7

FORGE

the LED. To achieve this with Arduino code we can
create the two arrays to store ambient light sensor
data and LED state data. Using a variable to track
the current index of the arrays to write to next, and
a variable to track if the arrays have been filled yet,
we effectively create a circular array that will begin to
overwrite the old data when the whole array is filled.

const   int data_store_size =   10;

int sensor_val_store[data_store_size];

int state_store[data_store_size];

int store_cursor    =   0;

bool    store_filled    =   false;

void    save_data_point(int sensor_val, bool    led_

state){

        sensor_val_store[store_cursor]  =   sensor_val;

        state_store[store_cursor]   =   led_state;

        store_cursor++;

    if  (store_cursor   >=  data_store_size){

            store_filled    =   true;

            store_cursor    =   0;

}
}
The reading of the data is done when we want to
score a set of prediction model variables:

int score(int   thresh, bool    direction_is_greater){

        int correct_count   =   0;

        //  if  the store   array   isn’t   full,   then    only    read

upto    the store_cursor

        int count_to    =   store_filled    ?   data_store_size :

store_cursor;

        for (int    i   =   0;  i   <=  count_to;   i++){

                bool    pred    =   predict_with_params(sensor_val_

store[i],   thresh, direction_is_greater);

                if  (pred   ==  state_store[i]){

                        correct_count++;

}

}

        return  correct_count;

}

Finally, we can tie all of the processes together. First

we’ll take a reading of our light sensor and feed it into

our predictor function. Feed the output of the predictor

(which will be a Boolean representative if the LED

should be on or off) and set the LED to it. If the user

presses the toggle button, we want to save the new

data point and then run the model optimisation process

(because you have just added new data).

void    loop(){

        sensor_value    =   analogRead(sensor);

        led_state   =   predict(sensor_value);

        led_state=      predict(sensor_val, learnt_thresh,

learnt_flip_behaviour_flag)

        digitalWrite(led,   led_state);

        if  (button_pressed()){

                debounce_button();

                save_data_point(sensor_value,   !led_state);

                optimise();

}

}

Play around with a night light to see how the

system quickly learns whatever your night light

preferences are!

The full code can be found ready for your tinkering

delight over at hsmag.cc/sqSuaL.

Left

These days,

marketers tell us

that every item with

digital electronics

is smart, but can it

learn? Our light can!