# Smart Learning [SmartLearning] (c) (r) (tm)
# Copyright Dean Van Greunen 2018 to infinity, All rights Reserved.


# Smart Learning is an Artificial Intelligence System which was designed and inspired by:
#   - The Human Brain
#   - The Deep Learning System
#   - The MatLab Programming Language
#   - Javascript Programming Language with Server Code Execution design of Node.js
#   - The Go Programming Language
# Author: Dean Van Greunen
# Email: deanvg9000@gmail.com
# 027 885 5371 (for South Africa local calls Only)
# +27 72 885 5371  (for International calls)

class SmartLearning:
    #################
    # NN class Code #   (this is a SmartLearning NN)
    #################
    # Neural Network, Simple and elegant, unconventional,
    # however it is:
    #  - organized
    #  - well formated
    #  - easy to (while running):
    #       - import
    #       - export
    #       - modify
    #  - supports:
    #       - Unbounded distance between neurons
    #       - Unbounded hooks (neurons, talking to other neurons, more just means more processing) [only allowed math ops are: + - / *], if more complicated functions are need
    #         then that means it will be like such,
    #           [
    #               {
    #                   "a": {
    #                       "b":"a+1",
    #                       "c":"b*0.90"
    #                   },
    #                   "d": {
    #                       "e":"c+1",
    #                       "f":"e-0.90"
    #                   },
    #               },
    #           ]
    #           This NN will take a single node input from "a" and get our output for the required f in this example the value can be anything and it will evaluate to:
    #               ((((a+1)*0.90)+1)-0.90)
    #               let a = 2:
    #                   ((((2+1)*0.90)+1)-0.90) = ((((3)*0.90)+1)-0.90) = ((2.7+1)-0.90) = (3.7 - 0.90) = (2.8) = 2.8
    #           after x time, it could be optimized by access to:
    #           [
    #               {
    #                   "a": {
    #                       "f":"(((a+1)*0.90)+1)-0.90"
    #                   },
    #               },
    #           ]
    #           this access time would go from, 8 node accesses to 2 which is
    class NN:
        # format of
        # NN = [{"x": {"y"}}] or [{"x": {"y":"y+1", "y+n":"y+n*2.90"}}]
        # NN = [
        #       {"x": {"y":"x*0.87"}},   # maps x to y which is mapped by (x * 0.87)
        #       {"x+1": {"y":"x*0.67"}}  # maps x+1 to y (not math but the Xn var...) which is mapped by (x * 0.67) as a math op using parent var...
        #   ]
        # or
        #  [
        #       {"x": {"y":"y+1", "y+n":"y*0.90"},
        #       {"x+1": {"y":"y+1", "y+n":"y*1.09"}} # maps x+1 to y (not math but the Xn var...)
        #  ]
        working_neurons = []    # Working Memory, always optimized for performance, has pointers to memory_neurons as mem_x or mem_y, name for x and y are not relevant and can change,
                                # as it is only a mathematical function.
        hard_memory_neurons = []        # Memory, not optimized, when saving the working neurons it is then un-optimized and then added to memory.
        def reset(self): # good for wiping / clear working memory, a.k.a temp memory, it will reset to what it was just before "waking up"
            self.neurons = []   # Clears ALL Neurons In Memory, a.k.a "Black Out"

        def wipe(self): # WARNING HARD RESET, MAKES A COMPLETELY EMPTY NN, like a brand new NN, useful if NN has gone rouge.
            self.hard_memory_neurons = []  # Clears ALl Neurons in Old
            self.reset()

        def map(self, n_x, n_y, steps=1): # Solves for Y by X using brute force in steps, if set to 1, it will do a initial rough, y = x from current memory.
            #                               Note with this method "learning" takes as long as a human would with the exception of speed which an electronic computing device would provide
            #                               Over what a biological computing device would perform at (the brain, human or not)
            # TODO: Map Properly, as [X->Y]....Must perform a search, each output is mapped to its unique input, etc... and any math logic is re-used
            pass # This mapping function is not available in this version of the code, please contact deanvg9000@gmail.com for help or support or donations for this project

