How to set minimum learning rate inside pipeline.config (tensorflow 1.15 + object detection API)

Question

I'm training a model with this piece of pipeline.config:

  optimizer {
    momentum_optimizer: {
      learning_rate: {
        cosine_decay_learning_rate {
          learning_rate_base: 0.3
          total_steps: 200000
          warmup_learning_rate: 0.13333
          warmup_steps: 2000
        }
      }
      momentum_optimizer_value: 0.9
    }
    use_moving_average: false
  }

after 200k epochs the learning rate fall to 0

two questions:

• is there a way to set a minimum learning rate (ex. 0.0001) without reaching zero after the aforementioned 200k epochs?
• is there a way to add another bunch of epochs (ex. 100k) with a fixed learning rate in order to continue the training?

I'm using tensorflow 1.15, object_detection API and python 3.6

Answer 1

You can't. You can only do that with ExponentialDecayLearningRate. See the source.

message ExponentialDecayLearningRate {
  optional float initial_learning_rate = 1 [default = 0.002];
  optional uint32 decay_steps = 2 [default = 4000000];
  optional float decay_factor = 3 [default = 0.95];
  optional bool staircase = 4 [default = true];
  optional float burnin_learning_rate = 5 [default = 0.0];
  optional uint32 burnin_steps = 6 [default = 0];
> optional float min_learning_rate = 7 [default = 0.0];
}