L1Decay

paddle.fluid.regularizer.L1Decay(regularization_coeff=0.0)

L1Decay实现L1权重衰减正则化，用于模型训练，使得权重矩阵稀疏。

该类生成的实例对象，需要设置在 ParamAttr 或者 optimizer (例如 SGDOptimizer )中，在 ParamAttr 中设置时， 只对该网络层中的参数生效；在 optimizer 中设置时，会对所有的参数生效；如果同时设置， 在 ParamAttr 中设置的优先级会高于在 optimizer 中设置。

具体实现中，L1权重衰减正则化的计算公式如下：

L1Decay - 图1

参数

  • regularization_coeff (float) – L1正则化系数，默认值为0.0。

代码示例 1

  1. import paddle.fluid as fluid
  3. main_prog = fluid.Program()
  4. startup_prog = fluid.Program()
  5. with fluid.program_guard(main_prog, startup_prog):
  6.     data = fluid.layers.data(name='image', shape=[3, 28, 28], dtype='float32')
  7.     label = fluid.layers.data(name='label', shape=[1], dtype='int64')
  8.     hidden = fluid.layers.fc(input=data, size=128, act='relu')
  9.     prediction = fluid.layers.fc(input=hidden, size=10, act='softmax')
  10.     loss = fluid.layers.cross_entropy(input=prediction, label=label)
  11.     avg_loss = fluid.layers.mean(loss)
  12. optimizer = fluid.optimizer.Adagrad(
  13.     learning_rate=1e-4,
  14.     regularization=fluid.regularizer.L1Decay(
  15.         regularization_coeff=0.1))
  16. optimizer.minimize(avg_loss)

代码示例 2

  1. # 在 ParamAttr 和 optimizer 中同时设置正则化
  2. import paddle.fluid as fluid
  3. l1 = fluid.regularizer.L1Decay(regularization_coeff=0.1)
  4. l2 = fluid.regularizer.L2Decay(regularization_coeff=0.1)
  5. x = fluid.layers.uniform_random([3,4])
  7. # 在ParamAttr中设置L1正则化
  8. w_param = fluid.ParamAttr(regularizer=l1)
  9. hidden1 = fluid.layers.fc(x, 8, param_attr=w_param)    # fc_0.w_0(L1), fc_0.b_0
  10. hidden2 = fluid.layers.fc(hidden1, 16, param_attr=w_param)   # fc_1.w_0(L1), fc_1.b_0
  11. predict = fluid.layers.fc(hidden2, 32)     # fc_3.w_0, fc_3.b_0
  12. avg_loss = fluid.layers.mean(predict)
  14. # 在optimizer中设置L2正则化
  15. optimizer = fluid.optimizer.SGD(learning_rate=1e-4, regularization=l2)
  16. optimizer.minimize(avg_loss)
  18. # 将会打印出提示信息:
  19. # Regularization of [fc_0.w_0, fc_1.w_0] have been set by ParamAttr or WeightNormParamAttr already.
  20. # So, the Regularization of Optimizer will not take effect for these parameters!