Package index • tanmai

Running the app

Entry points and the Shiny modules the interface is built from.

run_app(): Run the tanmai Shiny application
mod_analyze_server(): Analyze tab server
mod_analyze_ui(): Analyze tab UI
mod_board_server(): Interactive board server
mod_board_ui(): Interactive board UI
mod_calibrate_server(): Calibrate tab server
mod_calibrate_ui(): Calibrate tab UI
mod_live_server(): Live game tracking server
mod_live_ui(): Live game tracking UI
app_ui(): Top-level Shiny UI
app_server(): Top-level Shiny server
cv_theme(): The application's Bootstrap theme
cv_brand(): Wordmark for the navbar
cv_section_title(): A section heading with an optional lead paragraph

Reading a board

Screenshot to FEN: cropping and splitting the image, matching pieces against template libraries, working out which way round the board is, and assembling a position.

recognize_position(): Recognize a board screenshot end-to-end
prepare_board(): Normalize a screenshot to an exact board-sized RGB image
autocrop_board(): Best-effort crop to the board by trimming near-uniform margins
split_board(): Split a normalized board into 64 square images
split_squares(): Prepare and split a screenshot into 64 square images in one step
classify_square(): Classify one square against a single template library
recognize_symbols(): Classify all 64 squares against a single template library
recognize_symbols_auto(): Recognize 64 squares against all available sets at once
template_matrix(): Stack several libraries' normalized templates into one matrix
build_fen(): Build a full FEN string from recognized symbols
fen_complete(): Fill in the FEN fields people leave off
grid_to_placement(): Convert a 64-symbol grid to a FEN piece-placement field
infer_castling(): Infer castling rights from piece placement
resolve_orientation(): Resolve board orientation
detect_flip(): Detect board orientation from rank labels along the left edge
detect_flip_piece_mass(): Detect orientation from where each army sits
detect_flip_start_brightness(): Detect orientation of a starting position from piece brightness
fix_template_colour_swap(): Assert (and if necessary repair) the white/black brightness invariant
resolve_calibration_flip(): Resolve the orientation of a calibration (starting-position) screenshot

Piece sets and calibration

The bundled cburnett set, the Lichess sets fetched on demand, and teaching the recognizer art it has never seen.

piece_set_manifest(): Manifest of every Lichess piece set
available_piece_sets(): Names of all piece sets available locally
fetch_piece_set(): Download one piece set's 12 SVGs into the cache
setup_piece_sets(): Fetch and build templates for every set in the manifest
build_set_templates(): Build and cache a set's template library
build_from_start_position(): Calibrate a template library from a starting-position board
load_template_library(): Load a template library from an RDS file
load_all_template_libraries(): Load every available template library
save_template_library(): Save a template library to an RDS file
render_position(): Render an 8x8 board with pieces placed from a symbol grid
piece_svg_path(): Resolve the image file path for a piece symbol within a set
read_piece_image(): Rasterize one piece image at a given square size
piece_set_path(): Locate the directory holding a set's 12 SVGs
piece_set_ext(): Find which file extension a downloaded piece set uses
piece_sets_dir(): Cache directory for downloaded piece sets
piece_theme_base(): Piece-image URL template for the interactive board
active_templates_path(): Path to the currently active template library
custom_templates_path(): Path for user-calibrated ("custom") templates
set_templates_path(): Path where a set's calibrated template library lives

Engines

Stockfish over UCI, both one-shot and as a session that keeps searching, plus the chess.js bridge used for legality and SAN.

