20260419_workflow_v1

Pending

Results

1) Competition: UCL - EURO - FWC. 
2) partition data: train - test.  
2) Math models : Bayesian - Monte Carlo.  
3) Proccess: Estimation - Forecast.  

Workflow

  1) Load matches data.  
    1.1) Select data to run models.  
  
  2) Build features (forma + fatiga).   
  
  3) Split temporal (train / test). 
  
  4) Modelo Bayesiano using train.    
    4.1) Run: fit_bayesian_model (train). 
    4.2) Model evaluation (train).
    4.3) Run: Evaluate model full (test).  
    4.4) Check calibration results (test). 
    4.5) Check metrics
    
  5) Prediccion futura + Monte Carlo (test)   
    5.1) Check test data. 
    5.2) Build features (test)
    5.3) Run fit_bayesian_model (test). 
    5.4) Crear partidos futuros. 
    5.5) Predecir probabilidades. 
    
  6) Simulate Monte Carlo (test) 
    6.1) Fun: simulate_season_montecarlo. 
    6.2) Create new_matches.1. 
    6.3) Run: simulate_season_montecarlo. 
    6.4) Means simulation results.  
    6.5) Simulate partial results . 

1. Matches data

1.1 Select data to run models

• competition == “UCL”,
• season == “2025_2026”

matches <- matches %>% 
  filter(competition == "UCL", 
         season == "2025_2026") %>% 
  drop_na(goals_home)

# ===============================
# VALIDAR COLUMNAS REQUERIDAS
# ===============================

  required_cols <- c(
    "date",
    "competition",
    "season",
    "home_team",
    "away_team",
    "goals_home",
    "goals_away"
  )

  missing_cols <- setdiff(required_cols, 
                          names(matches))

  if(length(missing_cols) > 0){
    stop(paste("Faltan columnas:", paste(missing_cols, collapse=", ")))
  }

  # ===============================
  # CONVERSIONES DE TIPO
  # ===============================
  matches <- matches %>%
    mutate(
      competition = factor(competition),
      season = factor(season),
      home_team = factor(home_team),
      away_team = factor(away_team),
      goals_home = as.integer(goals_home),
      goals_away = as.integer(goals_away)
    )

  # ===============================
  # VALIDACIONES IMPORTANTES
  # ===============================
  if(any(is.na(matches$goals_home)) | any(is.na(matches$goals_away))){
    stop("Existen goles NA — revisar datos.")
  }

  if(any(matches$goals_home < 0 | matches$goals_away < 0)){
    stop("Los goles no pueden ser negativos.")
  }

  # ordenar temporalmente
  matches <- matches %>% arrange(date)
  
# str(matches, list.len = ncol(matches))

glimpse(matches)

Rows: 180
Columns: 8
$ cup         <chr> "ucl2526", "ucl2526", "ucl2526…
$ date        <date> 2025-09-16, 2025-09-16, 2025-…
$ competition <fct> UCL, UCL, UCL, UCL, UCL, UCL, …
$ season      <fct> 2025_2026, 2025_2026, 2025_202…
$ home_team   <fct> PSV, Athletic Club, Tottenham,…
$ away_team   <fct> Union SG, Arsenal, Villarreal,…
$ goals_home  <int> 1, 0, 1, 2, 4, 2, 2, 0, 3, 0, …
$ goals_away  <int> 3, 2, 0, 3, 4, 1, 2, 0, 2, 2, …

2. Build features (forma + fatiga)

2.1 Run build_features

# Run build_features

df <- matches

matches <- build_features(matches)

glimpse(matches)

Rows: 180
Columns: 14
$ cup         <chr> "ucl2526", "ucl2526", "ucl2526…
$ date        <date> 2025-09-16, 2025-09-16, 2025-…
$ competition <fct> UCL, UCL, UCL, UCL, UCL, UCL, …
$ season      <fct> 2025_2026, 2025_2026, 2025_202…
$ home_team   <fct> PSV, Athletic Club, Tottenham,…
$ away_team   <fct> Union SG, Arsenal, Villarreal,…
$ goals_home  <int> 1, 0, 1, 2, 4, 2, 2, 0, 3, 0, …
$ goals_away  <int> 3, 2, 0, 3, 4, 1, 2, 0, 2, 2, …
$ rest_home   <dbl> 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, …
$ rest_away   <dbl> 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, …
$ goal_diff   <int> -2, -2, 1, -1, 0, 1, 0, 0, 1, …
$ form_home   <dbl> 0.0, 0.0, 0.0, 0.0, -0.8, -0.2…
$ form_away   <dbl> 0.0, 0.0, 0.0, 0.0, 0.8, 0.2, …
$ matchweek   <int> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10,…

