[SYMBOL][A][B][C][D][E][F][G][H][I][J][K][L][M][N][O][P][Q][R][S][T][U][V][W][X][Z]
accuracy
action-value functions, 2nd
activating bots at OGS
activation argument, 2nd
activation layers
actor-critic algorithm
actor-critic methods
designing neural networks for actor-critic learning
playing games with agents
reinforcement learning with
training agents from experience data
Adadelta optimizer, 2nd
Adagrad optimizer
adaptive gradients
refining with Adadelta
training with
decay and momentum in SGD
optimizing neural networks with Adagrad
add_liberty function
advantage
calculating during self-play
overview of
to determine important decisions
affine linear
AGA (American Go Association)
Agent class
agents
loading from disk
saving from disk
AGZ (AlphaGo Zero)
additional resources for
building neural network for tree search
guiding tree search with neural networks
expanding trees
selecting moves
walking down trees
improving exploration with Dirichlet noise
modern techniques for deeper neural networks
batch normalization
residual networks
training
AI (artificial intelligence)
algorithms
alpha-beta pruning, 2nd
AlphaGo program
board encoder
bootstrapping self-play from policy networks
deriving value networks from self-play data
network architectures in
policy networks for improved searches
implementing AlphaGo search algorithm
neural networks to improve Monte Carlo rollouts
tree search with combined value function
practical considerations for training
training deep neural networks for
training policy networks
value networks for improved searches
implementing AlphaGo search algorithm
neural networks to improve Monte Carlo rollouts
tree search with combined value function
AlphaGo Zero.
See AGZ.
alphago_model function
alphago_sl_agent
AlphaGoEncoder
AlphaGoMCTS class, 2nd
AlphaGoNode
Amazon Machine Image (AMI)
Amazon Web Services (AWS).
See AWS.
American Go Association (AGA)
AMI (Amazon Machine Image)
ANNs (artificial neural networks), 2nd
apply_move function
architectures, testing
area counting
area scoring
argsort function
array type
artificial intelligence (AI)
artificial neural networks (ANNs), 2nd
average pooling
average_digit function
AveragePooling2D
avg_eight function
AWS (Amazon Web Service)
deploying bots with
hosting bots on
model training on
overview of, 2nd, 3rd
training bots with
axis argument
backends
backpropagation
computational challenges with
for feed-forward networks
for sequential neural networks
notation in
overview of
backward passes, 2nd
batch normalization
batch_size parameter
begin_episode method
benchmarking Go AI (artificial intelligence)
BensonDarr
best_black value
best_result function
best_white value
BetagoBot account
bias term, 2nd
binary classification
binary features
binomial test
black_player function
board
AlphaGo encoders
implementing in Python
overview of
Board class, 2nd, 3rd
board-game AI, overall structure of
evaluating game states
reducing number of moves to consider
searching game states
selecting opening moves
boardsize command
bootstrapping self-play from policy networks
bot account
bot_v_bot function
bots
activating at OGS
competing against other bots locally
deploying
GTP (Go Text Protocol)
in cloud
serving to web frontend
to online servers
with AWS
hosting on AWS over HTTP
losing gracefully
playing against
playing against other Go programs
registering at OGS
overview of
registering bots at OGS
submitting to OGS
testing locally
training in cloud
training with AWS
when bots should pass or resign.
See also deep-learning bots; ; Go bot.
calculus
derivatives
finding maxima
callbacks
can_add_child property
categorical cross-entropy loss function
chain rule, 2nd
chained layers
channel
channels first convention
channels_first format
channels_last format
child node
children property
Chinese counting
chunking
class imbalance
classifying
cross-entropy loss for problems with
games
handwritten digits
MNIST data preprocessing
MNIST data set of handwritten digits
clear_deltas method
CLI (command-line interface)
clipnorm parameter
clipping
overview of
probability distributions
cloud
deploying bots in
training bots in
clustering
combined value functions
command_and_response method
command-line interface (CLI)
competing against other bots locally
bots playing against other Go programs
when bots should pass or resign
complete_episode method, 2nd
completing captures
computing the gradient
Concatenate layer
concentration parameter
consolidate_games function
Conv2D layer, 2nd
convolutional kernel
convolutional layer
convolutional networks
analyzing space with
building with Keras
overview of convolutions
convolutions
counting
covariate shift
credit assignment problem, 2nd
cross-entropy loss
curvature of function
cutoff
data
building data generators
building data processors
for training networks
human game-play data
one-dimensional data
parallel data processing and generators
preparing for deep learning
parallel data processing and generators
replaying Go game from an SGF record
training and test data
two-dimensional data.
