a given number of 1s
|
155 |
"""
|
156 |
def __init__(self, data_config):
|
157 |
super().__init__(data_config)
|
|
|
289 |
return super().sample()
|
290 |
|
291 |
|
292 |
+
class AdderSampler(BinaryInputSampler):
|
293 |
+
def __init__(self, data_config):
|
294 |
+
super().__init__(data_config)
|
295 |
+
self.name = 'addition'
|
296 |
+
|
297 |
+
if 'n_addends' not in data_config:
|
298 |
+
data_config['n_addends'] = 2
|
299 |
+
self.n_addends = data_config['n_addends']
|
300 |
+
self.addend_scales = np.array([2**i for i in range(self.n_addends)]).reshape(-1, 1)
|
301 |
+
|
302 |
+
if 'label_type' not in data_config:
|
303 |
+
data_config['label_type'] = 'state'
|
304 |
+
self.label_type = data_config['label_type']
|
305 |
+
|
306 |
+
self.__info__ = f'Adder of n={self.n_addends} binary numbers:\n' \
|
307 |
+
+f"- Inputs: {self.n_addends} binary numbers, encoded as the int for the {self.n_addends}-bit binary number.\n" \
|
308 |
+
+ "- Labels: depending on the label_type.\n" \
|
309 |
+
+ "- Config:\n" \
|
310 |
+
+ " - n_addends (int): number of binary numbers to be added; default as 2.\n" \
|
311 |
+
+ " - label_type (str): choosing from the following options: \n" \
|
312 |
+
+f" - 'state': the state id, i.e. the int for the base-{self.n_addends} int corresponding to the number (carry, digit). \n" \
|
313 |
+
+f" - 'digit': the current output base-{self.n_addends} digit, without the carry. \n" \
|
314 |
+
+ " - 'position': the current carry bit.\n" \
|
315 |
+
+ self.__info__
|
316 |
+
|
317 |
+
|
318 |
+
def f(self, x):
|
319 |
+
outputs, carries = [], []
|
320 |
+
carry = 0
|
321 |
+
T = x.shape[-1]
|
322 |
+
for i in range(T):
|
323 |
+
curr_sum = x[:, i].sum() + carry
|
324 |
+
# NOTE: 'mod n_addends' makes sure the carry is binary
|
325 |
+
output, carry = curr_sum % self.n_addends, curr_sum // self.n_addends
|
326 |
+
outputs += output,
|
327 |
+
carries += carry,
|
328 |
+
outputs = np.array(outputs).astype(np.int64)
|
329 |
+
carries = np.array(carries).astype(np.int64)
|
330 |
+
|
331 |
+
if self.label_type == 'state':
|
332 |
+
return outputs + self.n_addends*carries
|
333 |
+
elif self.label_type == 'digit':
|
334 |
+
return outputs
|
335 |
+
elif self.label_type == 'carry':
|
336 |
+
return carries
|
337 |
+
|
338 |
+
def sample_addend(self, T):
|
339 |
+
a = self.np_rng.binomial(1, self.prob1, size=T)
|
340 |
+
return a
|
341 |
+
|
342 |
+
def sample(self):
|
343 |
+
T = self.sample_length()
|
344 |
+
x = np.stack([self.sample_addend(T) for _ in range(self.n_addends)])
|
345 |
+
# Pad the most significant bit (rightmost position, i.e. we're reversing the number) with 0 to handle the potential carry
|
346 |
+
pad = np.zeros((self.n_addends, 1))
|
347 |
+
x = np.concatenate([x, pad], 1)
|
348 |
+
|
349 |
+
x_encode = (self.addend_scales * x).sum(0)
|
350 |
+
return x_encode, self.f(x)
|
351 |
|
352 |
|
353 |
class FlipFlopSampler(AutomatonSampler):
|
|
|
364 |
|
365 |
self.__info__ = f"Flipflop with n={self.n_states} states:\n" \
|
366 |
+f"- Inputs: tokens are either 0 (read) or 1:{self.n} (write).\n" \
|
367 |
+
+ "- Labels: the state id.\n" \
|
368 |
+ "- Config:\n" \
|
369 |
+ " - n (int): number of write states; i.e. the states are 1,2,...,n, plus a default start state 0.\n" \
|
370 |
+ self.__info__
|
|
|
699 |
return x, self.f(x)
|
700 |
|
701 |
|
702 |
+
|
703 |
+
|
704 |
+
|
705 |
+
|
706 |
dataset_map = {
|
707 |
'abab': ABABSampler,
|
708 |
+
'add': AdderSampler,
|
709 |
'alternating': AlternatingSampler,
|
710 |
'cyclic': CyclicSampler,
|
711 |
'dihedral': DihedralSampler,
|
|
|
714 |
'parity': ParitySampler,
|
715 |
'quaternion': QuaternionSampler,
|
716 |
'symmetric': SymmetricSampler,
|
717 |
+
# TODO: add Dyck
|
718 |
}
|
719 |
|