ale-sys 0.1.2

# ale_python_interface.py
# Author: Ben Goodrich
# This directly implements a python version of the arcade learning
# environment interface.
__all__ = ["ALEInterface"]

import os
import sys
import warnings
import platform

import numpy as np

from ctypes import *
from numpy.ctypeslib import as_ctypes


def _load_cdll(path, name):
    if sys.platform.startswith("linux"):
        ext_suffix = ".so"
    elif sys.platform.startswith("darwin"):
        ext_suffix = ".dylib"
    elif sys.platform.startswith("win"):
        ext_suffix = ".dll"
    else:
        raise RuntimeError(
            'Platform "{}" not recognized while trying to resolve shared library'.format(
                sys.platform))

    library_format = "{}{}" if platform.system() == "Windows" else "lib{}{}"
    library_path = os.path.join(path, library_format.format(name, ext_suffix))

    try:
        return cdll.LoadLibrary(library_path)
    except Exception as ex:
        raise RuntimeError(
            "Failed to load library {}. Attempted to load {}.\n{}".format(
                name, library_path, ex))


ale_lib = _load_cdll(os.path.dirname(__file__), "ale_c")

ale_lib.ALE_new.argtypes = None
ale_lib.ALE_new.restype = c_void_p
ale_lib.ALE_del.argtypes = [c_void_p]
ale_lib.ALE_del.restype = None
ale_lib.getString.argtypes = [c_void_p, c_char_p]
ale_lib.getString.restype = c_char_p
ale_lib.getInt.argtypes = [c_void_p, c_char_p]
ale_lib.getInt.restype = c_int
ale_lib.getBool.argtypes = [c_void_p, c_char_p]
ale_lib.getBool.restype = c_bool
ale_lib.getFloat.argtypes = [c_void_p, c_char_p]
ale_lib.getFloat.restype = c_float
ale_lib.setString.argtypes = [c_void_p, c_char_p, c_char_p]
ale_lib.setString.restype = None
ale_lib.setInt.argtypes = [c_void_p, c_char_p, c_int]
ale_lib.setInt.restype = None
ale_lib.setBool.argtypes = [c_void_p, c_char_p, c_bool]
ale_lib.setBool.restype = None
ale_lib.setFloat.argtypes = [c_void_p, c_char_p, c_float]
ale_lib.setFloat.restype = None
ale_lib.loadROM.argtypes = [c_void_p, c_char_p]
ale_lib.loadROM.restype = None
ale_lib.act.argtypes = [c_void_p, c_int]
ale_lib.act.restype = c_int
ale_lib.game_over.argtypes = [c_void_p]
ale_lib.game_over.restype = c_bool
ale_lib.reset_game.argtypes = [c_void_p]
ale_lib.reset_game.restype = None
ale_lib.getAvailableModes.argtypes = [c_void_p, c_void_p]
ale_lib.getAvailableModes.restype = None
ale_lib.getAvailableModesSize.argtypes = [c_void_p]
ale_lib.getAvailableModesSize.restype = c_int
ale_lib.setMode.argtypes = [c_void_p, c_int]
ale_lib.setMode.restype = None
ale_lib.getAvailableDifficulties.argtypes = [c_void_p, c_void_p]
ale_lib.getAvailableDifficulties.restype = None
ale_lib.getAvailableDifficultiesSize.argtypes = [c_void_p]
ale_lib.getAvailableDifficultiesSize.restype = c_int
ale_lib.setDifficulty.argtypes = [c_void_p, c_int]
ale_lib.setDifficulty.restype = None
ale_lib.getLegalActionSet.argtypes = [c_void_p, c_void_p]
ale_lib.getLegalActionSet.restype = None
ale_lib.getLegalActionSize.argtypes = [c_void_p]
ale_lib.getLegalActionSize.restype = c_int
ale_lib.getMinimalActionSet.argtypes = [c_void_p, c_void_p]
ale_lib.getMinimalActionSet.restype = None
ale_lib.getMinimalActionSize.argtypes = [c_void_p]
ale_lib.getMinimalActionSize.restype = c_int
ale_lib.getFrameNumber.argtypes = [c_void_p]
ale_lib.getFrameNumber.restype = c_int
ale_lib.lives.argtypes = [c_void_p]
ale_lib.lives.restype = c_int
ale_lib.getEpisodeFrameNumber.argtypes = [c_void_p]
ale_lib.getEpisodeFrameNumber.restype = c_int
ale_lib.getScreen.argtypes = [c_void_p, c_void_p]
ale_lib.getScreen.restype = None
ale_lib.getRAM.argtypes = [c_void_p, c_void_p]
ale_lib.getRAM.restype = None
ale_lib.getRAMSize.argtypes = [c_void_p]
ale_lib.getRAMSize.restype = c_int
ale_lib.getScreenWidth.argtypes = [c_void_p]
ale_lib.getScreenWidth.restype = c_int
ale_lib.getScreenHeight.argtypes = [c_void_p]
ale_lib.getScreenHeight.restype = c_int
ale_lib.getScreenRGB.argtypes = [c_void_p, c_void_p]
ale_lib.getScreenRGB.restype = None
ale_lib.getScreenGrayscale.argtypes = [c_void_p, c_void_p]
ale_lib.getScreenGrayscale.restype = None
ale_lib.saveState.argtypes = [c_void_p]
ale_lib.saveState.restype = None
ale_lib.loadState.argtypes = [c_void_p]
ale_lib.loadState.restype = None
ale_lib.cloneState.argtypes = [c_void_p]
ale_lib.cloneState.restype = c_void_p
ale_lib.restoreState.argtypes = [c_void_p, c_void_p]
ale_lib.restoreState.restype = None
ale_lib.cloneSystemState.argtypes = [c_void_p]
ale_lib.cloneSystemState.restype = c_void_p
ale_lib.restoreSystemState.argtypes = [c_void_p, c_void_p]
ale_lib.restoreSystemState.restype = None
ale_lib.deleteState.argtypes = [c_void_p]
ale_lib.deleteState.restype = None
ale_lib.saveScreenPNG.argtypes = [c_void_p, c_char_p]
ale_lib.saveScreenPNG.restype = None
ale_lib.encodeState.argtypes = [c_void_p, c_void_p, c_int]
ale_lib.encodeState.restype = None
ale_lib.encodeStateLen.argtypes = [c_void_p]
ale_lib.encodeStateLen.restype = c_int
ale_lib.decodeState.argtypes = [c_void_p, c_int]
ale_lib.decodeState.restype = c_void_p
ale_lib.setLoggerMode.argtypes = [c_int]
ale_lib.setLoggerMode.restype = None


