Avoid importing GUI by default to avoid issues on headless systems.

README.md

@@ -79,7 +79,7 @@ This is the simplest way to visualize the environments, unless you need to handl
 ### Using GUI class
 We also include a slightly more complex GUI to visualize the environments and optionally handle user input. This GUI is used in examples/human_play.py to play as a human and examples/agent_play.py to visualize the performance of trained agents. To use the GUI you can import it in your code with:
 ```python
-from minatar import GUI
+from minatar.gui import GUI
 ```
 Initialize an instance of the GUI class by providing a name for the window, and the integer number of input channels for the minatar environment to be visualized. For example:
 ```python

examples/agent_play.py

@@ -10,7 +10,8 @@
 import torch
 import argparse, numpy
 
-from minatar import Environment, GUI
+from minatar import Environment
+from minatar.gui import GUI
 from dqn import QNetwork, get_state
 from AC_lambda import ACNetwork
 

examples/human_play.py

@@ -8,7 +8,8 @@
 
 import argparse
 import tkinter as Tk
-from minatar import Environment, GUI
+from minatar import Environment
+from minatar.gui import GUI
 
 ################################################################################################################
 # Script that allows a human to play any of the MinAtar games. Use arrow keys to move and space to fire. 

minatar/__init__.py

@@ -1,2 +1 @@
-from .environment import Environment
-from .gui import GUI
+from .environment import Environment

setup.py

@@ -27,7 +27,7 @@
 
 setup(
     name='MinAtar',
-    version='1.0.8',
+    version='1.0.9',
     description='A miniaturized version of the arcade learning environment.',
     url='https://github.com/kenjyoung/MinAtar',
     author='Kenny Young',