Finish markov chain generation impl

* Markov chains will train and generate chains correctly now
* Implement Markov.save_chains/0
* Add a couple more utils that help accomplish the above

Signed-off-by: Alek Ratzloff <alekratz@gmail.com>
This commit is contained in:
2020-07-15 16:25:25 -07:00
parent c43c075588
commit 4c93b42fdc
5 changed files with 150 additions and 21 deletions

View File

@@ -1,5 +1,5 @@
defmodule Omnibot.Contrib.Markov.Chain do
alias Omnibot.Contrib.Markov.Chain
alias Omnibot.{Contrib.Markov.Chain, Util}
@enforce_keys [:order]
defstruct order: 2, chain: []
@@ -14,17 +14,18 @@ defmodule Omnibot.Contrib.Markov.Chain do
Enum.filter(words, &(String.length(&1) > 0))
|> Enum.chunk_every(order + 1, 1) # this gives us a "sliding window" effect
|> Enum.reduce(chain, &case Enum.split(&1, order) do
{words, []} -> if length(words) == order,
# Null case for the chain; this is an "end" state
do: add_weight(&2, words, nil),
else: &2 # TODO ? train [a, nil] -> b ?
{words, []} when length(words) == order -> add_weight(&2, words, nil)
{words, []} -> add_weight(&2, Util.pad_trailing(words, nil, order), nil)
{words, [next]} -> add_weight(&2, words, next)
end)
end
def add_weight(%Chain {chain: chain, order: order}, key, word, increment \\ 1) do
if length(key) != order, do: raise(ArgumentError, message: "invalid key (length #{length(key)} vs. order #{order})")
chain = case Enum.find_index(chain, fn {listkey, _} -> listkey == key end) do
if length(key) != order do
raise(ArgumentError, message: "invalid key (length #{length(key)} vs. order #{order})")
end
chain = case find_index(chain, key) do
# Insert weight
nil -> [{key, %{word => increment}} | chain]
# Update weight
@@ -36,4 +37,38 @@ defmodule Omnibot.Contrib.Markov.Chain do
end
%Chain{chain: chain, order: order}
end
def get(chain, key) when is_list(chain) do
item = Enum.find(chain, fn {listkey, _} -> listkey == key end)
case item do
nil -> nil
{_, weights} -> weights
end
end
def get(chain, key), do: get(chain.chain, key)
def find_index(chain, key) when is_list(chain) do
Enum.find_index(chain, fn {listkey, _} -> listkey == key end)
end
def find_index(chain = %Chain{}, key), do: find_index(chain.chain, key)
def generate(chain) do
{seed, _} = Stream.filter(chain.chain, fn {key, _} -> length(key) == chain.order end)
|> Enum.random()
generate(chain, seed)
end
def generate(chain, key) do
do_generate(chain, key) |> Enum.join(" ")
end
defp do_generate(_chain, [nil | _]), do: []
defp do_generate(chain, key) do
weights = get(chain, key) || []
[next | key] = key ++ [Util.weighted_random(weights)]
[next | do_generate(chain, key)]
end
end

View File

@@ -1,28 +1,44 @@
defmodule Omnibot.Contrib.Markov do
use Omnibot.Plugin
alias Omnibot.Contrib.Markov.Chain
alias Omnibot.{Contrib.Markov.Chain, Util}
require Logger
@default_config path: "markov", order: 2, save_every: 5 * 60
@default_config path: "markov.ets", order: 2, save_every: 5 * 60
command "!markov", ["force"] do
# Choose a random value from the sender
Irc.send_to(irc, channel, "TODO")
end
command "!markov", ["all"] do
Irc.send_to(irc, channel, "TODO")
end
command "!markov", ["status"] do
Irc.send_to(irc, channel, "TODO")
end
@impl true
def children(cfg) do
[{Task, fn ->
Stream.timer(cfg[:save_every] * 1000)
|> Stream.cycle()
|> Stream.each(fn _ -> save_chains() end)
|> Stream.run()
end}]
[
{Task, fn ->
Stream.timer(cfg[:save_every] * 1000)
|> Stream.cycle()
|> Stream.each(fn _ -> save_chains() end)
|> Stream.run()
end}
]
end
@impl true
def on_init(_cfg) do
# Create the markov database
path = String.to_atom(cfg()[:path])
{:ok, db} = :dets.open_file(path, [:named_table])
chains = :ets.new(:markov_chains, [:public])
{:ok, db} = :dets.open_file(path, [])
chains = :ets.new(:markov_chains, [:named_table, :public])
:dets.to_ets(db, chains)
:dets.close(db)
:ok = :dets.close(db)
chains
end
@impl true
@@ -48,7 +64,7 @@ defmodule Omnibot.Contrib.Markov do
db = state()
case user_chain(channel, user) do
nil -> :ets.insert_new(db, {{channel, user}, chain})
chain -> :ets.insert(db, {{channel, user}, chain})
_old_chain -> :ets.insert(db, {{channel, user}, chain})
end
end
@@ -58,7 +74,14 @@ defmodule Omnibot.Contrib.Markov do
end
def save_chains() do
# TODO
Logger.info("Saved markov chains")
start = Util.now_unix()
Logger.debug("Saving markov chains")
{:ok, db} = :dets.open_file(cfg()[:path], [])
:ets.to_dets(state(), db)
:ok = :dets.close(db)
stop = Util.now_unix()
Logger.info("Saved markov chains in #{stop - start} seconds")
end
end