See also experience data; Go data, designing neural networks for.
data generator
dead ReLUs problem
decay in SGD
decision threshold
decode_move_index function
decode_point_index function
decoding operation
deep learning
overview of
preparing data for
building data generators
building data processors
parallel data processing and generators
replaying Go game from an SGF record
deep neural networks
creating move-prediction agents from
modern techniques for
batch normalization
residual networks
deep-learning bots
building data encoders
importing Go game records
downloading and replaying Go game records from KGS
SGF file format
preparing data for deep learning
building data generators
building data processors
parallel data processing and generators
replaying Go game from an SGF record, 6th
building data generators
building data processors
parallel data processing and generators
replaying Go game from an SGF record
running experiments
guidelines for testing architectures
guidelines for testing hyperparameters
training deep-learning models on human game-play data
training with adaptive gradients
decay and momentum in SGD
optimizing neural networks with Adagrad
refining adaptive gradients with Adadelta
deep-learning library, in Keras
design principles
installing
move prediction with feed-forward neural networks
running first example
DeepLearningAgent, 2nd, 3rd, 4th
dense layers, 2nd, 3rd
deploying
bots
competing against other bots locally
creating move-prediction agents from deep neural network
GTP (Go Text Protocol)
in cloud
serving to web frontend
to online servers
with AWS
OGS on AWS
depth pruning, 2nd
derivatives
descriptive model
deterministic games
diagonal matrix
differentiable function
Dirichlet noise
discounting technique
disks
loading agents from
saving agents from
dlgo module, 2nd, 3rd, 4th
dlgo.data.parallel_processor
dlgo.encoders module
dlgo.networks module
does_move_violate_ko
dot product, 2nd
download_files method
downloading Go game records from KGS
dropout technique, dropping neurons for regularization
ϵ-greedy policy, 2nd
EC2 (Elastic Compute Cloud)
Elastic Compute Cloud (EC2)
empty points
encode_move function
encode_point function
Encoder class, 2nd
encoders
AlphaGo board encoder
building data encoders
encoding Go game positions
episodes
epochs
error terms, propagating
evaluate_generator function
evaluation functions
evaluations
position evaluation
rollout policies for
expanding trees
experience data
overview of
representing
training actor-critic agents from
experience_simulation function
ExperienceBuffer class
ExperienceCollector class, 2nd, 3rd
expert systems
exploitation, 2nd
exploration, 2nd
fast policy network
feature maps
feature planes
feature vectors
features
feed-forward networks
backpropagation for
dense layers in Python as building blocks for
predicting moves in Keras
filter
fit function, 2nd, 3rd
fit_generator function
fitting models
Flatten layer
forgetting
forward passes, 2nd, 3rd
FreeGoban
frontend.
See web frontend, serving bots to.
frozenset
Fuego
fully connected layer
fuzzy logic
game states
capturing
evaluating
overview of
with MCTS
searching
game tree
game_state property
games
as machine-learning problems
classifying
playing with actor-critic agents
playing with Q-learning
simulating
GameState class, 2nd, 3rd, 4th
generate method
generators, data generators
GenericLinearFunction
genmove command
get_backward_input function
get_encoder_by_name function
get_forward_input function
get_go_string function
get_web_app method
global minimum
GNU Go program, 2nd
Go AI
benchmarking
measuring strength of
Go bot
capturing game states
checking for illegal moves
ko
self-capture
creating first bot
ending games
playing against bots
speeding up game play with Zobrist hashing
Go game
additional resources for
as machine-learning problem
measuring Go AI strength
overall structure of board-game AI
downloading records from KGS
encoding positions for neural networks
handicap stones
importing records
overview of
capturing stones
counting
ending game
placing stones
understanding board
understanding ko
replaying from SGF records
replaying records from KGS
representing in Python
capturing stones on board
implementing board
placing stones on board
tracking connected groups of stones
traditional Go ranks
Go programs
GNU Go
Pachi
Go servers
IGS (Internet Go Server)
OGS (Online Go Server)
Tygem
Go Text Protocol (GTP), 2nd, 3rd
GoDataProcessor, 2nd, 3rd
GOFAI (good old-fashioned AI)
gosgf submodule
GoString class
gradient ascent algorithm
gradient descent algorithm
modifying neural network policies with
overview of
to find minima
gradients, propagate back through network.
