Install Box2D using conda instead of pip, and add note for Windows+pip users

Author: ageron

environment.yml

@@ -3,6 +3,7 @@ channels:
   - conda-forge
   - defaults
 dependencies:
+  - box2d-py  # used only in chapter 18, exercise 8
   - ftfy=6.0  # used only in chapter 16 by the transformers library
   - graphviz  # used only in chapter 6 for dot files
   - ipython=7.28  # a powerful Python shell
@@ -43,7 +44,7 @@ dependencies:
     - tfx~=1.3.0  # platform to deploy production ML pipelines
     - transformers~=4.11.3  # Natural Language Processing lib for TF or PyTorch
     - urlextract~=1.4.0  # optionally used in chapter 3, exercise 4
-    - gym[box2d,atari,accept-rom-license]~=0.21.0  # used only in chapter 18
+    - gym[atari,accept-rom-license]~=0.21.0  # used only in chapter 18
 
 # Specific lib versions to avoid conflicts
   - attrs=20.3

requirements.txt

@@ -1,6 +1,6 @@
-# TensorFlow is much easier to install using Anaconda, especially
-# on Windows or when using a GPU. Please see the installation
-# instructions in INSTALL.md
+# WARNING: Using Anaconda instead of pip is highly recommended, especially on
+# Windows or when using a GPU. Please see the installation instructions in
+# INSTALL.md
 
 
 ##### Core scientific packages
@@ -49,6 +49,11 @@ tensorflow-addons~=0.14.0
 # https://github.com/openai/gym#installing-everything
 gym[Box2D,atari,accept-rom-license]~=0.21.0
 
+# WARNING: on Windows, installing Box2D this way requires:
+# * Swig: http://www.swig.org/download.html
+# * Microsoft C++ Build Tools: https://visualstudio.microsoft.com/visual-cpp-build-tools/
+# It's much easier to use Anaconda instead.
+
 tf-agents~=0.10.0
 
 ##### Image manipulation