| /external/tensorflow/tensorflow/contrib/opt/python/training/ |
| nadam_optimizer_test.py | 39 beta2=0.999,
41 alpha_t = alpha * np.sqrt(1 - beta2**t) / (1 - beta1**t)
44 v_t = beta2 * v + (1 - beta2) * g_t * g_t
|
| adam_gs_optimizer.py | 15 """Adam rewrite to use global step for computing beta1 & beta2 accumulation."""
44 beta2=0.999,
51 global step for computing beta1 and beta2 accumulators, instead of having
52 optimizer keep its own independent beta1 and beta2 accumulators as non-slot
92 beta2: A float value or a constant float tensor. The exponential decay
100 enabled, `learning_rate`, `beta1`, `beta2`, and `epsilon` can each be a
108 self._beta2 = beta2
132 beta2 = self._call_if_callable(self._beta2)
137 self._beta2_t = ops.convert_to_tensor(beta2, name="beta2")
[all...] |
| weight_decay_optimizers_test.py | 38 beta2=0.999, epsilon=1e-8):
39 lr_t = lr * np.sqrt(1 - beta2**t) / (1 - beta1**t)
42 v_t = beta2 * v + (1 - beta2) * g_t * g_t
|
| adam_gs_optimizer_test.py | 45 beta2=0.999,
47 alpha_t = alpha * np.sqrt(1 - beta2**t) / (1 - beta1**t)
50 v_t = beta2 * v + (1 - beta2) * g_t * g_t
193 beta2 = lambda: 0.999
198 beta2 = beta2()
|
| lazy_adam_gs_optimizer_test.py | 45 beta2=0.999,
47 alpha_t = alpha * np.sqrt(1 - beta2**t) / (1 - beta1**t)
50 v_t = beta2 * v + (1 - beta2) * g_t * g_t
212 beta2 = lambda: 0.999
217 beta2 = beta2()
|
| lazy_adam_optimizer_test.py | 45 beta2=0.999,
47 alpha_t = alpha * np.sqrt(1 - beta2**t) / (1 - beta1**t)
50 v_t = beta2 * v + (1 - beta2) * g_t * g_t
188 beta2 = lambda: 0.999
193 beta2 = beta2()
|
| weight_decay_optimizers.py | 359 def __init__(self, weight_decay, learning_rate=0.001, beta1=0.9, beta2=0.999,
370 beta2: A float value or a constant float tensor.
380 weight_decay, learning_rate=learning_rate, beta1=beta1, beta2=beta2,
|
| adamax.py | 40 def __init__(self, learning_rate=0.001, beta1=0.9, beta2=0.999, epsilon=1e-8,
59 v_t <- max(beta2 * v_{t-1}, abs(g))
79 beta2: A float value or a constant float tensor.
86 super(AdaMaxOptimizer, self).__init__(learning_rate, beta1, beta2,
151 # u_t = max(beta2 * u, abs(g_t))
|
| adamax_test.py | 44 beta2=0.999,
47 v_t = np.maximum(beta2 * v, np.abs(g_t))
60 beta2=0.999,
64 v_t_slice = np.maximum(beta2 * v[indices], np.abs(g_t))
|
| /external/tensorflow/tensorflow/python/keras/optimizer_v2/ |
| adam_test.py | 45 beta2=0.999,
47 lr_t = lr * np.sqrt(1 - beta2**(t + 1)) / (1 - beta1**(t + 1))
50 v_t = beta2 * v + (1 - beta2) * g_t * g_t
64 beta2=0.999,
66 lr_t = lr * np.sqrt(1 - beta2**(t + 1)) / (1 - beta1**(t + 1))
69 v_t = beta2 * v + (1 - beta2) * g_t * g_t
85 beta2=0.999,
89 lr_t = lr * np.sqrt(1 - beta2**(t + 1)) / (1 - beta1**(t + 1)
[all...] |
| nadam_test.py | 57 beta2=0.999,
65 v_t = beta2 * v + (1 - beta2) * g_t * g_t
68 v_prime_t = v_t / (1 - beta2**(t + 1))
|
| adamax_test.py | 43 beta2=0.999,
46 v_t = np.maximum(beta2 * v, np.abs(g_t))
59 beta2=0.999,
63 v_t_slice = np.maximum(beta2 * v[indices], np.abs(g_t))
|
| /external/tensorflow/tensorflow/python/training/ |
| training_ops_test.py | 277 beta2 = np.array(0.999, dtype=var.dtype)
279 beta2_power = beta2**t
283 beta2_t = constant_op.constant(beta2, self._toType(var.dtype), [])
292 beta2, epsilon)
300 def _adamUpdateNumpy(self, param, g_t, t, m, v, alpha, beta1, beta2, epsilon):
301 alpha_t = alpha * np.sqrt(1 - beta2**t) / (1 - beta1**t)
304 v_t = beta2 * v + (1 - beta2) * g_t * g_t