3. Split temporal (train / test)

# Split temporal

  # ==========================
  # 1. SPLIT TEMPORAL
  # ==========================
  
  matches <- matches %>% arrange(date)
  test_size = 0.2
  
  split_index <- floor((1 - test_size) * nrow(matches))
  
  train <- matches[1:split_index, ]
  test  <- matches[(split_index+1):nrow(matches), ]
  
  message("Train size:", nrow(train))

Train size:144

  message("Test size:", nrow(test))

Test size:36

4. Modelo Bayesiano train.

4.1 Run: Fit bayesian model (train)

• create ucl_20260418

# Run Fit Bayesian Model

# file path for actual model

file_path <- "~/Documents/bayes_soccer/output/models/bayes_ucl_20260418.rds"

start <- Sys.time()
  
  if (file.exists(file_path)) {
    # Load the object if the file exists
    bayes_ucl_20260418 <- readRDS(file_path)
  } else {
    # Create the object
    bayes_ucl_20260418 <- fit_bayesian_model(train)
    
    # Save it for future use
    saveRDS(bayes_ucl_20260418, file = file_path)
  }

end <- Sys.time()

print(end - start)

Time difference of 0.3158648 secs

summary("ucl_20260418")

   Length     Class      Mode 
        1 character character 

4.3 Run: Evaluate model full (test)

• Using Bayesian model_ucl & test data

# Run: Evaluate model using train data

# file path for actual model

file_path <- "~/Documents/bayes_soccer/output/models/eval_bayes_ucl_20260418.rds"

start <- Sys.time()
  
  if (file.exists(file_path)) {
    # Load the object if the file exists
    eval_bayes_ucl_20260418 <- readRDS(file_path)
  } else {
    # Create the object
    eval_bayes_ucl_20260418 <-
      evaluate_model_full(bayes_ucl_20260418, # actual model ucl
                          test,
                          test_size=0.2)
    
    # Save it for future use
    saveRDS(eval_bayes_ucl_20260418, file = file_path)
  }
end <- Sys.time()

print(end - start)

Time difference of 0.4185491 secs

summary("eval_bayes_ucl_20260418")

   Length     Class      Mode 
        1 character character 

4.4 Check calibration results (test)

• using test data

# check calibration results

calibration_results <- cbind(
  eval_bayes_ucl_20260418$calibration_data, test) %>%
  select(cup:goals_away,pH:actual, correct) %>%
  mutate_if(is.numeric, round, 2)

glimpse(calibration_results)

Rows: 36
Columns: 13
$ cup         <chr> "ucl2526", "ucl2526", "ucl2526…
$ date        <date> 2026-02-17, 2026-02-17, 2026-…
$ competition <fct> UCL, UCL, UCL, UCL, UCL, UCL, …
$ season      <fct> 2025_2026, 2025_2026, 2025_202…
$ home_team   <fct> Benfica, Galatasaray, Monaco, …
$ away_team   <fct> Real Madrid, Juventus, PSG, At…
$ goals_home  <dbl> 0, 5, 2, 2, 3, 3, 1, 0, 4, 3, …
$ goals_away  <dbl> 1, 2, 3, 0, 1, 3, 6, 2, 1, 2, …
$ pH          <dbl> 0.48, 0.57, 0.55, 0.55, 0.47, …
$ pD          <dbl> 0.08, 0.10, 0.09, 0.10, 0.10, …
$ pA          <dbl> 0.43, 0.33, 0.35, 0.35, 0.43, …
$ actual      <chr> "A", "H", "A", "H", "H", "D", …
$ correct     <lgl> FALSE, TRUE, FALSE, TRUE, TRUE…

4.5 Check metrics (logloss, brier, etc.).

• logloss, brier, accuracy & baseline_log_loss.

