| /external/tensorflow/tensorflow/contrib/opt/python/training/ |
| D | ggt_test.py | 45 m_t = m * beta1 + (1 - beta1) * g_t
46 grad_buffer[((t - 1) % window), :] = m_t
57 np.transpose(m_matrix), m_t
63 new_step += (m_t - np.linalg.multi_dot([
67 np.transpose(m_matrix), m_t
71 return param_t, m_t, grad_buffer
105 m_t = opt._get_moment1()
108 self.assertTrue(m_t is not None)
111 self.assertIn(m_t, opt_variables)
125 m_t = opt._get_moment1()
[all …]
|
| D | nadam_optimizer.py | 83 m_t = state_ops.assign(m, m * beta1_t, use_locking=self._use_locking)
84 with ops.control_dependencies([m_t]):
85 m_t = scatter_add(m, indices, m_scaled_g_values)
87 m_bar = m_scaled_g_values + beta1_t * array_ops.gather(m_t, indices)
|
| D | adam_gs_optimizer.py | 191 m_t = state_ops.assign(m, m * beta1_t, use_locking=self._use_locking)
192 with ops.control_dependencies([m_t]):
193 m_t = scatter_add(m, indices, m_scaled_g_values)
202 var, lr * m_t / (v_sqrt + epsilon_t), use_locking=self._use_locking)
203 return control_flow_ops.group(*[var_update, m_t, v_t])
|
| D | lazy_adam_gs_optimizer.py | 62 m_t = state_ops.scatter_update(m, grad.indices,
75 m_t_slice = array_ops.gather(m_t, grad.indices)
81 return control_flow_ops.group(var_update, m_t, v_t)
|
| D | lazy_adam_optimizer.py | 62 m_t = state_ops.scatter_update(m, grad.indices,
75 m_t_slice = array_ops.gather(m_t, grad.indices)
81 return control_flow_ops.group(var_update, m_t, v_t)
|
| D | adamax_test.py | 46 m_t = beta1 * m + (1 - beta1) * g_t
48 param_t = param - (alpha / (1 - beta1**t)) * (m_t / (v_t + epsilon))
49 return param_t, m_t, v_t
62 m_t, v_t, param_t = np.copy(m), np.copy(v), np.copy(param)
67 m_t[indices] = m_t_slice
70 return param_t, m_t, v_t
|
| D | addsign.py | 141 m_t = state_ops.assign(
147 array_ops.gather(math_ops.sign(m_t), sign_g.indices) * sign_g.values,
168 return control_flow_ops.group(* [var_update, m_t])
|
| D | powersign.py | 144 m_t = state_ops.assign(
150 array_ops.gather(math_ops.sign(m_t), sign_g.indices) * sign_g.values,
172 return control_flow_ops.group(* [var_update, m_t])
|
| D | nadam_optimizer_test.py | 43 m_t = beta1 * m + (1 - beta1) * g_t
46 m_bar = (1 - beta1) * g_t + beta1 * m_t
49 return param_t, m_t, v_t
|
| /external/clang/test/PCH/ |
| D | cxx-member-init.cpp | 19 struct m_t struct
22 m_t() { } in m_t() argument
33 m_t *test() { in test()
34 return new m_t; in test()
|
| /external/tensorflow/tensorflow/python/keras/optimizer_v2/ |
| D | nadam.py | 150 m_t = beta_1_t * m + (1 - beta_1_t) * grad
151 m_t = state_ops.assign(m, m_t, use_locking=self._use_locking)
152 m_t_prime = m_t / (1. - self.m_schedule_next)
174 m_t = state_ops.assign(m, m * beta_1_t, use_locking=self._use_locking)
175 with ops.control_dependencies([m_t]):
176 m_t = self._resource_scatter_add(m, indices, m_scaled_g_values)
177 m_t_slice = array_ops.gather(m_t, indices)
|
| D | adam.py | 218 m_t = state_ops.assign(m, m * beta_1_t, use_locking=self._use_locking)
219 with ops.control_dependencies([m_t]):
220 m_t = self._resource_scatter_add(m, indices, m_scaled_g_values)
232 var, lr * m_t / (v_sqrt + epsilon_t), use_locking=self._use_locking)
233 return control_flow_ops.group(*[var_update, m_t, v_t])
243 lr * m_t / (v_hat_sqrt + epsilon_t),
245 return control_flow_ops.group(*[var_update, m_t, v_t, v_hat_t])
|
| D | adam_test.py | 49 m_t = beta1 * m + (1 - beta1) * g_t