|
| adam.py | 42 beta2=0.999,
84 beta2: A float value or a constant float tensor. The exponential decay
92 enabled, `learning_rate`, `beta1`, `beta2`, and `epsilon` can each be a
100 self._beta2 = beta2
119 # Create the beta1 and beta2 accumulators on the same device as the first
137 beta2 = self._call_if_callable(self._beta2)
142 self._beta2_t = ops.convert_to_tensor(beta2, name="beta2")
194 # v_t = beta2 * v + (1 - beta2) * (g_t * g_t
[all...] |
| adam_test.py | 44 beta2=0.999,
46 alpha_t = alpha * np.sqrt(1 - beta2**t) / (1 - beta1**t)
49 v_t = beta2 * v + (1 - beta2) * g_t * g_t
184 beta2 = lambda: 0.999
189 beta2 = beta2()
|
| /external/speex/libspeexdsp/ |
| scal.c | 156 float beta, beta2; local
186 beta2 = beta;
205 if (max_alpha > .98/(1.+beta2))
206 max_alpha = .98/(1.+beta2);
|
| /external/tensorflow/tensorflow/compiler/tests/ |
| adam_test.py | 41 beta2=0.999,
43 alpha_t = alpha * np.sqrt(1 - beta2**t) / (1 - beta1**t)
46 v_t = beta2 * v + (1 - beta2) * g_t * g_t
|
| adamax_test.py | 40 beta2=0.999,
43 v_t = np.maximum(beta2 * v, np.abs(g_t))
|
| /external/tensorflow/tensorflow/contrib/optimizer_v2/ |
| adam.py | 37 def __init__(self, learning_rate=0.001, beta1=0.9, beta2=0.999, epsilon=1e-8,
81 beta2: A float hyperparameter. The exponential decay rate for the 2nd
94 self._set_hyper("beta2", beta2)
108 initial_value=lambda: state.get_hyper("beta2"), name="beta2_power")
127 state.get_hyper("beta2", var.dtype.base_dtype),
144 state.get_hyper("beta2", grad.dtype.base_dtype),
155 beta2_t = state.get_hyper("beta2", var.dtype.base_dtype)
164 # v_t = beta2 * v + (1 - beta2) * (g_t * g_t
[all...] |
| adam_test.py | 44 beta2=0.999,
46 alpha_t = alpha * np.sqrt(1 - beta2**t) / (1 - beta1**t)
49 v_t = beta2 * v + (1 - beta2) * g_t * g_t
|
| /external/tensorflow/tensorflow/core/kernels/ |
| training_ops_gpu.cu.cc | 133 typename TTypes<T>::ConstScalar beta2,
144 v + (beta2.constant(one) - beta2).reshape(single).broadcast(bcast) *
177 typename TTypes<T>::ConstScalar beta2,
188 v + (beta2.constant(one) - beta2).reshape(single).broadcast(bcast) *
208 typename TTypes<T>::ConstScalar beta2,
219 (beta2.reshape(single).broadcast(bcast) * v).cwiseMax(grad.abs());
|
| training_ops.h | 146 typename TTypes<T>::ConstScalar beta2,
160 typename TTypes<T>::ConstScalar beta2,
172 typename TTypes<T>::ConstScalar beta2,
|
| training_ops_test.cc | 176 auto beta2 = Scalar(g, 0.99); local
181 {var, m, v, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad});
|
| training_ops.cc | 310 typename TTypes<T>::ConstScalar beta2,
316 // beta2 == ?
321 v.device(d) += (grad.square() - v) * (T(1) - beta2());
336 T beta1_power, T beta2_power, T lr, T beta1, T beta2,
341 v.device(d) += (grad.square() - v) * (T(1) - beta2);
359 typename TTypes<T>::ConstScalar beta2,
366 v.device(d) += (grad.square() - v) * (T(1) - beta2());
379 typename TTypes<T>::ConstScalar beta2,
384 v.device(d) = (beta2() * v).cwiseMax(grad.abs());
2840 const Tensor& beta2 = ctx->input(7); variable
2938 T beta2 = 0; variable
3109 const Tensor& beta2 = ctx->input(8); variable
3243 const Tensor& beta2 = ctx->input(6); variable
[all...] |
| /external/tensorflow/tensorflow/python/tpu/ |
| tpu_embedding.py | 154 beta2=0.999,
165 beta2: A float value.
181 if beta2 < 0. or beta2 >= 1.:
182 raise ValueError('beta2 must be between 0. and 1; got {}.'.format(beta2))
190 self.beta2 = beta2
[all...] |
