Advances, Systems and Applications
From: Online architecture for predicting live video transcoding resources
Model | Parameters |
---|---|
RL | Neural network = Input: 25 Integers; 3*(32 Unit layers+RELUs), output: Integer (linear activation) DQN Agent (target_model_update = 1e-3, nb_steps_warmup = 50, policy = Boltzmann Q Policy), Adam Optimizer (learning rate = 1e-2), Training steps = 4000 Reward: 0.4 (goal achieved), −0.5 (goal not achieved) |
RF | RandomForestRegressor(n_estimators = 100) |
SGD | SGDRegressor (max_iter = 1000) |