---
title: "@abresler's 2016 NCAA Tournament Topological Data Analysis Exploration"
author: "Alex Bresler"
output:
  flexdashboard::flex_dashboard:
    orientation: columns
    social: menu
    source_code: embed
---

```{r setup, include=FALSE}
packages <-
  c('kohonen',
    'igraph',
    'visNetwork',
    'readr',
    'tidyr',
    'stringr',
    'dplyr',
    'magrittr',
    'knitr',
    'rbokeh',
    'ggplot2',
    'flexdashboard',
    'plotly',
    'kohonen',
    'broom',
    'htmlwidgets',
    'materializeR')
lapply(packages, library, character.only = T)
all_data <-
  'https://gist.githubusercontent.com/abresler/cd488749a84f2aa54e73/raw/c019d0981bb68865810a8f79832462d5af32f51a/asb_ncaa_summary_data.csv' %>%
  read_csv
clusters <-
  6
xdim <-
  4
ydim <-
  5
kohonen.data <-
    all_data %>%
  dplyr::select(team.win_probability:tov.opp.avg)

plot_data <-
  all_data

som_grid <-
  somgrid(xdim = xdim,
          ydim = ydim,
          topo = "hexagonal")

kohonen.matrix <-
  kohonen.data %>%
  as.matrix  %>%
  scale(center = T)

names(kohonen.matrix) <-
  names(kohonen.data)

rownames(kohonen.matrix) <-
  all_data$team

som_model_ncaa <-
  kohonen.matrix %>%
  som(
    grid = som_grid,
    rlen = 100,
    # could be 100
    alpha = c(.05, .01),
    keep.data = T,
    n.hood = 'circular'
  )

som_cluster <-
  som_model_ncaa$codes %>%
  dist %>%
  hclust %>%
  cutree(clusters)

cluster.team <-
  som_cluster[som_model_ncaa$unit.classif]

group.team <-
  som_model_ncaa$unit.classif

plot_data %<>%
  mutate(cluster.team,
         group.team)

group_averages <-
  plot_data %>%
  group_by(cluster.team, group.team) %>%
  summarise_each_(funs(mean),
                  vars =
                    plot_data %>% dplyr::select(number.seed, team.win_probability:tov.opp.avg) %>% names
                  ) %>%
  left_join(
    plot_data %>%
  group_by(cluster.team, group.team) %>%
    summarise(seed.min = number.seed %>% min,
              seed.max = number.seed %>% max)
  ) %>%
  dplyr::select(cluster.team, group.team, seed.min, seed.max, everything()) %>%
  arrange((number.seed))
plot.data <-
  plot_data %>%
  dplyr::select(cluster.team,
                group.team,
                team,
                name.region,
                number.seed,
                team.win_probability,
                everything()) %>%
  arrange(desc(team.win_probability))

plot.data %<>%
  ungroup

team.nodes <-
    plot.data %>%
    ungroup %>%
    dplyr::select(label = team, group = name.region) %>%
    mutate(id = paste0("P", 1:nrow(.))) %>%
    dplyr::select(id, label, group)



group.nodes <-
    plot.data %>%
    dplyr::select(group.team) %>%
    distinct() %>%
    arrange(group.team) %>%
    mutate(label = paste('Group', group.team),
           id = paste0('G', 1:nrow(.))) %>%
    dplyr::select(id, label)

cluster.nodes <-
    plot.data %>%
    dplyr::select(cluster.team) %>%
    distinct() %>%
    arrange(cluster.team) %>%
    mutate(label = paste('Cluster', cluster.team),
           id = paste0('C', 1:nrow(.))) %>%
    dplyr::select(id, label)

  nodes <-
    cluster.nodes %>%
    bind_rows(group.nodes, team.nodes)

  id.clusters <-
    plot.data %>%
    mutate(label = paste("Cluster", cluster.team)) %>%
    dplyr::select(label, cluster.team) %>%
    distinct() %>%
    left_join(nodes) %>%
    rename(id.cluster = id)

  id.groups <-
    plot.data %>%
    mutate(label = paste("Group", group.team)) %>%
    dplyr::select(label, group.team) %>%
    distinct() %>%
    left_join(nodes) %>%
    rename(id.group = id)

  id.teams <-
    plot.data %>%
    dplyr::select(label = team, team, group = name.region) %>%
    dplyr::select(label, group) %>%
    left_join(nodes) %>%
    rename(id.team = id)


  cluster.group.edges <-
    plot.data %>%
    dplyr::select(cluster.team, group.team) %>%
    distinct() %>%
    left_join(id.clusters)

  cluster.group.edges %<>%
    dplyr::select(group.team) %>%
    left_join(id.groups) %>%
    dplyr::select(-label, -group) %>%
    left_join(cluster.group.edges) %>%
    dplyr::select(from = id.cluster, to = id.group)

  team.edges <-
    plot.data %>%
    dplyr::select(label = team, group.team) %>%
    left_join(id.teams) %>%
    dplyr::select(group.team, to = id.team)

  team.edges %<>%
    left_join(id.groups) %>%
    dplyr::select(from = id.group, to) %>%
    distinct


  edges <-
    bind_rows(cluster.group.edges,
              team.edges)

  new_nodes <-
    nodes %>%
    left_join(
      plot.data %>%
        mutate(label = paste0('\n',team,'\n', "Seed: ", number.seed,'\nTeam Win Probability: ', team.win_probability * 100 %>% digits(2), '%')) %>%
        dplyr::select(label = team, label_new = label)
    ) %>%
    mutate(label_new = ifelse(label_new %>% is.na, label, label_new)) %>%
    dplyr::select(-label) %>%
    dplyr::rename(label = label_new) %>%
    mutate(group = ifelse(label %>% str_detect("Cluster"), id, group))

  ggplot_data <-
  kohonen.matrix %>% data.frame %>% mutate(team = rownames(.)) %>% tbl_df %>%
  left_join(
    plot.data %>% dplyr::select(cluster.team, group.team, team, name.region, number.seed, actual.team_win_prob = team.win_probability )
  ) %>%
  suppressMessages() %>%
  dplyr::select(number.seed, name.region, team, cluster.team, group.team, everything()) %>%
  arrange(desc(team.win_probability))


    kohonen.group <-
      new_nodes %>%
      visNetwork(edges, width = '100%') %>%
      visGroups(groupname = "South", color = "red", ) %>%
      visGroups(groupname = "Midwest", color = "lightblue") %>%
      visGroups(groupname = "East", color = "grey") %>%
      visGroups(groupname = "West", color = "green") %>%
      visLegend()
```

