flatnet

Flatnet is an attempt to create a machine learning algorithm that reduces a standard model down to a list of actions to execute on the model outputs, rather than a system of nodes and links.

npm i flatnet

TLDR: To quickly jump in and start training a network on data, see Train

Module

{Model,Actions,Network,Train,Ext}=require('flatnet)

• Model
• Actions
• Network
• Train
• Ext

Command Line

• Simple command line functionality
• Entering with no args node flatnet.js will print command line paths available:

  ~> node flatnet.js
  
      Flatnet:
          cli root: 'actions' 'eval'
          flags: 'ptr_count=2'

• If using the flag ptr_count=int, it must be anywhere after the 3rd argument: actions or model
  • Default ptr_count=2 if not entered
  • For more info on ptr_count flag, see Actions.New and Model.New

  ~> node flatnet.js actions ptr_count=1 args
  ~> node flatnet.js actions args ptr_count=2
  
  ~> node flatnet.js model ptr_count=3 arg1 arg2
  ~> node flatnet.js model arg1 ptr_count=4 arg2
  ~> node flatnet.js model arg1 arg2 ptr_count=5

• node flatnet.js actions args
  • Also accessed by node actions.js, see Actions._CLI
  • If args = 0, will print and return actions.length
  • If args > 0, will print names of actions by integer lookup
  • Returns action array
• node flatnet.js model arg1 arg2
  • Also accessed by node model.js, see Model._CLI
  • Accepts arguments as comma separated strings
    • node model.js "0,1,2,3,4" "0,1,2,3,4"
  • Or will accept map argument as a file name without a file extension
    • node model.js mymapfile "0,1,2,3,4"
  • Prints the model.out array
  • Returns the model object