def _str_as_bytes(arg):
    if isinstance(arg, str):
        return arg.encode("utf-8")
    return arg


class ALEInterface(object):
    # Logger enum
    class Logger:
        Info = 0
        Warning = 1
        Error = 2

    def __init__(self):
        self.obj = ale_lib.ALE_new()

    def getString(self, key):
        return ale_lib.getString(self.obj, _str_as_bytes(key))

    def getInt(self, key):
        return ale_lib.getInt(self.obj, _str_as_bytes(key))

    def getBool(self, key):
        return ale_lib.getBool(self.obj, _str_as_bytes(key))

    def getFloat(self, key):
        return ale_lib.getFloat(self.obj, _str_as_bytes(key))

    def setString(self, key, value):
        ale_lib.setString(self.obj, _str_as_bytes(key), _str_as_bytes(value))

    def setInt(self, key, value):
        ale_lib.setInt(self.obj, _str_as_bytes(key), int(value))

    def setBool(self, key, value):
        ale_lib.setBool(self.obj, _str_as_bytes(key), bool(value))

    def setFloat(self, key, value):
        ale_lib.setFloat(self.obj, _str_as_bytes(key), float(value))

    def loadROM(self, rom_file):
        ale_lib.loadROM(self.obj, _str_as_bytes(rom_file))

    def act(self, action):
        return ale_lib.act(self.obj, int(action))

    def game_over(self):
        return ale_lib.game_over(self.obj)

    def reset_game(self):
        ale_lib.reset_game(self.obj)

    def getLegalActionSet(self):
        act_size = ale_lib.getLegalActionSize(self.obj)
        act = np.zeros((act_size), dtype=np.intc)
        ale_lib.getLegalActionSet(self.obj, as_ctypes(act))
        return act

    def getMinimalActionSet(self):
        act_size = ale_lib.getMinimalActionSize(self.obj)
        act = np.zeros((act_size), dtype=np.intc)
        ale_lib.getMinimalActionSet(self.obj, as_ctypes(act))
        return act

    def getAvailableModes(self):
        modes_size = ale_lib.getAvailableModesSize(self.obj)
        modes = np.zeros((modes_size), dtype=np.intc)
        ale_lib.getAvailableModes(self.obj, as_ctypes(modes))
        return modes

    def setMode(self, mode):
        ale_lib.setMode(self.obj, int(mode))

    def getAvailableDifficulties(self):
        difficulties_size = ale_lib.getAvailableDifficultiesSize(self.obj)
        difficulties = np.zeros((difficulties_size), dtype=np.intc)
        ale_lib.getAvailableDifficulties(self.obj, as_ctypes(difficulties))
        return difficulties

    def setDifficulty(self, difficulty):
        ale_lib.setDifficulty(self.obj, int(difficulty))

    def getLegalActionSet(self):
        act_size = ale_lib.getLegalActionSize(self.obj)
        act = np.zeros((act_size), dtype=np.intc)
        ale_lib.getLegalActionSet(self.obj, as_ctypes(act))
        return act

    def getMinimalActionSet(self):
        act_size = ale_lib.getMinimalActionSize(self.obj)
        act = np.zeros((act_size), dtype=np.intc)
        ale_lib.getMinimalActionSet(self.obj, as_ctypes(act))
        return act

    def getFrameNumber(self):
        return ale_lib.getFrameNumber(self.obj)

    def lives(self):
        return ale_lib.lives(self.obj)

    def getEpisodeFrameNumber(self):
        return ale_lib.getEpisodeFrameNumber(self.obj)

