abhinav

-module(markov_distributed).
-compile(export_all).
-author({jha, abhinav}).
-define(INITPOINT, 50).
-define(ARITY, 3).

start() -> start("corpus.txt").
start(Corpus) ->
    {ok, Bin} = file:read_file(Corpus),
    Words = string:tokens(binary_to_list(Bin), "\n "),
    Tab = ets:new(words, [bag]),
    form_dictionary(Words, Tab, ?ARITY),
    Start = mrandom(?INITPOINT),
    State0 = lists:sublist(Words, Start, ?ARITY - 1),
    Sentence = spew(Tab, State0, []),
    ets:delete(Tab),
    io:format("~p~n", [string:join(Sentence, " ")]).

spew(Tab, [_|L]=State, Acc)->
    % End at a word that ends with a fullstop, once we're greater than 20 words.
    case (length(Acc) >= 15) and (string:rchr(lists:last(L), $.) =:= length(lists:last(L))) of
        true -> lists:reverse(Acc);
        false ->
            Objects = ets:lookup(Tab, list_to_tuple(State)),
            {_,Snew} = lists:nth(mrandom(length(Objects)), Objects),
            spew(Tab, L ++ [Snew], [Snew | Acc])
    end.

mrandom(Max) ->
    {A, B, C} = now(),
    random:seed(A,B,C),
    random:uniform(Max).

form_dictionary([], _, _) -> void;
form_dictionary([H|T], Tab, N)->
    Self = self(),
    Pids = [spawn(fun()->ngrams(lists:nthtail(Y-1, [H|T]), N, [], Self) end) || Y<- lists:seq(1, length([H|T]))],
    gather(Tab, length(Pids)).

gather(_, 0) -> void;
gather(Tab, X)->
    receive
        {ok, Pid, Ngram} ->
            ets:insert(Tab, Ngram),
            %io:format("Received ~p from ~p. X = ~p ~n", [Ngram, Pid, X]),
            gather(Tab, X-1);
        _-> gather(Tab, X-1)
    end.


ngrams([], _, _, Self) -> Self!void;
ngrams([H|_], 1, Acc, Self)->
    %io:format("~p: ~p~n", [self(), {list_to_tuple(lists:reverse(Acc)), H}]),
    Self ! {ok, self(), {list_to_tuple(lists:reverse(Acc)), H}};
ngrams([H|T], X, Acc, Self)-> ngrams(T, X-1, [H|Acc], Self).