        def optimize(): # this optimizes memory into working memory.
            pass # This optimiz(e/ing) function is not available in this version of the code, please contact deanvg9000@gmail.com for help or support or donations for this project

        def save(self, soft_save=true, hard_save=true, path="", filename="default.pyai-nn"):
            # Soft Load, loads only working memory.
            if soft_save:
                self.soft_save(path, filename)
            # Hard Load, Loads
            if hard_save:
                self.hard_save(path, filename)

        def soft_load(self, path="", filename="default.pyai-nn"):
            # Loads Optimized data
            f_mem_opt_data = open(path + filename + "-opt", "w+")
            f_mem_opt_data.write(toJSON(self.working_neurons))
            f_mem_opt_data.close()

        def hard_load(self, path="", filename="default.pyai-nn"):
            # Loads Complete UnOptimized Data
            f_mem_data = open(path + filename + "-mem", "w+")
            f_mem_data.write(toJSON(self.hard_memory_neurons))
            f_mem_data.close()

        def load(self, soft_load=true, hard_load=true, path="", filename="default.pyai-nn"):
            # Soft Load, loads only working memory.
            if soft_load:
                self.soft_load(path, filename)
            # Hard Load, Loads
            if hard_load:
                self.hard_load(path, filename)

        def soft_load(self, path="", filename="default.pyai-nn"):
            # Loads Optimized data
            f_mem_opt_data = open(path + filename + "-opt", "w+")
            mem_opt_data = f_mem_opt_data.read()
            f_mem_opt_data.close()
            self.working_neurons = fromJSON(mem_opt_data)

        def hard_load(self, path="", filename="default.pyai-nn"):
            # Loads Complete UnOptimized Data
            f_mem_data = open(path + filename + "-mem", "w+")
            mem_data = f_mem_data.read()
            f_mem_data.close()
            self.hard_memory_neurons = fromJSON(mem_data)

        def smartlearn(self):
            pass

    ####################
    # EVent class Code #
    ####################
    # Event, Clever way to exchange data and format it without actually changing it, just the way it is perceived.
    class Event:
        chains = []  # All events to chain with this event
        func = None  # function to call after formating is complete
        formatting = None  # formatting function to call, passes in data, expects data
        def init(self, func=None, formatting=None, chains=[]):  # Events can be empty
            # This creates a data formating matrix, which takes the func to be called and applies the formatting as matrix which converts the data, events can be chained.
            if func:
                self.func = func
            else:
                self.func = self.ret
            if formatting:
                self.formatting = formatting
            else:
                self.formatting = self.ret
            self.chains = chains

        def bindEvent(self, event):  # binds an event to chain (stack first in, first called)
            self.chains[] = event

        def call(self, data):
            if len(self.chains) >= 1:  # check if there are items in the chain
                for ev in self.chains: # loop through events in chain
                    if type(ev) is SmartLearning().Event():  # check if item in chain is an event else skip it
                        ev.call(data)  # if any Subscribing Events have chains they will all be called as well.
            self.func(self.formatting(data)) # Data is not destroyed

        def ret(self, x):
            return x

    #####################
    # Input class Code  #
    #####################
    # Input which has a NN, which monitors the current environment
    class Input:
        parent_events = SmartLearning().Event()
        stream = None
        queue = []

        def init(self):
            startThread(self.notifier)

        def setStream(self, stream):
            self.stream = stream

        def bindEvent(self, event):
            self.parent_events.bindEvent(event)

        def notifier(self):
            while self.stream:
                if self.stream.canFetch():
                    self.queue = self.stream.fetch()
                    for i,q in self.queue:
                        self.stream.pop(q)
                        self.queue[i] = None


