Markov (markov v4.1.3)
Public API
Example workflow:
# The model will be stored under this path
{:ok, model} = Markov.load("./model_path", sanitize_tokens: true, store_log: [:train])
# train using four strings
:ok = Markov.train(model, "hello, world!")
:ok = Markov.train(model, "example string number two")
:ok = Markov.train(model, "hello, Elixir!")
:ok = Markov.train(model, "fourth string")
# generate text
{:ok, text} = Markov.generate_text(model)
IO.puts(text)
# commit all changes and unload
Markov.unload(model)
# these will return errors because the model is unloaded
# Markov.generate_text(model)
# Markov.train(model, "hello, world!")
# load the model again
{:ok, model} = Markov.load("./model_path")
# enable probability shifting and generate text
:ok = Markov.configure(model, shift_probabilities: true)
{:ok, text} = Markov.generate_text(model)
IO.puts(text)
# print uninteresting stats
model |> Markov.dump_partition(0) |> IO.inspect
model |> Markov.read_log |> IO.inspect
# this will also write our new just-set option
Markov.unload(model)
Link to this section Summary
Types
Model options that could be set during creation in a call to load/3
or with configure/2
If data was tagged when training, you can use tag queries to alter the probabilities of certain generation paths
Functions
Reconfigures a loaded model. See model_option/0
for a thorough
description of the options
Generates a string. Will raise an exception if the model was trained on non-textual tokens at least once
Generates a list of tokens
Gets the configuration of a loaded model
Loads an existing model under path path
. If none is found, a new model with
the specified options will be created and loaded, and if that fails, an error
will be returned.
Reads the log file and returns a list of entries in chronological order
Trains model
using text or a list of tokens.
Unloads a loaded model
Link to this section Types
log_entry_type()
@type log_entry_type() :: :start | :end | :train | :gen
model_option()
@type model_option() :: {:store_log, [log_entry_type()]} | {:shift_probabilities, boolean()} | {:sanitize_tokens, boolean()} | {:order, integer()}
Model options that could be set during creation in a call to load/3
or with configure/2
:
store_log
: determines what data to put in the operation log, all of them by default::start
- model is loaded:end
- model is unloaded:train
: training requests:gen
: generation results
shift_probabilities
: gives less popular generation paths more chance to get used, which makes the output more original but may produce nonsense; false by defaultsanitize_tokens
: ignores letter case and punctuation when switching states, but still keeps the output as-is; false by default, can't be changed once the model is createdorder
: order of the chain, i.e. how many previous tokens the next one is based on; 2 by default, can never be changed once the model is created
@opaque model_reference()
tag_query()
@type tag_query() :: %{required(term()) => non_neg_integer()}
If data was tagged when training, you can use tag queries to alter the probabilities of certain generation paths
examples
Examples:
# training
iex> Markov.train(model, "hello earth", [
{:action, :saying_hello}, # <- terms of any type can function as tags
{:subject_type, :planet},
{:subject, "earth"},
:lowercase
])
:ok
iex> Markov.train(model, "Hello Elixir", [
{:action, :saying_hello},
{:subject_type, :programming_language},
{:subject, "Elixir"},
:uppercase
])
:ok
# simple generation - both paths have equal probabilities
iex> Markov.generate_text(model)
{:ok, "hello earth"}
iex> Markov.generate_text(model)
{:ok, "hello Elixir"}
# All generation paths have a score of 1 by default. Here we're telling
# Markov to add 1 point to paths tagged with `:uppercase`;
# "hello Elixir" now has a score of 2 and "hello earth" has a score of 1.
# Thus, "hello Elixir" has a probability of 2/3, and "hello earth" has
# that of 1/3
iex> Markov.generate_text(model, %{uppercase: 1})
{:ok, "hello Elixir"}
iex> Markov.generate_text(model, %{uppercase: 1})
{:ok, "hello Elixir"}
iex> Markov.generate_text(model, %{uppercase: 1})
{:ok, "hello earth"}
Link to this section Functions
configure(model, opts)
@spec configure(model :: model_reference(), opts :: [model_option()]) :: :ok | {:error, term()}
Reconfigures a loaded model. See model_option/0
for a thorough
description of the options
generate_text(model, tag_query \\ %{})
@spec generate_text(model_reference(), tag_query()) :: {:ok, binary()} | {:error, term()}
Generates a string. Will raise an exception if the model was trained on non-textual tokens at least once
iex> Markov.generate_text(model)
{:ok, "hello world"}
See type tag_query/0
for more info about tags
generate_tokens(model, tag_query \\ %{})
@spec generate_tokens(model_reference(), tag_query()) :: {:ok, [term()]} | {:error, term()}
Generates a list of tokens
iex> Markov.generate_tokens(model)
{:ok, ["hello", "world"]}
See type tag_query/0
for more info about tag_query
get_config(model)
@spec get_config(model :: model_reference()) :: {:ok, [model_option()]} | {:error, term()}
Gets the configuration of a loaded model
load(path, create_options \\ [])
@spec load(path :: String.t(), options :: [model_option()]) :: {:ok, model_reference()} | {:error, term()}
Loads an existing model under path path
. If none is found, a new model with
the specified options will be created and loaded, and if that fails, an error
will be returned.
read_log(model)
@spec read_log(model_reference()) :: [ %Markov.Operation{arg: term(), date_time: term(), type: term()} ]
Reads the log file and returns a list of entries in chronological order
iex> Markov.read_log(model)
{:ok,
[
%Markov.Operation{date_time: ~U[2022-10-02 16:59:51.844Z], type: :start, arg: nil},
%Markov.Operation{date_time: ~U[2022-10-02 16:59:56.705Z], type: :train, arg: ["hello", "world"]}
]}
train(model, text, tags \\ [:"$none"])
@spec train(model_reference(), String.t() | [term()], [term()]) :: :ok | {:error, term()}
Trains model
using text or a list of tokens.
:ok = Markov.train(model, "Hello, world!")
:ok = Markov.train(model, "this is a string that's broken down into tokens behind the scenes")
:ok = Markov.train(model, [
:this, "is", 'a token', :list, "where",
{:each_element, :is, {:taken, :as_is}},
:and, :can_be, :erlang.make_ref(), "<-- any term"
])
See tag_query/0
for more info about tags
unload(model)
@spec unload(model :: model_reference()) :: :ok
Unloads a loaded model