See also policy gradients.
graphical interfaces
greedy policy, 2nd
groups
GTP (Go Text Protocol), 2nd, 3rd
gtp2ogs library
GTPFrontend, 2nd, 3rd
Hadamard product, 2nd
handicap stones
handwritten digits, classifying, 2nd
MNIST data preprocessing
MNIST data set of handwritten digits
hash values
heavy rollouts
hidden information games
hidden layers
hidden units
HTTP (Hypertext Transfer Protocol), 2nd
httpfrontend module
human game-play data
hyperbolic tangent function
hyperparameters
overview of, 2nd
testing
Hypertext Transfer Protocol (HTTP), 2nd
IGS (Internet Go Server)
illegal moves, checking for
ko
self-capture
importing Go game records
init_ac_agent.py script
inner product
input image
input neurons
installing Keras deep-learning library
Internet Go Server
is_over function
is_point_an_eye function
is_terminal property
JagoClient
Japanese rules
jgoboard library
Keras API
building convolutional neural networks with
building two-input networks in
deep-learning library in
installing
move prediction with feed-forward neural networks
running first example with Keras
design principles
functional API
implementing E-greedy policy with
with Q-learning
Keras library
keras.datasets module
keras.layers module
kernel
KGS Go Server
downloading Go game records from
overview of
replaying Go game records from
KGSIndex class
ko
checking for
overview of
komi
labels
ladders
Layer class
layers
activation layers in neural networks
dense layers in Python
neural network layers in Python
overview of, 2nd, 3rd
pooling layers
softmax activation function in
LeakyReLU function
learning rate
learning.
See RL (reinforcement learning); ; supervised learning; ; unsupervised learning.
Leela Zero
legal_moves function
liberties
light rollouts
load_experience function
load_go_data method
load_policy_agent function
loading agents from disk
local maximum
local minimum, 2nd
LocalGtpBot
logic production systems
logistic regression, 2nd
long-running process
loss functions
finding minima in
gradient descent to find minima
MSE for
overview of, 2nd, 3rd
propagate gradients back through network
SGD for
loss weights
loss_derivative function
machine learning
AI and
applications of
deep learning
games as
Go game as
overall structure of board-game AI
in software applications
limits of
overview of
reinforcement learning
supervised learning
unsupervised learning
matrices
matrix transposition
max pooling
max_depth parameter
maxima
MaxPooling2D
MCTS (Monte Carlo tree search)
applying to Go
bot losing gracefully
improved rollout policies for better evaluations
evaluating game states with
selecting branches to explore
implementing in Python
overview of
mcts module
MCTSNode class
mean squared error (MSE)
measuring
Go AI strength
benchmarking Go AI
traditional Go ranks
small differences in strength
merged_with method, 2nd
meta-information
mini-batch
Minigo
minima
finding in loss functions
gradient descent to find
minimax search algorithm
anticipating opponent with
example of
overview of
MLP (multilayer perceptron)
MNIST (Modified National Institute of Standards and Technology)
data preprocessing
data set of handwritten digits
overview of
ModelCheckpoint tool
models
calibrating
deep-learning
training on AWS
Modified National Institute of Standards and Technology.
See MNIST.
momentum in SGD
Monte Carlo rollouts
Monte Carlo tree search (MCTS).
See MCTS.
Move class
move property
move-prediction agents
move-prediction networks, 2nd
moves
implementing move selection
predicting probabilities of
cross-entropy loss for classification problems
softmax activation function in last layer
predicting with feed-forward neural networks in Keras
reducing number of moves to consider
selecting
opening moves
overview of
MSE (mean squared error)
multilayer neural network
multilayer perceptron (MLP)
neural networks
activation layers in
backpropagation for
building for tree search
classifying handwritten digits
MNIST data preprocessing
MNIST data set of handwritten digits
designing for actor-critic learning
encoding Go game positions for
feed-forward networks
guiding tree search with
expanding trees
selecting moves
walking down trees
loss functions
finding minima in
gradient descent to find minima
MSE
overview of
propagate gradients back through network
SGD for
modifying policies with gradient descent
optimizing with Adagrad
overview of, 2nd
logistic regression as simple artificial neural network
networks with more than one output dimension
to improve Monte Carlo rollouts
training in Python
applying network handwritten digit classification
dense layers in Python as building blocks for feed-forward networks
neural network layers in Python
sequential neural networks with Python.
