Differences

This shows you the differences between two versions of the page.

--- brain:arcade_games [2017/01/05 00:19] – peter
+++ brain:arcade_games [2020/07/15 09:30] (current) – external edit 127.0.0.1
@@ Line 19: / Line 19: @@
 ale> make -j 4
 </code>
-===== Ideas =====
-Brain learns from images, i.e. image snapshots, or differences between one image and another.
-Images therefore need to be readable by the Brain.  Use jpg, png, bmp, own format libraries to do this.
-Standardise all images to a standard size to standardise learning.  Do we therefore assume X*Y is pre-set for any image.  The larger that X and Y is will result in more pixels, so more training / longer training needed to process, but more accurate.
-Perhaps different type neurons for different size images with some sort of standardisation comparison between unequal image sizes.
@@ Line 67: / Line 55: @@
 This should display some details about the game.
+READ the manual at /ale/doc/manual/manual.pdf
 <code bash>
@@ Line 102: / Line 91: @@
 ./ale -display_screen true -game_controller fifo roms/breakout.bin
 </code>
+===== Install ImageMagick and supporting libs =====
+apt-get install imagemagick
+apt-get install libmagick++-dev
+===== Ideas =====
+Brain learns from images, i.e. image snapshots, or differences between one image and another.
+Images therefore need to be readable by the Brain.  Use jpg, png, bmp, own format libraries to do this.
+Standardise all images to a standard size to standardise learning.  Do we therefore assume X*Y is pre-set for any image.  The larger that X and Y is will result in more pixels, so more training / longer training needed to process, but more accurate.
+Perhaps different type neurons for different size images with some sort of standardisation comparison between unequal image sizes.
+Images - Convolutions
+update resolver's epsilon (chance of random action instead of optimal one)
+# epsilon decreases over time
+current_epsilon = 0.05 + 0.95 * np.exp(-epoch / 2500.)
+resolver.epsilon.set_value(np.float32(current_epsilon))
+# gamma - delayed reward coefficient - what fraction of reward is retained if it is obtained one tick later
+gamma = theano.shared(np.float32(1), name='q_learning_gamma')
@@ Line 109: / Line 126: @@
 https://www.nervanasys.com/deep-reinforcement-learning-with-neon/
+http://www.imagemagick.org/Magick++/tutorial/Magick++_tutorial.pdf