deep_logistic: Deep Learning Classification with Automated Parameter Tuning
In tianwei-zhang/easyAI: Democratize AI

Deep Learning Classification with Automated Parameter Tuning

deep_logistic(x, y, option, num_layer = seq(1, 3, 1), max_units = NULL,
  start_unit = 5, max_dropout = 0.2, min_dropout = 0, max_lr = 0.2,
  min_lr = 0.001, iteration_per_layer = 10, validation_split = 0.2,
  num_epoch = 10, num_patience = 3, machine_type = "standard")

`x`	training feature matrix
`y`	target matrix
`option`	either local or google for hyper-parameter tuning
`num_layer`	a vector of integers indicating the number of hidden layers to test. Default to seq(1,3,1)
`max_units`	the maximum number of hidden units in a layer. Default to an optimized value based on data
`start_unit`	the minimum number of hiddent units in a layer. Default to 5
`max_dropout`	A number between 0 and 1 indicating the maximum dropoff rate in a layer. Default to 0.2
`min_dropout`	A number between 0 and 1 indicating the minimum dropoff rate in a layer. Default to 0
`max_lr`	maximum learning rate in a run. Default to 0.2
`min_lr`	minimum learning rate in a run. Default to 0.001
`iteration_per_layer`	Number of parameter randomizations for a given number of hidden layers. More iterations will explore a larger parameter space
`validation_split`	Percent of data used for validation. Default to 20 percent
`num_epoch`	number of epoches to go through during training. Default to 10
`num_patience`	number of patience in early stopping criteria. Default to 3
`machine_type`	type of server to use. Could be standard, standard_gpu, standard_p100. For more visit https://cloud.google.com/ml-engine/docs/training-overview#machine_type_table