    #####################
    # Output class Code #
    #####################
    # Output which has a NN, affects the current environment
    class Output:
        parent_events = SmartLearning().Event()
        stream = None
        queue = []
        def init(self):
            startThread(self.sender)

        def setStream(self, stream):
            self.stream = stream

        def bindEvent(self, event):
            self.parent_events.bindEvent(event)

        def sender(self):
            while self.stream:
                if self.stream.canPush():
                    for i,q in self.queue:
                        self.stream.push(q)
                        self.queue[i] = None

    ##########################
    # Environment class Code #
    ##########################
    # Environment contains input and output which both have NNs
    class Environment:
        io = None
        parent_events = SmartLearning().Event()
        ev_notify = SmartLearning().Event()
        def init(self):
            self.io = {
                "i": SmartLearning().Input(),
                "o": SmartLearning().Output()
            }
            self.ev_notify.func = self.notify
            self.io["i"].init()
            self.io["o"].init()
            self.io["i"].bindEvent(self.ev_notify)
            self.io["o"].bindEvent(self.ev_notify)

        def bindEvent(self, event):
            self.parent_events.bindEvent(event)

        def notify(self, data):
            self.parent_events.call(data)


    #########################
    # Processing class Code #
    #########################
    # Processor contains the environment, which contains input and output which both have NNs
    class Processing:
        nn = None
        environment = None
        parent_events = None
        ev_notify = None
        def init(self):
            self.nn = SmartLearning().NN()
            self.environment = SmartLearning().Environment()
            self.parent_events = SmartLearning().Event()
            self.ev_notify = SmartLearning().Event()
            self.environment.init()

        def bindIO(self, i, o):
            self.environment.io["i"].setStream(i)
            self.environment.io["o"].setStream(o)

        def bindEvent(self, event):
            self.parent_events.bindEvent(event)

        def notify(self, data):
            self.parent_events.call(data)

    ###################
    # Main class Code #
    ###################
    processor = None
    parent_events = None
    ev_notify = None
    def init(self):
        self.processor = SmartLearning().Processing()
        self.parent_events = SmartLearning().Event()
        self.ev_notify = SmartLearning().Event()
        self.processor.init()

    def bindEvent(self, event):
        self.parent_events.bindEvent(event)

    def notify(self, data):
        self.parent_events.call(data)

    def loadStreams(self, i_stream, o_stream):
        self.processor.bindIO(i_stream, o_stream)

    def boot(self): # loads hard memory only (usually done a few times during AI Up time or after hitting the kill switch)
        self.processor.hard_memory_neurons = []
        self.processor.load(soft_load=False, hard_load=False)
        self.wake()

    def wake(self):  # loads working memory only.
        self.processor.working_neurons = []
        self.processor.load(soft_load=True, hard_load=False)

    def sleep(self):  # slowly pauses tasks one by one from least to most important,
        #               and then de-optimizes working memory and store in into hard memory,
        #               then saves hard memory and clears optimied memory
        self.processor.optimize()
        self.processor.save(soft_save=True, hard_save=False)
        self.processor.working_neurons = []

    def shutdown(self): # runs sleep and then clears both hard and working memory after saving, then it shutdown the AI completely.
        #                 useful for debugging and going through the "AI's brain" data files
        self.sleep()
        self.processor.save(soft_save=False, hard_save=True)
        self.processor.hard_memory_neurons = []
        quit()
        pass

    def kill(self): # Crashes AI, good kill switch for a rouge AI
        quit()



#############################################
#               USAGE DEMO                  #
#############################################
# Note Run the following in a different thread, while keeping access to sl_ai

# Create SmartLearning AI System
sl_ai = SmartLearning()
# Initialize SubSystems
sl_ai.init()

# Load The World Information Stream (o_stream)
# AI Actor/Agent Control Stream (i_stream)
sl_ai.loadStreams(i_stream=StreamObject, o_stream=StreamObject)

# Start Up AI for the first time (Booting)
sl_ai.boot()

# every 3 to 9 hours do a
# sl_ai.sleep()
# sl_ai.wake()

# The AI should be functional now. it will take rough 9 to 48 human months to learn how to speak, and walk.
# after 20 years roughly it will be able to program a version of it's self.