ClaraBing commited on
Commit
c0d4681
·
1 Parent(s): 6a23009

add option 'random_length'

Browse files
Files changed (1) hide show
  1. automata.py +21 -7
automata.py CHANGED
@@ -117,7 +117,12 @@ class AutomatonSampler:
117
  data_config['length'] = 20
118
  self.T = self.data_config['length']
119
 
120
- self.__info__ = " - T (int): sequence length"
 
 
 
 
 
121
 
122
  def f(self, x):
123
  """
@@ -128,6 +133,11 @@ class AutomatonSampler:
128
  def sample(self):
129
  raise NotImplementedError()
130
 
 
 
 
 
 
131
  def help(self):
132
  print(self.__info__)
133
 
@@ -146,7 +156,8 @@ class BinaryInputSampler(AutomatonSampler):
146
  raise NotImplementedError()
147
 
148
  def sample(self):
149
- x = self.np_rng.binomial(1, self.prob1, size=self.T)
 
150
  return x, self.f(x)
151
 
152
  class ParitySampler(BinaryInputSampler):
@@ -258,8 +269,9 @@ class ABABSampler(BinaryInputSampler):
258
  def sample(self):
259
  pos_sample = np.random.random() < self.prob_abab_pos_sample
260
  if pos_sample:
261
- x = [0,1,0,1] * (self.T//4)
262
- x += [0,1,0,1][:(self.T%4)]
 
263
  x = np.array(x)
264
  return x, self.f(x)
265
  else:
@@ -295,9 +307,10 @@ class FlipFlopSampler(AutomatonSampler):
295
  return np.array(states)
296
 
297
  def sample(self):
298
- rand = np.random.uniform(size=self.T)
 
299
  nonzero_pos = (rand < 0.5).astype(np.int64)
300
- writes = np.random.choice(range(1, self.n_states+1), size=self.T)
301
  x = writes * nonzero_pos
302
  return x, self.f(x)
303
 
@@ -389,7 +402,8 @@ class SymmetricSampler(AutomatonSampler):
389
  return np.array(labels)
390
 
391
  def sample(self):
392
- x = np.random.choice(range(self.n_actions), replace=True, size=self.T)
 
393
 
394
  return x, self.f(x)
395
 
 
117
  data_config['length'] = 20
118
  self.T = self.data_config['length']
119
 
120
+ if 'random_length' not in data_config:
121
+ data_config['random_length'] = 0
122
+ self.random_length = data_config['random_length']
123
+
124
+ self.__info__ = " - T (int): sequence length.\n" \
125
+ + " - random_length (int in {0, 1}): whether to randomly sample a length per sample.\n"
126
 
127
  def f(self, x):
128
  """
 
133
  def sample(self):
134
  raise NotImplementedError()
135
 
136
+ def sample_length(self):
137
+ if self.random_length:
138
+ return self.np_rng.choice(range(1, self.T+1))
139
+ return self.T
140
+
141
  def help(self):
142
  print(self.__info__)
143
 
 
156
  raise NotImplementedError()
157
 
158
  def sample(self):
159
+ T = self.sample_length()
160
+ x = self.np_rng.binomial(1, self.prob1, size=T)
161
  return x, self.f(x)
162
 
163
  class ParitySampler(BinaryInputSampler):
 
269
  def sample(self):
270
  pos_sample = np.random.random() < self.prob_abab_pos_sample
271
  if pos_sample:
272
+ T = self.sample_length()
273
+ x = [0,1,0,1] * (T//4)
274
+ x += [0,1,0,1][:(T%4)]
275
  x = np.array(x)
276
  return x, self.f(x)
277
  else:
 
307
  return np.array(states)
308
 
309
  def sample(self):
310
+ T = self.sample_length()
311
+ rand = np.random.uniform(size=T)
312
  nonzero_pos = (rand < 0.5).astype(np.int64)
313
+ writes = np.random.choice(range(1, self.n_states+1), size=T)
314
  x = writes * nonzero_pos
315
  return x, self.f(x)
316
 
 
402
  return np.array(labels)
403
 
404
  def sample(self):
405
+ T = self.sample_length()
406
+ x = np.random.choice(range(self.n_actions), replace=True, size=T)
407
 
408
  return x, self.f(x)
409