git submodule update --init --recursive
sudo apt install build-essential libgtk-3-dev xclip python3-pip # or brew for mac
pip3 install --user -r requirements.txt # or in a venv

make # build in ./bin

Find words in a grid. Reads from stdin if GRID_FILE not specified. Prints the
grid highlighted with entries followed by a list of results.
Usage:
  wordsearch [OPTION...] [GRID_FILE]

  -h, --help               help
  -d, --dict DICT_FILE     dictionary file (default: /usr/share/dict/words)
  -l, --min_length LENGTH  min length of words to consider from dictionary
                           (default: 5)

 search options:
      --max_turns N_TURNS  max turns to take, -1 for infinite (default: 0)
      --inserted N_SKIPS   number of letters inserted into words placed in
                           the grid (default: 0)
      --removed N_SKIPS    number of letters removed from words placed in the
                           grid (default: 0)
      --allow_less         allow less than the max letters inserted/removed
      --card_only          search in cardinal directions only

 display options:
      --no_list         do not list results
      --no_grid         do not show the grid
      --show_artifacts  show inserted and removed letters in list

usage: xword [-h] [--image IMAGE] [--clues CLUES] [--entries ENTRIES] [--output OUTPUT]
             [--force-number] [--clip] [--fft]

Server for automated crossword solving.

This tracks changes on the clipboard to update source data. After an image is copied, the server
will generate the board as html, which can be pasted into Google Sheets. After an image and a set of
clues are copied, the server will generate its best guess at filling out the board along with board
and clue information.

The server will keep running in the background, taking whatever is put on the clipboard, solving
crosswords, and putting the results back on the clipboard. In particular, it does not matter whether
the image of the board or the text of the clues is copied first. Once finished, it is recommended to
stop the server so that it will not change new clipboard contents.

On repeated uses, all crossword answer queries are cached. Additionally, the last image, clues,
generated entry answers, and output are all stored to last-image.png, etc, under the queso project
root. The last inputs are reused by default. To ignore a file, use the options to set the input file
to /dev/null.

implementation details:
This program has two parts, board extraction and crossword solving. A board is extracted from an
image by performing autocorrelation with itself and then cross-correlation with a constructed grid
of the detected size. Crossword solving is done by scraping a series of online crossword clue
databases for potential answers (via proxies so that there are no rate limits). The answer scores
are aggreated and used as priors for Bayesian inference (using Markov Random Fields). Belief
propagation is performed using the sum-product algorithm where the values of cells in the crossword
are the variables and the probability distributions over the answer candidates are the factors.
Unknown answers are accounted for by a special answer candidate that uses a smoothed trigram model
over past NYT answers.

optional arguments:
  -h, --help            show this help message and exit
  --image IMAGE, -i IMAGE
                        Image file to read and extract a board from.
  --clues CLUES, -c CLUES
                        Text file to read clues from.
  --entries ENTRIES, -e ENTRIES
                        Text file to read entry scores from.
  --output OUTPUT, -o OUTPUT
                        Output file to write html (same data that is copied to the clipboard).
  --force-number        Insert numbers by board structure rather than detecting text in a corner.
  --clip                Use the clipboard contents on startup instead of loading from files.
                        (Changes in clipboard contents are always used.)
  --fft                 Use an fft based board detection instead of line detection. Requires that
                        square sizes are exactly equal.

Attempt to decode a QR code from text input. Input should be lines of symbols, with [1#+*@Xx] for black, [ .0Oo_-] for white, and [=?] for unknown.
Usage:
  qr [OPTION...] [PATH]

  -h, --help                help
  -a, --all_format_options  try all 32 possible format
  -d, --debug               show debugging steps