See deep neural networks.
neurons, 2nd
next() function
node probabilities
nodes
nondeterministic games
nonsequential neural networks
normalizing
notation in backpropagation
np.array function
np.clip function
np.random.choice function
np.random.dirichlet function
num_cols function
num_liberties function, 2nd
num_rollouts property
num_rows function
num_simulations
num_total_examples function
numerical input
numpy arrays
NumPy library
NVIDIA
objective functions
OCR (optical character recognition)
OGS (Online Go Server)
activating bots at
deploying on AWS
overview of, 2nd, 3rd
registering bots at, 2nd
submitting bots to
testing bots locally
one-dimensional data
oneplane encoder
OnePlaneEncoder, 2nd, 3rd
online servers, deploying bots to
OpenAI Gym
opening moves, selecting
optical character recognition (OCR)
optimization, starting point for
optimizers
overview of
tuning
optimizing neural networks with Adagrad
other method
outlier detection
output dimensions
output neurons
output probabilities
overfitting, 2nd
Pachi program, 2nd, 3rd
parallel data processing
parallelized search
parent property
parse function
passing
patches
pattern recognition
perfect information games
performance, evaluating
PhoenixGo
picture labels
place_stone method
planes
play command
play_our_move method
play_their_move method
Player class
Point class
Polanyi’s paradox
policy function
policy gradients
overview of
reinforcement learning with
identifying good decisions with random games
modifying neural network policies with gradient descent
training with self-play
policy learning
policy networks
AlphaGo-style, training
bootstrapping self-play from
for improved searches
implementing AlphaGo search algorithm
neural networks to improve Monte Carlo rollouts
tree search with combined value function
overview of
policy_probabilities
policy_rollout
PolicyAgent class, 2nd, 3rd
pool size
pooling layers
position evaluation
predict labels
predicting
move probabilities
cross-entropy loss for classification problems
softmax activation function in last layer
moves with feed-forward neural networks in Keras.
See also move-prediction agents.
predictions
calculating
computing
cutoff value in
evaluating
prepare_experience_data function
previous_states variable
print_board function
print_move function
prior probabilities
probabilities
clipping distributions
of moves, predicting
cross-entropy loss for classification problems
softmax activation function in last layer
sampling from distributions
process method
process_zip function
propagating error terms
propagating gradients
pruning, reducing search space with
reducing search depth with position evaluation
reducing search width with alpha-beta pruning
Python programming language
and machine learning
implementing MCTS in
neural network layers in
representing Go game in
capturing stones on board
implementing board
placing stones on board
tracking connected groups of stones
sequential neural networks with
training neural networks in
Q-learning technique
playing games with
with Keras
building two-input networks
implementing E-greedy policy
training action-value functions
QAgent class, 2nd
random games
RandomAgent
RandomBot
rank 3 tensors
rank 4 tensors
record_decision method, 2nd
rectified linear units (ReLU) activation function, 2nd
refining adaptive gradients with Adadelta
registering bots at OGS, 2nd
regression problems
regularization
REINFORCE method
reinforcement-learning techniques
ReLU (rectified linear units) activation function, 2nd
remove_liberty function
replaying
Go game from an SGF record
Go game records from KGS
residual networks
resigning
ResignLargeMargin strategy
rewards, calculating
RL (reinforcement learning)
building agents that can learn
clipping probability distributions
implementing move selection
loading agents from disk
sampling from probability distributions
saving agents from disk
cycle of
experience data
integrating tree search with
building neural network for tree search
guiding tree search with neural networks
improving exploration with Dirichlet noise
modern techniques for deeper neural networks
training
self-play
representing experience data
simulating games
with actor-critic methods
advantage to determine important decisions
designing neural networks for actor-critic learning
playing games with actor-critic agents
training actor-critic agents from experience data
with policy gradients
identifying good decisions with random games
modifying neural network policies with gradient descent
training with self-play
with value methods
playing games with Q-learning
Q-learning with Keras
rollout policies, 2nd
sampling from probability distributions
saving agents from disk
scalar
search algorithm
search depth
search space, reducing with pruning
reducing search depth with position evaluation
reducing search width with alpha-beta pruning
search width
searching game states
select_move function, 2nd, 3rd, 4th, 5th, 6th
self-capture, 2nd
self-play
bootstrapping from policy networks
calculating advantage during
deriving value networks from data
overview of
representing experience data
simulating games
training with
evaluating progress
measuring small differences in strength
tuning SGD optimizers
self.previous_states
Sensei’s Library, 2nd
sequential neural networks
backpropagation for
overview of
with Python
SequentialNetwork class, 2nd
serialize method
servers.