# metrics

eval_bayes_ucl_20260418$metrics

$log_loss
[1] 1.002739

$brier_score
[1] 0.2000082

$accuracy
[1] 0.4722222

$baseline_log_loss
[1] 0.9730874

5. Prediccion futura (test).

• usando test data

5.1 check test data

Rows: 36
Columns: 14
$ cup         <chr> "ucl2526", "ucl2526", "ucl2526…
$ date        <date> 2026-02-17, 2026-02-17, 2026-…
$ competition <fct> UCL, UCL, UCL, UCL, UCL, UCL, …
$ season      <fct> 2025_2026, 2025_2026, 2025_202…
$ home_team   <fct> Benfica, Galatasaray, Monaco, …
$ away_team   <fct> Real Madrid, Juventus, PSG, At…
$ goals_home  <int> 0, 5, 2, 2, 3, 3, 1, 0, 4, 3, …
$ goals_away  <int> 1, 2, 3, 0, 1, 3, 6, 2, 1, 2, …
$ rest_home   <dbl> 20, 20, 20, 20, 21, 21, 21, 21…
$ rest_away   <dbl> 20, 20, 20, 20, 21, 21, 21, 21…
$ goal_diff   <int> -1, 3, -1, 2, 2, 0, -5, -2, 3,…
$ form_home   <dbl> 0.8, 1.4, 1.4, 1.2, 1.0, 1.2, …
$ form_away   <dbl> -0.8, -1.4, -1.4, -1.2, -1.0, …
$ matchweek   <int> 145, 146, 147, 148, 149, 150, …

5.2 build_features (test)

# build_features

df_test <- build_features(test)
glimpse(df_test)

Rows: 36
Columns: 14
$ cup         <chr> "ucl2526", "ucl2526", "ucl2526…
$ date        <date> 2026-02-17, 2026-02-17, 2026-…
$ competition <fct> UCL, UCL, UCL, UCL, UCL, UCL, …
$ season      <fct> 2025_2026, 2025_2026, 2025_202…
$ home_team   <fct> Benfica, Galatasaray, Monaco, …
$ away_team   <fct> Real Madrid, Juventus, PSG, At…
$ goals_home  <int> 0, 5, 2, 2, 3, 3, 1, 0, 4, 3, …
$ goals_away  <int> 1, 2, 3, 0, 1, 3, 6, 2, 1, 2, …
$ rest_home   <dbl> 7, 7, 7, 7, 7, 7, 7, 7, 6, 6, …
$ rest_away   <dbl> 7, 7, 7, 7, 7, 7, 7, 7, 6, 6, …
$ goal_diff   <int> -1, 3, -1, 2, 2, 0, -5, -2, 3,…
$ form_home   <dbl> 0.0, 0.0, 0.0, 0.0, 1.0, 1.2, …
$ form_away   <dbl> 0.0, 0.0, 0.0, 0.0, -1.0, -1.2…
$ matchweek   <int> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10,…

5.3 Fit Bayesian Model (test)

Time difference of 0.673301 secs
   Length     Class      Mode 
        1 character character 

5.4 Crear partidos futuros

# crear partidos futuros

new_matches <- data.frame(
  home_team=c("Bayern Munich","Liverpool",
              "Atletico Madrid","Arsenal"),
  away_team=c("Real Madrid","PSG",
              "Barcelona","Sporting"),
  competition=c("UCL"),
  season=c("2025_2026"),
  form_home=c(0.6,0.7), 
  form_away=c(0.5,0.65),
  rest_home=c(5,4),
  rest_away=c(4,3)
    
)

glimpse(new_matches)

Rows: 4
Columns: 8
$ home_team   <chr> "Bayern Munich", "Liverpool", …
$ away_team   <chr> "Real Madrid", "PSG", "Barcelo…
$ competition <chr> "UCL", "UCL", "UCL", "UCL"
$ season      <chr> "2025_2026", "2025_2026", "202…
$ form_home   <dbl> 0.6, 0.7, 0.6, 0.7
$ form_away   <dbl> 0.50, 0.65, 0.50, 0.65
$ rest_home   <dbl> 5, 4, 5, 4
$ rest_away   <dbl> 4, 3, 4, 3

5.5 Predecir probabilidades

# Predecir probabilidades

preds <- lapply(1:nrow(new_matches), function(i){
  predict_match_probabilities(bayes_ucl_test_20260418,
                              new_matches[i,])
})

head(preds,3)

