Looking for chess cheating in centipawn dataset

chesscheat.jl

using CSV
using Plots


function load_all_games()
  CSV.read("/Users/tallakt/Downloads/15982_games_with_centipawn_metrics/15982_games_with_centipawn_metrics.csv", DataFrame)
end


function player_games(all_games, player_name)
  black = filter(r -> !isnothing(match(Regex(player_name), r["Black Name"])), all_games)
  white = filter(r -> !isnothing(match(Regex(player_name), r["White Name"])), all_games)
  black.player_name = black[:, "Black Name"]
  white.player_name = white[:, "White Name"]
  black.opponent_name = black[:, "White Name"]
  white.opponent_name = white[:, "Black Name"]
  black.player_elo = black[:, "Black ELO"]
  white.player_elo = white[:, "White ELO"]
  black.opponent_elo = black[:, "White ELO"]
  white.opponent_elo = white[:, "Black ELO"]
  black.player_avg_cp_loss = black[:, "Black Av CP Loss"]
  white.player_avg_cp_loss = white[:, "White Av CP Loss"]
  black.opponent_avg_cp_loss = black[:, "White Av CP Loss"]
  white.opponent_avg_cp_loss = white[:, "Black Av CP Loss"]
  black.player_cp_loss = JSON.parse.(String.(black[:, "Black CP Loss List"]))
  white.player_cp_loss = JSON.parse.(String.(white[:, "White CP Loss List"]))
  black.opponent_cp_loss = JSON.parse.(String.(black[:, "White CP Loss List"]))
  white.opponent_cp_loss = JSON.parse.(String.(white[:, "Black CP Loss List"]))
  black.evaluations_list = JSON.parse.(String.(black[:, "Evaluations List"]))
  white.evaluations_list = JSON.parse.(String.(white[:, "Evaluations List"]))
  black.player_evaluations = [-x for x = [e[2:2:end] for e=JSON.parse.(String.(black[:, "Evaluations List"]))]]
  white.player_evaluations = [e[1:2:end] for e=JSON.parse.(String.(white[:, "Evaluations List"]))]
  black.opponent_evaluations = [e[1:2:end] for e=JSON.parse.(String.(black[:, "Evaluations List"]))]
  white.opponent_evaluations = [1000 .- x for x = [e[2:2:end] for e=JSON.parse.(String.(white[:, "Evaluations List"]))]]
  black.player_signed_evaluations = [-el for el=black.evaluations_list]
  white.player_signed_evaluations = white.evaluations_list
  black.player_color .= :black
  white.player_color .= :white
  vcat(black, white)
end


function plot_player_vs_opponent_moves_histogram(games; firstmove = 1)
  player_cp_losses = vcat([l[firstmove:end] for l = games.player_cp_loss]...)
  opponent_cp_losses = vcat([l[firstmove:end] for l = games.opponent_cp_loss]...)
  ymax = mean([count(player_cp_losses .== 1), count(opponent_cp_losses .== 1)])
  plot( 
       histogram(player_cp_losses, bins = 0:150, ylims = (0, ymax * 1.3), title = games.player_name[1])
       , histogram(opponent_cp_losses, bins = 0:150, ylims = (0, ymax * 1.3), title = "Opponents", xlabel="centipawn loss")
       , size=(1100, 600)
       , layout = (2,1)
       , legend = false
      )
end


function scatter_player_standing_vs_move(games)
  n = games.Date |> length
  all_player_cp_loss = vcat(games.player_evaluations...)
  move_no = vcat([1:length(g) for g=games.player_evaluations]...)
  println(length(move_no), length(vcat(games.player_evaluations...)))
  scatter(
          all_player_cp_loss
          , vcat(games.player_cp_loss...)
          , legend = false
          , msw = 0
          , msa = 0
          , ma = min(1.0, (0.2 * 15_000 / n))
          , ms = 2.5
          , ylims = (0, 500)
          , xlims = (-1000, 1000)
          , size=(1100, 600)
          , marker_z = move_no
          , colorbar = true
         );
end


function scatter_player_move_vs_opponent_move(games)
  n = games.Date |> length
  move_no = vcat([1:length(g) for g=games.player_evaluations]...)
  scatter(
          vcat(games.player_cp_loss...)
          , vcat(games.opponent_cp_loss...)
          , legend = false
          , msw = 0
          , msa = 0
          , ma = min(1.0, (0.1 * 15_000 / n))
          , ms = 2.5
          , xlims = (0, 750)
          , ylims = (0, 750)
          , aspectratio = :equal
          , size=(1100, 600)
          , xlabel = games.player_name[1]
          , marker_z = move_no
         );
end


function scatter_player_avg_move_vs_opponent_move(games)
  n = games.Date |> length
  scatter(
          [mean(l) for l=games.player_cp_loss]
          , [mean(l) for l=games.opponent_cp_loss]
          , legend = false
          , msw = 0
          , msa = 0
          , ma = min(1.0, (0.4 * 2_000 / n))
          , ms = 2.5
          , xlims = (0, 75)
          , ylims = (0, 75)
          , aspectratio = :equal
          , size=(1100, 600)
          , xlabel = games.player_name[1]
          , marker_z = datetime2unix.(games.Date .|> DateTime)
         );
end


