_targets.R

# === Snails --------------------------------------------------------------



# Functions ---------------------------------------------------------------
lapply(dir('R', '*.R', full.names = TRUE), source)


# Renv --------------------------------------------------------------------
snapshot()
restore()


# Options -----------------------------------------------------------------
tar_option_set(format = 'qs', 
               workspace_on_error = TRUE)

set.seed(53)



# Data --------------------------------------------------------------------
if(!dir.exists('input')) dir.create('input')
if(!dir.exists('output')) dir.create('output')
if(!dir.exists('figures')) dir.create('figures')

# Snail movement
path <- file.path('input', 'SnailDataUTM.csv')

# Proximity rasters
edge <- raster(file.path('input', 'edgedist.tif'))
brickedge1 <- raster(file.path('input', 'brickedge1.tif'))
brickedge2 <- raster(file.path('input', 'brickedge2.tif'))
brickedge3 <- raster(file.path('input', 'brickedge3.tif'))



# Variables ---------------------------------------------------------------
# Columns
id <- 'snail'
datetime <- 't'
x <- 'x'
y <- 'y'

# CRS
crs <- CRS(st_crs(32621)$wkt)

# Column to split analysis by
splitBy <- id

# Resampling rate 
rate <- hours(1)

# Tolerance rate 
tolerance <- minutes(2)

# Levels
treatment_levels <- c('C', '1', '2', '4')
stage_levels <- c('B', 'A')
ghostbrick_levels <- c('g1', 'g2', 'g3', '1', '2', '3')



# Targets: prep -----------------------------------------------------------
targets_prep <- c(
  # Read input data
  tar_target(
    input,
    fread(path)
  ),
  
  # Prep columns
  tar_target(
    prepared,
    prep_cols(input)
  ),
  
  # Remove duplicated and incomplete observations
  tar_target(
    mkunique,
    make_unique_complete(prepared, id, datetime, x, y)
  )
)



# Targets: splits ---------------------------------------------------------
targets_splits <- c(
  # Set up split -- these are our iteration units
  tar_target(
    splits,
    mkunique[, tar_group := .GRP, by = splitBy],
    iteration = 'group'
  ),
  
  tar_target(
    splitsnames,
    unique(mkunique[, .(path = path), by = splitBy])
  )
)



# Targets: iSSA -----------------------------------------------------------
targets_issa <- c(
  ## Make tracks 
  # Note from here on, when we want to iterate use pattern = map(x)
  # where x is the upstream target name
  tar_target(
    tracks,
    make_track(splits, x, y, t, crs = crs, id = snail),
    pattern = map(splits)
  ),
  
  ## Resample sampling rate
  # Regular
  tar_target(
    resamples,
    resample_tracks(tracks, rate, tolerance, binomial = FALSE),
    pattern = map(tracks)
  ),
  # Binomial move/not
  tar_target(
    binomial_resamples,
    resample_tracks(tracks, rate, tolerance, binomial = TRUE),
    pattern = map(tracks)
  ),
  
  ## Create random steps and extract covariates
  # Regular
  tar_target(
    random_steps,
    make_random_steps(resamples, 45, brickedge1, brickedge2, brickedge3, edge),
    pattern = map(resamples)
  ),

  
  ## Create step ID across individuals
  # Regular
  tar_target(
    step_id,
    make_step_id(random_steps),
    pattern = map(random_steps)
  ),

  
  ## Merge prep data back
  # Regular
  tar_target(
    merge_prep,
    merge_steps(step_id, splits, limit_edge = TRUE),
    pattern = map(step_id, splits)
  ),
  # Binomial
  tar_target(
    binomial_merge_prep,
    merge_steps(binomial_resamples, splits, limit_edge = FALSE),
    pattern = map(binomial_resamples, splits)
  ),
  
  # Prep binomial for model
  tar_target(
    prep_binomial,
    prep_model_binomial(binomial_merge_prep)
  )
)



# Targets: treatments -----------------------------------------------------
targets_treatments <- c(
  tar_target(
    brick_treats,
    brick_treatments(
      merge_prep
    )
  ),
  tar_target(
    control_treats,
    control_treatments(
      merge_prep
    )
  ),
  tar_target(
    combined_treatments,
    bind_treatments(brick_treats, control_treats)
  )
)