See Go servers; ; online servers, deploying bots to.
serving bots to web frontend
set_collector method, 2nd
set_handicap method
set_temperature method
SevenPlaneEncoder
SGD (stochastic gradient descent)
decay and momentum in
for loss functions
tuning optimizers
SGF (Smart Game Format) file format, 2nd
Sgf_game class
SGFWriter
shape attribute
shoulder hit
sigmoid function, 2nd, 3rd
sigmoid_double function
simple heuristics
simple models
simulate_game function
simulate_random_game function
simulating games
situational superko rule
skip connection
Smart Game Format (SGF) file format, 2nd
Smart Go Format
snapback
Sobel kernel
softmax activation function, 2nd
splitting data
ssh command
standardized protocols
star points
steps_per_epoch function
stochastic gradient descent (SGD).
See SGD.
stochastic policy, 2nd
stones
capturing
on board in Python
overview of
placing
on board in Python
overview of
tracking connected groups of in Python
strings
strong policy network
suggest_tags function
supervised learning
tanh activation
temperature, 2nd
Tencent
TensorFlow library
tensorflow-gpu library
tensors
overview of
rank 3 tensors
rank 4 tensors
termination strategies
TerminationAgent class
territory scoring
test data, splitting
testing
architectures
hyperparameters
OGS bots locally
text-based protocols
Theano library
to_buffer method
train function, 2nd
train_batch method
train_data method
trainable parameters
training
action-value functions
actor-critic agents from experience data
AlphaGo
AlphaGo-style policy networks
and test data
bots
in cloud
with AWS
deep neural networks for AlphaGo
AlphaGo board encoder
network architectures in AlphaGo
training AlphaGo-style policy networks
deep-learning models on human game-play data
models on AWS
networks
with adaptive gradients
decay and momentum in SGD
optimizing neural networks with Adagrad
refining adaptive gradients with Adadelta
with self-play
evaluating progress
measuring small differences in strength
tuning SGD optimizers
training models
tree search
building neural networks for
classifying games
evaluating game states with MCTS
applying MCTS to Go
implementing MCTS in Python
selecting branches to explore
generating games as network training data
guiding with neural networks
expanding trees
selecting moves
walking down trees
integrating with reinforcement learning
improving exploration with Dirichlet noise
training
minimax algorithm
minimax search algorithm
reducing search space with pruning
reducing search depth with position evaluation
reducing search width with alpha-beta pruning
with combined value function
trees
expanding
walking down
tuning SGD optimizers
two-dimensional data
two-input networks
Tygem Go server, 2nd
UCT (upper confidence bound for trees) formula, 2nd
uniform random policy
unsupervised learning
unvisited_moves property
update rules, for parameters
upper confidence bound for trees (UCT) formula, 2nd
use_generator flag
validation data
validation set
value methods
overview of
reinforcement learning with
playing games with Q-learning
Q-learning with Keras
value networks
deriving from self-play data
for improved searches
implementing AlphaGo search algorithm
neural networks to improve Monte Carlo rollouts
tree search with combined value function
overview of
value-estimating function
ValueAgent
vectors
one-dimensional data
overview of
vertical edges
visible units
web frontend, serving bots to
weight initializers
weights, 2nd
white_player function
win_counts property
winner function
winning_frac property
ZeroAgent
ZeroPadding2D layer, 2nd
ZeroTreeNode class
Zobrist hashing