function scatter_player_quantile_move_vs_opponent_move(games, p; lim = 25)
  n = games.Date |> length
  scatter(
          [quantile(l, p) for l=games.player_cp_loss]
          , [quantile(l, p) for l=games.opponent_cp_loss]
          , legend = false
          , msw = 0
          , msa = 0
          , ma = min(1.0, (0.2 * 2_000 / n))
          , ms = 3.5
          , xlims = (0, lim)
          , ylims = (0, lim)
          , aspectratio = :equal
          , size=(1100, 600)
          , xlabel = games.player_name[1]
          , marker_z = datetime2unix.(games.Date .|> DateTime)
         );
end


function filter_year(games, from, to)
  filter(g -> Date(from, 1, 1) <= g.Date <= Date(to, 12, 31), games)
end


function histogram_per_tournament(games)
  for g = groupby(games, "Event Name")
    println("Event: $(g[1,"Event Name"]), Player: $(g.player_name[1])")
    plot_player_vs_opponent_moves_histogram(g) |> display
    println("")
  end
end


function histogram_per_year(games)
  (y0, y1) = extrema([year(d) for d=games.Date])
  for y = y0:y1
    println("Year: $(y), Player: $(games.player_name[1])")
    plot_player_vs_opponent_moves_histogram(filter_year(games, y, y)) |> display
    println("")
  end
end


# "suspicious" tournamenet for HN
function filter_world_opp_49th(games)
  filter(r -> !isnothing(match(r"World op 49th", r."Event Name")), games)
end

# "suspicious" tournamenet for HN
function filter_mix_200(games)
  filter(r -> !isnothing(match(r"Novi Sad TS Mix 200th", r."Event Name")), games)
end

# "suspicious" tournamenet for HN
function filter_cccsa_20(games)
  filter(r -> !isnothing(match(r"Charlotte CCCSA Fall GM", r."Event Name")), games)
end


function filter_suspicios(games)
  vcat(filter_world_opp_49th(games), filter_mix_200(games), filter_cccsa_20(games))
end


function plot_some_players_meadian_cp_loss_per_event(allgames)
  six = [player_games(allgames, n) for n=["Carlsen", "Niemann", "Caruana", "Erigaisi", "Firouzja", "Praggnanandhaa"]]

  function plot_player(games)
    by_event = groupby(games, "Event Name")
    mean_cp_loss = [mean(vcat(ev.player_cp_loss...)) for ev = by_event]
    median_cp_loss = [median(vcat(ev.player_cp_loss...)) for ev = by_event]
    p0 = plot(mean_cp_loss, lab = "mean")
    plot!(p0, median_cp_loss, lab="median: $(games.player_name[1])")
  end


  plot(
       [plot_player(player) for player = [sort(df, "Date") for df = six]]...
       , layout = (2,3)
       , xlims = (0, 180)
       , ylims = (0, 40)
       , size = (1100, 650)
      )
end

function plot_some_players_meadian_cp_loss_per_game(allgames; xlims = (0, 2000))
  six = [player_games(allgames, n) for n=["Carlsen", "Niemann", "Caruana", "Erigaisi", "Firouzja", "Praggnanandhaa"]]

  function plot_player(games)
    mean_cp_loss = [mean(ev.player_cp_loss) for ev = eachrow(sort(games, "Date"))]
    median_cp_loss = [median(ev.player_cp_loss) for ev = eachrow(sort(games, "Date"))]
    p0 = plot(mean_cp_loss, lab = "mean")
    plot!(p0, median_cp_loss, lab="median: $(games.player_name[1])")
  end


  plot(
       [plot_player(player) for player = [sort(df, "Date") for df = six]]...
       , layout = (2,3)
       , xlims = xlims
       , ylims = (0, 90)
       , size = (1100, 650)
      )
end


function scatter_median_cp_loss_vs_std(games; firstmove = 1, ms = 8.0)
  cp_losses = [g.player_cp_loss[min(firstmove, length(g.player_cp_loss)):end] for g=eachrow(games)]
  scatter(
          [median(cpl) + randn(1)[1] for cpl=cp_losses]
          , [std(cpl) for cpl = cp_losses]
          , msw = 0, msa = 0, ma = 0.15 * 2000 / length(games.Date), ms = ms
          , size = (1100, 650)
          , xlabel = "median centipawn loss per game"
          , ylabel = "std centipawn loss per game"
          , ylims = (0, 300)
          , xlims = (0, 40)
          , label = games.player_name[1]
         )
end


function scatter_median_cp_loss_vs_std_for_some_players(games)
  players = [x[2] for x=([(length(g.Date), g[1, "White Name"]) for g=groupby(all_games, "White Name")] |> sort)[(end-8):end]]

  plot(
       [plot!(scatter_median_cp_loss_vs_std(filter_year(player_games(games, p), 2018, 2999), firstmove = 15, ms = 4), title = p, legend = false, titlefontsize = 8, ticks = false, xlabel = "", ylabel = "") for p=players]...
       , layout = (3, 3)
       , size = (1100, 650)
      );
end

nothing