# Targets: model ----------------------------------------------------------
targets_models <- c(
  tar_target(
    model_select,
    model_selection(combined_treatments),
    iteration = 'list'
  ),
  
  tar_target(
    model_move,
    model_movement(combined_treatments),
    iteration = 'list'
  ),
  tar_target(
    tidy_model_move,
    tidy_model(model_move, effect = 'ran_vals')
  ),
  
  tar_target(
    model_binom,
    model_binomial(prep_binomial),
    iteration = 'list'
  )
)



# Targets: speed ----------------------------------------------------------
targets_speed <- c(
  tar_target(
    cleaned_names,
    clean_model_names(tidy_model_move)
  ),
  
  tar_target(
    tidied_coefs,
    tidy_coefs(cleaned_names, distribution_parameters),
    pattern = map(distribution_parameters)
  ),
  
  tar_target(
    predict_seq,
    make_predict_seq(combined_treatments, tidy(model_move))
  ),
  
  tar_target(
    predicted_speed,
    predict_speed(tidied_coefs, predict_seq),
    pattern = map(tidied_coefs)
  )
)



# Targets: RSS ------------------------------------------------------------
targets_rss <- c(
  tar_target(
    predicted_means,
    predict_means(combined_treatments, model_select)
  ),
  tar_target(
    predicted_bricks,
    predict_brickdist(combined_treatments, model_select)
  ),
  tar_target(
    predicted_edges,
    predict_edgedist(combined_treatments, model_select)
  ),
  tar_target(
    rss,
    calc_rss(predicted_edges, predicted_bricks, predicted_means)
  ) 
)



# Targets: plots ----------------------------------------------------------
targets_plots <- c(
  tar_target(
    fig_rss_edge,
    plot_rss_edge(rss),
    format = 'file'
  ),
  tar_target(
    fig_rss_brick,
    plot_rss_brick(rss),
    format = 'file'
  ),
  tar_target(
    fig_binomial,
    plot_binomial(model_binom),
    format = 'file'
  ),
  tar_target(
    fig_speed_brick,
    plot_speed_brick(predicted_speed),
    format = 'file'
  ),
  tar_target(
    fig_speed_edge,
    plot_speed_edge(predicted_speed),
    format = 'file'
  )
)



# Targets: tables ---------------------------------------------------------
targets_tables <- c(
  tar_target(
    move_model_table,
    tidy_model_tables(
      model_move
    )
  ),
  tar_target(
    write_move_tab,
    fwrite(move_model_table, 
           file.path('output', 'move_model_table.csv'))
  ),
  tar_target(
    select_model_table,
    tidy_model_tables(
      model_select
    )
  ),
  tar_target(
    write_select_tab,
    fwrite(select_model_table, 
           file.path('output', 'select_model_table.csv'))
  ),
  tar_target(
    binom_model_table,
    tidy_model_tables(
      model_binom
    )
  ),
  tar_target(
    write_binom_tab,
    fwrite(binom_model_table, 
           file.path('output', 'binom_model_table.csv'))
  )
)



# Targets: distributions --------------------------------------------------
targets_distributions <- c(
  # Check step distributions
  #  iteration = 'list' used for returning a list of ggplots,
  #  instead of the usual combination with vctrs::vec_c()
  tar_target(
    distributions,
    ggplot(resamples, aes(sl_)) + geom_density(alpha = 0.4),
    pattern = map(resamples),
    iteration = 'list'
  ),
  # Distribution parameters
  tar_target(
    distribution_parameters,
    calc_distribution_parameters(random_steps),
    pattern = map(random_steps)
  )
)



# Targets: all ------------------------------------------------------------
# Automatically grab all the 'targets_*' lists above
lapply(grep('targets', ls(), value = TRUE), get)

# This is equivalent to 
# c(targets_distribution, targets_issa, ...)
# Because remember - the _targets.R file always needs to end 
# in a list of targets tar_target objects