TacticToe creates training examples from its own proof attempts (only loading)

removing backquotes in hhExportLib avoiding alternate parsing

remove trailing whitespaces

semi-local lambda lifting in holyhammer translation + s-expression export for tree neural networks

holyhammer translation: making lifting semi-local to avoid capturing type variables

ttt wip + filtering basis_emit in holyhammer export

exporting examples for the policy network + reordering tactic during search according to the policy network

nn term for sml expression

missing tttRecord. signature in tttUnfold code

summing the gradients for the node before backpropagating

fixing type error: flattening after translation

more handlers for type error

trying to fix a type error

training tnn from successful searches

tnn advice for the value during tactictoe search

nicer statistics before training

replacing tnnparam by tnndim

removing autoML

separting string tactic into applications

remove trailing whitespaces

exporting tptp examples

enigma for tactictoe: tptp export

removing trailing white spaces

save before battery crash

removing AI_develop

mcts for machine learning

supervised learning as a game

remove trailing white spaces

fixing type mismatch during creation of embedding variable

precomputation before starting big steps

removing profiling

removing profiling

faster ordering of term list

Diophantine synthesis experiment first version

easier examples

more work on combinator experiments

fixing mean square error messages

MCTS: including the initial status in the node

combinator experiments updated to reflect changes in mlReinforce

mlTune removed

TNN features restored

new TNN first working version

Refactoring mleTreeNeuralNetwork

remove trailing whitespaces

allowing pre-computation that affects the player before the search

removing other term_of_boards and defining an real vector embedding as a variable

trying to reproduce fast and successful synthesis experiment

testing difference in parallelization between server and local machine

removing cost

remove trailing white spaces

multiple players second experiment

more parallelization during neural network training (does not seem to help in the average case)

removing trailing whitespaces

fix threads used in training neural networks

updating files to reflect changes in smlParallel

updating mlTune

formula synthesis experiment adapted to new RL version

new reinforcement learning loop

removing global reference from smlParallel

more refactoring

refactoring in progress: removing global variables

allowing X in subterms

refractoring neural networks and probability

remove trailing whitespaces

biais for every layers working

biais for every layers

before removing external parallelism training failed experiment

batch external parallelization

levels for hanabi

null score

random search

faster target program generation

simplified programming task

only caching full forwrad pass through the network

caching neural network inference during mcts

better constraints on external parallelization

removing trailing whitespaces + fix cache errors in term generation

moving comments into code in arithmetical experiments

write/read for tree neural network

before tuning experiment

temporarily removing momentum

before temporarily removing momentum

mleCompute parameter tuning

moving, renaming, removing machine learning files

changing directory and build sequence + adding new entailment experiment

remove external parallelism from train_tnn

update on rlGameCopy

testset for rl

updating rl for ground arithmetic

printing real numbers with exact precision

global references for gencode directories

external parallelism during training: first working version

nn training with external process (wip)

different number of cores for training and exploration

export term list as a list and not as a set

rl exploration: external parallelism

removing unnecessary batch size argument

optimizing matrix addition

I/O for training examples

rl: more tests with parmap

nn: I/O

nn + mcts parallelism

removing all trailing white spaces

Fix a bunch of breakages caused by assuming term to be an eqtype

AIM paramodulation game specification

propositional and paramodulation provers

double headed treenns

fix mcts: before updating tree neural networks

standard interface for reinforcement learning + dirichlet noise for mcts

reinforcement learning mini-games + white spaces

minor

modifying some filenames

refactoring tactictoe/holyhammer (wip)