v1.1.00-unstable

[mochji]

NEW FEATURES

• New activation functions: ELU and Swish.
• Catch errors with invalid activation functions.
• Weights, biases and current tables are now stored with the rest of the neural network data. (_G[id]["weight"], _G[id]["bias"], _G[id]["current"] dictionary, might be removed on final release)
• New cost/error functions: Binary cross entropy and hinge loss.

DOCUMENTATION UPDATES

• Added new activation functions to documentation.
• Added new cost/error functions to documentation.
• Updated other stuff in documentation to match up with the new updates.
• Updated documentation to look better using markdown styling

OTHER STUFF

• Changed 'out' parameter on lnn.adjust() to 'output'.
• Changed 'expectedout' parameter on lnn.adjust() to 'expectedoutput'.

FUTURE PLANS

• Add the categorical cross entropy cost/error function in the next release. (v1.2.00)
• Group the activation functions into lnn.activation. (_G["lnn"]["activation"])

Releasing within the next 1-2 hours.