Kohonen Network
=======================================================================

Column {data-width=500}
-----------------------------------------------------------------------

### [Kohonen Network](https://en.wikipedia.org/wiki/Self-organizing_map)

```{r}
kohonen.group
```


Column {data-width=500}
-----------------------------------------------------------------------

### Team Win Probabilities

```{r}
plot.data %>%
  dplyr::select(
  seed = number.seed,
  region = name.region,
  team,
  cluster = cluster.team,
  group = group.team,
  win_prob = team.win_probability,
  win_pct = pct.wins
  ) %>%
  arrange(desc(win_prob), cluster) %>%
  knitr::kable()
```

Kohonen Cluster and Group Means
=======================================================================

Column {data-width=800}
-----------------------------------------------------------------------

### Cluster/Group Means
```{r}
group_averages %>%
  knitr::kable()
```

XY Plots
=======================================================================

Row
-----------------------------------------------------------------------

### Team vs Opponent Points
```{r}
p <-
  ggplot_data %>%
  arrange(desc(cluster.team)) %>%
  mutate(cluster.team = cluster.team %>% as.character) %>%
  plot_ly(x = pts.opp, y = pts, text = paste("Team: ", team, '
', "Seed:", number.seed,
                                             '
Win Probability:', actual.team_win_prob %>% percent(2),
                                             '
Kohonen Cluster:', cluster.team,
                                             '
Kohonen Group:', group.team
                                             ),
          mode = "markers", color = cluster.team,
          textfont =
            list(
              family = "sans serif",
              size = 8,
              color = toRGB("black")
            )) %>%
  layout(
    margin = list(l = 65),
    font = list(
      family = "Courier New, verdana",
      size = 8,
      color = "black"
    ),
    xaxis = list(nticks = 20,
                 title = "Opponent Points"),
    hovermode = "closest",
    yaxis = list(
      title = "Points",
      nticks = 20
    ),
    title = "Points vs Opposition Points by Kohonen Cluster
Scaled, Centered at Mean Zero",
    titlefont = list(size = 8)
  )
loess.df <-
  loess(pts ~ pts.opp, data = ggplot_data %>% dplyr::select(pts.opp, pts) %>% arrange(pts.opp) %>% data.frame) %>%
  broom::augment()

p %<>%
  add_trace(
    x = loess.df$pts.opp,
    y = loess.df$.fitted,
    #color = group,
    name = 'loess',
    mode = 'line',
    type = 'scatter',
    line = list(
      color = toRGB("black"),
      width = 1,
      shape = 'spline'
    ),
    showlegend = T
  )
p
```