[[1]]
[[1]]$prob_1X2
   Home    Draw    Away 
0.54400 0.17025 0.28575 

[[1]]$score_distribution
        [,1]    [,2]    [,3]    [,4]    [,5]
[1,] 0.01950 0.02400 0.02350 0.01125 0.00825
[2,] 0.03575 0.05300 0.04225 0.03050 0.01850
[3,] 0.04225 0.06300 0.05400 0.03175 0.01850
[4,] 0.03275 0.04875 0.04050 0.02675 0.01375
[5,] 0.02625 0.03625 0.02750 0.02000 0.01100
[6,] 0.01500 0.02150 0.01775 0.01575 0.00875
[7,] 0.00875 0.01425 0.01175 0.00775 0.00325
        [,6]    [,7]
[1,] 0.00275 0.00175
[2,] 0.00825 0.00475
[3,] 0.00700 0.00450
[4,] 0.00900 0.00300
[5,] 0.00625 0.00200
[6,] 0.00450 0.00075
[7,] 0.00125 0.00100

[[1]]$simulations

[[2]]
[[2]]$prob_1X2
  Home   Draw   Away 
0.4775 0.1745 0.3480 

[[2]]$score_distribution
        [,1]    [,2]    [,3]    [,4]    [,5]
[1,] 0.01550 0.02225 0.02575 0.01850 0.01125
[2,] 0.03075 0.05275 0.05175 0.03750 0.01750
[3,] 0.03150 0.05775 0.05650 0.04300 0.02100
[4,] 0.02825 0.04600 0.04550 0.03025 0.02000
[5,] 0.01675 0.03825 0.02975 0.02150 0.01300
[6,] 0.00875 0.01825 0.01700 0.01275 0.00650
[7,] 0.00400 0.01100 0.01125 0.00650 0.00450
        [,6]    [,7]
[1,] 0.00325 0.00200
[2,] 0.01050 0.00450
[3,] 0.01000 0.00450
[4,] 0.00725 0.00400
[5,] 0.00750 0.00325
[6,] 0.00600 0.00325
[7,] 0.00150 0.00050

[[2]]$simulations

[[3]]
[[3]]$prob_1X2
   Home    Draw    Away 
0.50425 0.17000 0.32575 

[[3]]$score_distribution
        [,1]    [,2]    [,3]    [,4]    [,5]
[1,] 0.01625 0.02550 0.02075 0.01575 0.00850
[2,] 0.03175 0.05250 0.04525 0.03525 0.01875
[3,] 0.03550 0.05825 0.05550 0.03125 0.01825
[4,] 0.02775 0.04800 0.04025 0.03050 0.01950
[5,] 0.01775 0.03025 0.02900 0.02125 0.01075
[6,] 0.01475 0.02025 0.01575 0.01850 0.00600
[7,] 0.00725 0.00850 0.01275 0.00600 0.00500
        [,6]    [,7]
[1,] 0.00550 0.00150
[2,] 0.00925 0.00425
[3,] 0.01200 0.00500
[4,] 0.00775 0.00500
[5,] 0.00850 0.00325
[6,] 0.00300 0.00100
[7,] 0.00300 0.00150

[[3]]$simulations

NANA

5.5.1 preds[[1]]$prob_1X2

   Home    Draw    Away 
0.54400 0.17025 0.28575 

5.5.2 preds[[1]]$score_distribution

     row col
[1,]   3   2

5.5.3 pred$score_distribution

        [,1]  [,2]    [,3]    [,4]    [,5]    [,6]
[1,] 0.01950 0.024 0.02350 0.01125 0.00825 0.00275
[2,] 0.03575 0.053 0.04225 0.03050 0.01850 0.00825
        [,7]
[1,] 0.00175
[2,] 0.00475

6. Simulate Monte Carlo.

6.1 Define data to be used.

# crear partidos futuros

new_matches.1 <- data.frame(
  home_team=c("Bayern Munich","Liverpool",
              "Atletico Madrid","Arsenal"),
  away_team=c("Real Madrid","PSG",
              "Barcelona","Sporting"),
  competition=c("UCL"),
  season=c("2025_2026"),
  form_home=c(0.6,0.7), 
  form_away=c(0.5,0.65),
  rest_home=c(5,4),
  rest_away=c(4,3)
    
)

glimpse(new_matches.1)

Rows: 4
Columns: 8
$ home_team   <chr> "Bayern Munich", "Liverpool", …
$ away_team   <chr> "Real Madrid", "PSG", "Barcelo…
$ competition <chr> "UCL", "UCL", "UCL", "UCL"
$ season      <chr> "2025_2026", "2025_2026", "202…
$ form_home   <dbl> 0.6, 0.7, 0.6, 0.7
$ form_away   <dbl> 0.50, 0.65, 0.50, 0.65
$ rest_home   <dbl> 5, 4, 5, 4
$ rest_away   <dbl> 4, 3, 4, 3

6.2 Run: simulate_season_montecarlo

• Using data: new_matches.2.
  