52 param_t = param - lr_t * m_t / (np.sqrt(v_t) + epsilon)
53 return param_t, m_t, v_t
68 m_t = beta1 * m + (1 - beta1) * g_t
72 param_t = param - lr_t * m_t / (np.sqrt(vhat_t) + epsilon)
73 return param_t, m_t, v_t, vhat_t
87 m_t, v_t, vhat_t, param_t = (np.copy(m), np.copy(v), np.copy(vhat),
92 m_t[indices] = m_t_slice
99 return param_t, m_t, v_t, vhat_t
|
| D | adamax_test.py | 45 m_t = beta1 * m + (1 - beta1) * g_t
47 param_t = param - (alpha / (1 - beta1**(t + 1))) * (m_t / (v_t + epsilon))
48 return param_t, m_t, v_t
61 m_t, v_t, param_t = np.copy(m), np.copy(v), np.copy(param)
66 m_t[indices] = m_t_slice
69 return param_t, m_t, v_t
|
| /external/tensorflow/tensorflow/contrib/optimizer_v2/ |
| D | adam.py | 161 m_t = state_ops.assign(m, m * beta1_t, use_locking=self._use_locking)
162 with ops.control_dependencies([m_t]):
163 m_t = scatter_add(m, indices, m_scaled_g_values)
172 var, lr * m_t / (v_sqrt + epsilon_t), use_locking=self._use_locking)
173 return control_flow_ops.group(*[var_update, m_t, v_t])
|
| /external/tensorflow/tensorflow/python/training/ |
| D | adam.py | 191 m_t = state_ops.assign(m, m * beta1_t, use_locking=self._use_locking)
192 with ops.control_dependencies([m_t]):
193 m_t = scatter_add(m, indices, m_scaled_g_values)
202 var, lr * m_t / (v_sqrt + epsilon_t), use_locking=self._use_locking)
203 return control_flow_ops.group(*[var_update, m_t, v_t])
|
| /external/tensorflow/tensorflow/compiler/tests/ |
| D | powersign_test.py | 48 m_t = beta * m + (1 - beta) * g_t
53 multiplier = base ** (sign_decayed * np.sign(g_t) * np.sign(m_t))
55 return params_t, m_t
|
| D | addsign_test.py | 47 m_t = beta * m + (1 - beta) * g_t
52 multiplier = alpha + sign_decayed * np.sign(g_t) * np.sign(m_t)
54 return params_t, m_t
|
| D | adamax_test.py | 42 m_t = beta1 * m + (1 - beta1) * g_t
44 param_t = param - (alpha / (1 - beta1**t)) * (m_t / (v_t + epsilon))
45 return param_t, m_t, v_t
|
| /external/tensorflow/tensorflow/core/api_def/base_api/ |
| D | api_def_ResourceApplyPowerSign.pbtxt | 55 m_t <- beta1 * m_{t-1} + (1 - beta1) * g
56 update <- exp(logbase * sign_decay * sign(g) * sign(m_t)) * g
|
| D | api_def_ApplyPowerSign.pbtxt | 61 m_t <- beta1 * m_{t-1} + (1 - beta1) * g
62 update <- exp(logbase * sign_decay * sign(g) * sign(m_t)) * g
|
| D | api_def_ResourceApplyAdaMax.pbtxt | 68 m_t <- beta1 * m_{t-1} + (1 - beta1) * g
70 variable <- variable - learning_rate / (1 - beta1^t) * m_t / (v_t + epsilon)
|
| D | api_def_ApplyAdaMax.pbtxt | 74 m_t <- beta1 * m_{t-1} + (1 - beta1) * g
76 variable <- variable - learning_rate / (1 - beta1^t) * m_t / (v_t + epsilon)
|
| D | api_def_ResourceApplyAdam.pbtxt | 80 $$m_t := beta_1 * m_{t-1} + (1 - beta_1) * g$$
82 $$variable := variable - lr_t * m_t / (\sqrt{v_t} + \epsilon)$$
|
| /external/tensorflow/tensorflow/python/keras/ |
| D | optimizers.py | 503 m_t = (self.beta_1 * m) + (1. - self.beta_1) * g
507 p_t = p - lr_t * m_t / (K.sqrt(vhat_t) + self.epsilon)
510 p_t = p - lr_t * m_t / (K.sqrt(v_t) + self.epsilon)
512 self.updates.append(state_ops.assign(m, m_t))
592 m_t = (self.beta_1 * m) + (1. - self.beta_1) * g
594 p_t = p - lr_t * m_t / (u_t + self.epsilon)
596 self.updates.append(state_ops.assign(m, m_t))
682 m_t = self.beta_1 * m + (1. - self.beta_1) * g
683 m_t_prime = m_t / (1. - m_schedule_next)
689 self.updates.append(state_ops.assign(m, m_t))
|