    def getScreenDims(self):
        """returns a tuple that contains (screen_width, screen_height)
        """
        width = ale_lib.getScreenWidth(self.obj)
        height = ale_lib.getScreenHeight(self.obj)
        return (width, height)

    def getScreen(self, screen_data=None):
        """This function fills screen_data with the RAW Pixel data
        screen_data MUST be a numpy array of uint8/int8. This could be initialized like so:
        screen_data = np.empty(w*h, dtype=np.uint8)
        Notice,  it must be width*height in size also
        If it is None,  then this function will initialize it
        Note: This is the raw pixel values from the atari,  before any RGB palette transformation takes place
        """
        if screen_data is None:
            width = ale_lib.getScreenWidth(self.obj)
            height = ale_lib.getScreenHeight(self.obj)
            screen_data = np.zeros(width * height, dtype=np.uint8)
        ale_lib.getScreen(self.obj, as_ctypes(screen_data))
        return screen_data

    def getScreenRGB(self, screen_data=None):
        """This function fills screen_data with the data in RGB format
        screen_data MUST be a numpy array of uint8. This can be initialized like so:
        screen_data = np.empty((height,width,3), dtype=np.uint8)
        If it is None,  then this function will initialize it.
        """
        if screen_data is None:
            width = ale_lib.getScreenWidth(self.obj)
            height = ale_lib.getScreenHeight(self.obj)
            screen_data = np.empty((height, width, 3), dtype=np.uint8)
        ale_lib.getScreenRGB(self.obj, as_ctypes(screen_data[:]))
        return screen_data

    def getScreenGrayscale(self, screen_data=None):
        """This function fills screen_data with the data in grayscale
        screen_data MUST be a numpy array of uint8. This can be initialized like so:
        screen_data = np.empty((height,width,1), dtype=np.uint8)
        If it is None,  then this function will initialize it.
        """
        if screen_data is None:
            width = ale_lib.getScreenWidth(self.obj)
            height = ale_lib.getScreenHeight(self.obj)
            screen_data = np.empty((height, width, 1), dtype=np.uint8)
        ale_lib.getScreenGrayscale(self.obj, as_ctypes(screen_data[:]))
        return screen_data

    def getRAMSize(self):
        return ale_lib.getRAMSize(self.obj)

    def getRAM(self, ram=None):
        """This function grabs the atari RAM.
        ram MUST be a numpy array of uint8/int8. This can be initialized like so:
        ram = np.array(ram_size, dtype=uint8)
        Notice: It must be ram_size where ram_size can be retrieved via the getRAMSize function.
        If it is None,  then this function will initialize it.
        """
        if ram is None:
            ram_size = ale_lib.getRAMSize(self.obj)
            ram = np.zeros(ram_size, dtype=np.uint8)
        ale_lib.getRAM(self.obj, as_ctypes(ram))
        return ram

    def saveScreenPNG(self, filename):
        """Save the current screen as a png file"""
        return ale_lib.saveScreenPNG(self.obj, _str_as_bytes(filename))

    def saveState(self):
        """Saves the state of the system"""
        return ale_lib.saveState(self.obj)

    def loadState(self):
        """Loads the state of the system"""
        return ale_lib.loadState(self.obj)

    def cloneState(self):
        """This makes a copy of the environment state. This copy does *not*
        include pseudorandomness, making it suitable for planning
        purposes. By contrast, see cloneSystemState.
        """
        return ale_lib.cloneState(self.obj)

    def restoreState(self, state):
        """Reverse operation of cloneState(). This does not restore
        pseudorandomness, so that repeated calls to restoreState() in
        the stochastic controls setting will not lead to the same
        outcomes.  By contrast, see restoreSystemState.
        """
        ale_lib.restoreState(self.obj, state)

    def cloneSystemState(self):
        """This makes a copy of the system & environment state, suitable for
        serialization. This includes pseudorandomness and so is *not*
        suitable for planning purposes.
        """
        return ale_lib.cloneSystemState(self.obj)

    def restoreSystemState(self, state):
        """Reverse operation of cloneSystemState."""
        ale_lib.restoreSystemState(self.obj, state)

    def deleteState(self, state):
        """ Deallocates the ALEState """
        ale_lib.deleteState(state)

    def encodeStateLen(self, state):
        return ale_lib.encodeStateLen(state)

    def encodeState(self, state, buf=None):
        if buf == None:
            length = ale_lib.encodeStateLen(state)
            buf = np.zeros(length, dtype=np.uint8)
        ale_lib.encodeState(state, as_ctypes(buf), c_int(len(buf)))
        return buf

    def decodeState(self, serialized):
        return ale_lib.decodeState(as_ctypes(serialized), len(serialized))

    def __del__(self):
        ale_lib.ALE_del(self.obj)

    @staticmethod
    def setLoggerMode(mode):
        dic = {"info": 0, "warning": 1, "error": 2}
        mode = dic.get(mode, mode)
        assert mode in [
            0,
            1,
            2,
        ], "Invalid Mode! Mode must be one of 0: info, 1: warning, 2: error"
        ale_lib.setLoggerMode(mode)