### Team Win Percentage versus Vegas Win Probability
```{r}
p <-
  ggplot_data %>%
  arrange(desc(cluster.team)) %>%
  mutate(cluster.team = cluster.team %>% as.character) %>%
  plot_ly(x = pct.wins, y = team.win_probability,
          text = paste("Team: ", team, '
', "Seed:", number.seed,
                                             '
Win Probability:', actual.team_win_prob %>% percent(2),
                                             '
Kohonen Cluster:', cluster.team,
                                             '
Kohonen Group:', group.team
                                             ),
          mode = "markers", color = cluster.team,
          textfont =
            list(
              family = "sans serif",
              size = 8,
              color = toRGB("black")
            )) %>%
  layout(
    margin = list(l = 65),
    font = list(
      family = "Courier New, verdana",
      size = 8,
      color = "black"
    ),
    xaxis = list(nticks = 20,
                 title = "Win Percetntage"),
    hovermode = "closest",
    yaxis = list(
      title = "Vegas Win Probability",
      nticks = 20
    ),
    title = "Vegas Win Probability vs Team Win Percentage by Kohonen Cluster
Scaled, Centered at Mean Zero",
    titlefont = list(size = 8)
  )
loess.df <-
  loess(pts ~ pts.opp, data = ggplot_data %>% dplyr::select(pts.opp, pts) %>% arrange(pts.opp) %>% data.frame) %>%
  broom::augment()

p %<>%
  add_trace(
    x = loess.df$pts.opp,
    y = loess.df$.fitted,
    #color = group,
    name = 'loess',
    mode = 'line',
    type = 'scatter',
    line = list(
      color = toRGB("black"),
      width = 1,
      shape = 'spline'
    ),
    showlegend = T
  )
p
```

Row
-----------------------------------------------------------------------

### 3PT vs 2PT Shooting Teams
```{r}
p <-
  ggplot_data %>%
  arrange(desc(cluster.team)) %>%
  mutate(cluster.team = cluster.team %>% as.character) %>%
  plot_ly(x = fg2m, y = fg3m, text = paste("Team: ", team, '
', "Seed:", number.seed,
                                             '
Win Probability:', actual.team_win_prob %>% percent(2),
                                             '
Kohonen Cluster:', cluster.team,
                                             '
Kohonen Group:', group.team
                                             ),
          mode = "markers", color = cluster.team,
          textfont =
            list(
              family = "sans serif",
              size = 8,
              color = toRGB("black")
            )) %>%
  layout(
    margin = list(l = 65),
    font = list(
      family = "Courier New, verdana",
      size = 8,
      color = "black"
    ),
    xaxis = list(nticks = 20,
                 title = "2PT Makes"),
    hovermode = "closest",
    yaxis = list(
      title = "3PT Makes",
      nticks = 20
    ),
    title = "3PT Makes vs 2PT Makes
Scaled, Centered at Mean Zero",
    titlefont = list(size = 8)
  )
loess.df <-
  loess(fg3m ~ fg2m, data = ggplot_data %>% dplyr::select(fg2m, fg3m) %>% arrange(fg2m) %>% data.frame) %>%
  broom::augment()

p %<>%
  add_trace(
    x = loess.df$fg2m,
    y = loess.df$.fitted,
    #color = group,
    name = 'loess',
    mode = 'line',
    type = 'scatter',
    line = list(
      color = toRGB("black"),
      width = 1,
      shape = 'spline'
    ),
    showlegend = T
  )
p
```

### Assists vs Vegas Win Probability
```{r}
p <-
  ggplot_data %>%
  arrange(desc(cluster.team)) %>%
  mutate(cluster.team = cluster.team %>% as.character) %>%
  plot_ly(x = ast, y = team.win_probability, text = paste("Team: ", team, '
', "Seed:", number.seed,
                                             '
Win Probability:', actual.team_win_prob %>% percent(2),
                                             '
Kohonen Cluster:', cluster.team,
                                             '
Kohonen Group:', group.team
                                             ),
          mode = "markers", color = cluster.team,
          textfont =
            list(
              family = "sans serif",
              size = 8,
              color = toRGB("black")
            )) %>%
  layout(
    margin = list(l = 65),
    font = list(
      family = "Courier New, verdana",
      size = 8,
      color = "black"
    ),
    xaxis = list(nticks = 20,
                 title = "Assists"),
    hovermode = "closest",
    yaxis = list(
      title = "Vegas Win Probability",
      nticks = 20
    ),
    title = "Vegas Win Probability vs Assists by Kohonen Group
Scaled, Centered at Mean Zero",
    titlefont = list(size = 8)
  )
loess.df <-
  loess(team.win_probability ~ ast, data = ggplot_data %>% dplyr::select(ast, team.win_probability) %>% arrange(ast) %>% data.frame) %>%
  broom::augment()

p %<>%
  add_trace(
    x = loess.df$ast,
    y = loess.df$.fitted,
    #color = group,
    name = 'loess',
    mode = 'line',
    type = 'scatter',
    line = list(
      color = toRGB("black"),
      width = 1,
      shape = 'spline'
    ),
    showlegend = T
  )
p
```

Scaled and Centered Team Data
=======================================================================

Column {data-width=800}
-----------------------------------------------------------------------

### [Data, Clickable for Raw Data Access](https://gist.github.com/abresler/cd488749a84f2aa54e73)
```{r}
kohonen.matrix %>% data.frame %>% mutate(team = rownames(.)) %>% tbl_df %>%
  left_join(
    plot.data %>% dplyr::select(cluster.team, group.team, team, name.region, number.seed)
  ) %>%
  suppressMessages() %>%
  dplyr::select(number.seed, name.region, team, cluster.team, group.team, everything()) %>%
  arrange(desc(team.win_probability)) %>%
  knitr::kable()
```