best_move_uci(): Ask Stockfish for the best move in a position
ensure_stockfish(): Return a path to a Stockfish binary, downloading it on first use
find_local_engine(): Find a Stockfish binary already present on this machine
stockfish_bin_dir(): Directory where the downloaded Stockfish binary is cached
engine_session_start(): Start a persistent Stockfish process
engine_session_analyse(): Point a persistent engine at a position and start an infinite search
engine_session_poll(): Drain pending engine output and fold it into the session state
engine_session_eval(): Evaluate one position on a persistent engine, waiting for the result
engine_session_stop(): Stop a persistent engine session
parse_info_line(): Parse one UCI info line into a principal-variation record
parse_uci_score(): Parse a UCI score from an engine info line
format_score(): Format an engine score for display
eval_bar_pct(): Convert a score to a 0-100 eval-bar percentage (White's share)
wait_for_line(): Read engine output until a line starting with a given prefix appears
new_chess_context(): Create a chess.js V8 context
validate_fen(): Validate a FEN via chess.js
is_valid_fen(): Test whether a FEN is legal
legal_moves(): List every legal move in a position, in UCI
fen_after_move(): The position that results from playing a move
fen_move_to_san(): Convert a UCI move to SAN for a position
fen_placement(): The piece-placement field of a FEN
fen_with_turn(): Rewrite a FEN's side to move
fen_turn(): Which side is to move in a FEN
same_position(): Test whether two FENs describe the same position to play from

The human model

Maia run through lc0, and what it is used for: scoring how likely a mistake is, and inferring how strong the opponent is from what they played.

maia_dir(): Directory holding lc0 and the Maia weight files
maia_weights_path(): Path to a Maia weights file for a rating
match_maia_rating(): Snap a requested rating to the nearest available Maia network
nearest_installed_rating(): Nearest Maia network whose weights are actually on disk
find_lc0(): Locate the lc0 binary
ensure_maia(): Download lc0 and the Maia weights if they are not already present
human_model_available(): Is a human model usable right now?
maia_session_start(): Start a persistent Maia session
maia_session_stop(): Stop a Maia session
human_move_probabilities(): Probability a modelled human plays each legal move
parse_policy_lines(): Parse lc0's verbose move statistics into per-move probabilities
blunder_risk(): Assess how likely a human is to go wrong in a position
evaluate_moves(): Evaluate a specific set of moves in one search
score_to_cp(): Convert a UCI score into a single centipawn number
trappiness(): Rate moves by how likely they are to induce an opponent mistake
radar_arrow_moves(): Rank your moves by how likely they are to induce an opponent error
maia_pool_start(): Start one Maia session per rating
maia_pool_stop(): Stop a pool of Maia sessions
move_log_likelihood(): Log-probability of one played move under each rating's network
rating_posterior(): Posterior over ratings from accumulated log-likelihoods
estimate_rating(): Estimate an opponent's rating from the moves they played
rating_estimate_is_confident(): Is a rating estimate strong enough to act on?

Following a game

Turning a stream of screenshots into a game: the legality gate that decides what counts as a move, and the summaries built on top of it.

game_new(): Start a new tracked game
game_apply_move(): Play a move onto a tracked game
game_current_fen(): The position a tracked game currently stands at
game_ply(): Number of plies played in a tracked game
game_accept(): Fold an accepted observation into the game
track_observation(): Decide what a freshly recognized position means for a tracked game
find_move_paths(): Every short legal path from one position to a target placement
move_quality(): Classify how bad a move was
game_move_losses(): What each move cost the player who made it
game_turning_points(): The moves where the game turned
game_from_pgn(): Build a tracked game from PGN
pgn_summary(): A one-line description of an imported game
format_loss(): Describe what a move cost, in words a player would use
eval_graph_geometry(): Geometry for the evaluation graph
eval_graph_svg(): Draw the evaluation across a game as an inline SVG

Internals

Helpers that support the above.

app_sys(): Locate a file installed under inst/
cfg_env(): Read a configuration setting from the environment
golem_add_external_resources(): Register static resource paths and add css/js includes
image_input_ui(): Drag-drop file input plus a clipboard-paste zone
latest_image_input(): Combine upload and paste events into one image reactive
decode_data_url_image(): Decode a data-URL image string into a magick image
to_gray_matrix(): Convert a magick image to a grayscale integer matrix
normalize_vec(): Zero-mean, unit-norm flatten of a matrix
sq_background(): Estimate a square's background level from its four corners
sq_core(): Background-subtracted central region of a square
sq_energy(): Mean absolute energy of a square core
sq_idx(): Map a 0-indexed board row/column to a 1-indexed symbol-vector position
squares_mean_brightness(): Mean background-subtracted brightness of the pieces in a set of squares
label_ink(): Foreground ink in the rank-label strip of a left-edge square
tanmai tanmai-package: tanmai: Chess Board Screenshot to Best Move