• model object: mc_ucl_test_20260418

file_path <- "~/Documents/bayes_soccer/output/models/mc_ucl_test_20260418.rds"

start <- Sys.time()
  
  if (file.exists(file_path)) {
    # Load the object if the file exists
    mc_ucl_test_20260418 <- readRDS(file_path)
  } else {
    # Create the object
    mc_ucl_test_20260418 =
      simulate_season_montecarlo(
        bayes_ucl_test_20260418,
                               new_matches.1, 
                               nsim=1000)
    
    # Save it for future use
    saveRDS(mc_ucl_test_20260418, file = file_path)
  }

end <- Sys.time()

print(end - start)

Time difference of 0.1416502 secs

6.3 Means simulation results

mc_ucl_test_20260418$simulation_results %>% 
  group_by(team) %>% 
  summarise(point = mean(points), 
            goals_for = mean(goals_for),
            goals_against = mean(goals_against),
            position = mean(position)) %>%
  ungroup() %>%
  mutate_if(is.numeric, round, 2) %>%
  arrange(position)

6.4 Simulate partial results

• Export results using matches.1 data as a input.

# Exportar resultados

results <- data.frame(
  match = paste(
    new_matches.1$home_team, "vs",new_matches.1$away_team),
  p_home=sapply(preds, function(x) x$prob_1X2["Home"]),
  p_draw=sapply(preds, function(x) x$prob_1X2["Draw"]),
  p_away=sapply(preds, function(x) x$prob_1X2["Away"])
)

results %>% mutate_if(is.numeric, round, 3)

7. Simulate Monte Carlo.

7.1 Define data to be used.

# crear partidos futuros

new_matches.2 <- data.frame(
  home_team=c("PSG","Atletico Madrid", 
              "Bayern Munich","Arsenal"),
  away_team=c("Bayern Munich","Arsenal", 
              "PSG","Atletico Madrid"),
  competition=c("UCL"),
  season=c("2025_2026"),
  form_home=c(0.6,0.7), 
  form_away=c(0.5,0.65),
  rest_home=c(5,4),
  rest_away=c(4,3)
    
)

glimpse(new_matches.2)

Rows: 4
Columns: 8
$ home_team   <chr> "PSG", "Atletico Madrid", "Bay…
$ away_team   <chr> "Bayern Munich", "Arsenal", "P…
$ competition <chr> "UCL", "UCL", "UCL", "UCL"
$ season      <chr> "2025_2026", "2025_2026", "202…
$ form_home   <dbl> 0.6, 0.7, 0.6, 0.7
$ form_away   <dbl> 0.50, 0.65, 0.50, 0.65
$ rest_home   <dbl> 5, 4, 5, 4
$ rest_away   <dbl> 4, 3, 4, 3

7.2 Run: simulate_season_montecarlo

• Using data: new_matches.2.
  
• model object: mc_ucl_test_20260418_v2

file_path <- "~/Documents/bayes_soccer/output/models/mc_ucl_test_20260418_v2.rds"

start <- Sys.time()
  
  if (file.exists(file_path)) {
    # Load the object if the file exists
    mc_ucl_test_20260418_v2 <- readRDS(file_path)
  } else {
    # Create the object
    mc_ucl_test_20260418_v2 =
      simulate_season_montecarlo(
        bayes_ucl_test_20260418,
                               new_matches.2, 
                               nsim=1000)
    
    # Save it for future use
    saveRDS(mc_ucl_test_20260418_v2, file = file_path)
  }

end <- Sys.time()

print(end - start)

Time difference of 23.13425 mins

7.3 Means simulation results

mc_ucl_test_20260418_v2$simulation_results %>% 
  group_by(team) %>% 
  summarise(point = mean(points), 
            goals_for = mean(goals_for),
            goals_against = mean(goals_against),
            position = mean(position)) %>%
  ungroup() %>%
  mutate_if(is.numeric, round, 2) %>%
  arrange(position)

7.4 Simulate partial results

• Export results using matches.2 data as a input.

# Exportar resultados

results.2 <- data.frame(
  match = paste(
    new_matches.2$home_team, "vs",new_matches.2$away_team),
  p_home=sapply(preds, function(x) x$prob_1X2["Home"]),
  p_draw=sapply(preds, function(x) x$prob_1X2["Draw"]),
  p_away=sapply(preds, function(x) x$prob_1X2["Away"])
)

results.2 %>% mutate_if(is.numeric, round, 3)