fpl_new.qmd

---
title: "fplnew"
format: html
editor: visual
---

```{r}
library(tidyverse)
library(jsonlite)
```

```{r}
url <- "https://fantasy.premierleague.com/api/bootstrap-static"
fpl_ret <- fromJSON(url)
summary(fpl_ret)
```

```{r}
player <- fpl_ret$elements
team <- fpl_ret$teams
pos <- fpl_ret$element_types
events <- fpl_ret$events

fpl_ret$element_stats

fpl_ret$elements

```

```{r}
cleaned_players <- read_csv(file = "../Fantasy-Premier-League/data/2023-24/cleaned_players.csv")
```

```{r}
cleaned_players -> last_season

# Add current prices to last season's data
combined <- last_season |> left_join(player |> select(first_name, second_name, "now_cost_new" = now_cost ,  points_per_game), by= c("first_name", "second_name"))

combined2 <- combined|> mutate(ppm_per_cost = as.numeric(points_per_game) / as.numeric(now_cost_new / 100))

library(ggrepel)
combined2 |>
  filter(minutes > (0.5 * max(minutes, na.rm = TRUE))) |> 
  ggplot(aes(x = minutes, 
             y = ppm_per_cost, 
             colour = element_type)) + 
  geom_smooth(method = "lm",
              se = FALSE, alpha = 0.4) + 
  geom_label_repel(aes(label = second_name), alpha = 0.8) +
  labs(title = "Points per match (2023/24) per £m (2024/25 prices)",
       subtitle = "Players who played a majority of minutes",
       y = "Points per match per £m",
       x = "Minutes played",
       colour = "Position") +
  theme_bw() +
  facet_wrap(~element_type)
# 
# m1 <- lm(data = df, formula = total_points ~ creativity + influence + threat)
# 
# m2 <- lm(data = df, formula = total_points ~ creativity * influence * threat)
# 
# m3 <- mgcv::gam(formula = total_points ~ s(creativity, k = 5, bs = "cr") + s(influence, k = 5, bs = "cr") + s(threat, k = 5, bs = "cr"), data = df)
```

```{r}

combined3 <- combined2 |>
  mutate(total_per_cost = as.numeric(total_points) / as.numeric(now_cost_new / 100))

combined3 |>
  filter(minutes > (0.5 * max(minutes, na.rm = TRUE))) |> 
  ggplot(aes(x = minutes, 
             y = total_per_cost, 
             colour = element_type)) + 
  geom_smooth(method = "lm",
              se = FALSE, alpha = 0.4) + 
  geom_point() +
  geom_label_repel(aes(label = second_name), alpha = 0.8) +
  labs(title = "Total points (2023/24) per £m (2024/25 prices)",
       subtitle = "Players who played a majority of minutes",
       y = "Total points per £m",
       x = "Minutes played",
       colour = "Position") +
  theme_bw() +
  facet_wrap(~element_type)
```

```{r}
gameweeks_csvs_paths <- list.files(path = "../Fantasy-Premier-League/data/2023-24/gws/", pattern = "^gw", full.names = TRUE)

gameweeks_csvs_list <- purrr::map(.x = gameweeks_csvs_paths, .f = rio::import)

combined_df <- purrr::imap_dfr(gameweeks_csvs_list, ~ mutate(.x, gw = .y))
```

```{r}

original_prices <- combined_df |>
  filter(gw == 1) |>
  select(name, "original_price" = value)

player2 <- player |> select(first_name, second_name,
                 "influence_new" = influence,
                 "creativity_new" = creativity,
                 "threat_new" = threat) |>
  mutate(name = paste0(first_name, " ", second_name)) |>
  left_join(original_prices)

combined_df2 <- combined_df |>
  left_join(player2 |> select(name,
                             influence_new,
                             creativity_new,
                             threat_new,
                             original_price),
            by = c("name"))

combined_df3 <- combined_df2 |>
  mutate(total_points = if_else(total_points < 0, 0, total_points))

m1 <- lme4::lmer(data = combined_df2, formula = total_points ~ position + original_price + was_home +(1 | team) + (1 | opponent_team))

m2 <- lme4::lmer(data = combined_df2, formula = total_points ~ position *  was_home + (1 | team) + (1 | name) + (1 | opponent_team))

m3 <- lme4::glmer(data = combined_df3,
                  family = poisson(link = "log"), formula = total_points ~ position *  was_home + (1 | team) + (1 | name) + (1 | opponent_team))

modelsummary::modelsummary(models = list(m1, m2, m3))

```

Idea: Bayesian model based on previous 4 GWs but priors set based on 2023/24 data, allowing prediction in first GWs.