params.yaml

version: "1.1.0"

# Container configuration
container:
  type: "singularity"  # "singularity" or "docker"
  name: "cit-sci-traits"
  docker_tag: "latest"
  def: "cit-sci-traits.def"
  image: "cit-sci-traits.sif"

model_res: "1km"
PFT: "Shrub_Tree_Grass"
random_seed: 42
base_resolution: 1000 # Shared best resolution of all EO datasets (used for masking)
target_resolution: 1000 # Target resolution for model training
extent: [-180, -60, 180, 90] # Extent of the final dataset
crs: "EPSG:6933" # Equal Area Cylindrical projection
tmp_dir: tmp
raw_dir: data/raw
interim_dir: data/interim
processed_dir: data/processed
dask_dashboard: ":39143"
trait_mapping: "reference/trait_mapping.json"
trait_stat_mapping: "reference/trait_stat_mapping.json"
try_version: 6
splot_open: false

pylos:
  "1km":
    harmonize_eo_data:
      n_workers: 12
    match_gbif_pfts:
      n_workers: 80
      n_partitions: 80
    build_gbif_maps:
      n_workers: 20
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 60
      dask:
        n_workers: 20
        memory_limit: 40GB
    featurize_train:
      n_chunks: 6
      n_workers: 30
      memory_limit: null
      threads_per_worker: 2
    build_predict:
      n_chunks: 5
      n_workers: null
      memory_limit: 100GB
      impute_chunks: 3
    calc_spatial_autocorr:
      n_workers: 60
      n_workers_variogram: 5
      n_chunks: 4
    skcv_splits:
      n_workers: 60
    aoa:
      device_ids: [0, 1]
      predict_sample: 0.5
      splot:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: true
        train_sample: 0.5
        avg_dist_batch_size: 5000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
  "001":
    harmonize_eo_data:
      n_workers: 12
    match_gbif_pfts:
      n_workers: 80
      n_partitions: 80
    build_gbif_maps:
      n_workers: 50
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 60
      dask:
        n_workers: 40
        memory_limit: 40GB
    featurize_train:
      n_chunks: 6
      n_workers: 30
      memory_limit: null
      threads_per_worker: 5
    build_predict:
      n_chunks: 5
      n_workers: null
      memory_limit: 100GB
      impute_chunks: 3
    calc_spatial_autocorr:
      n_workers: 60
      n_workers_variogram: 5
      n_chunks: 4
    skcv_splits:
      n_workers: 60
    aoa:
      device_ids: [0, 1]
      predict_sample: 0.5
      splot:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: true
        train_sample: 0.5
        avg_dist_batch_size: 5000
        predict_partitions: null
    build_final_product:
      n_workers: 3
      threads_per_worker: 1
  "22km":
    harmonize_eo_data:
      n_workers: 20
    match_gbif_pfts:
      n_workers: 80
      n_partitions: 80
    build_gbif_maps:
      n_workers: 50
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 60
      dask:
        n_workers: 60
        memory_limit: 40GB
    featurize_train:
      n_chunks: 1
      n_workers: 60
      memory_limit: null
      threads_per_worker: 5
    build_predict:
      n_chunks: 1
      n_workers: null
      memory_limit: 100GB
      impute_chunks: 1
    calc_spatial_autocorr:
      n_workers: 60
      n_workers_variogram: 5
      n_chunks: 4
    skcv_splits:
      n_workers: 60
    aoa:
      device_ids: [0, 1]
      predict_sample: 1
      splot:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
  "02":
    harmonize_eo_data:
      n_workers: 20
    match_gbif_pfts:
      n_workers: 80
      n_partitions: 80
    build_gbif_maps:
      n_workers: 50
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 60
      dask:
        n_workers: 60
        memory_limit: 40GB
    featurize_train:
      n_chunks: 1
      n_workers: 60
      memory_limit: null
      threads_per_worker: 5
    build_predict:
      n_chunks: 1
      n_workers: null
      memory_limit: 100GB
      impute_chunks: 1
    calc_spatial_autocorr:
      n_workers: 60
      n_workers_variogram: 5
      n_chunks: 4
    skcv_splits:
      n_workers: 60
    aoa:
      device_ids: [0, 1]
      predict_sample: 1
      splot:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
  "05":
    harmonize_eo_data:
      n_workers: 18
    match_gbif_pfts:
      n_workers: 80
      n_partitions: 80
    build_gbif_maps:
      n_workers: 50
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 60
      dask:
        n_workers: 60
        memory_limit: 40GB
    featurize_train:
      n_chunks: 1
      n_workers: 60
      memory_limit: null
      threads_per_worker: 5
    build_predict:
      n_chunks: 1
      n_workers: null
      memory_limit: 100GB
      impute_chunks: 1
    calc_spatial_autocorr:
      n_workers: 60
      n_workers_variogram: 60
      n_chunks: 4
    skcv_splits:
      n_workers: 60
    aoa:
      device_ids: [0, 1]
      predict_sample: 1
      splot:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
  "1":
    harmonize_eo_data:
      n_workers: 48
    match_gbif_pfts:
      n_workers: 80
      n_partitions: 80
    build_gbif_maps:
      n_workers: 50
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 60
      dask:
        n_workers: 60
        memory_limit: 40GB
    featurize_train:
      n_chunks: 1
      n_workers: 30
      memory_limit: null
      threads_per_worker: 5
    build_predict:
      n_chunks: 1
      n_workers: null
      memory_limit: 100GB
      impute_chunks: 1
    calc_spatial_autocorr:
      n_workers: 60
      n_workers_variogram: 5
      n_chunks: 4
    skcv_splits:
      n_workers: 60
    aoa:
      device_ids: [0, 1]
      predict_sample: 1
      splot:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
  "2":
    harmonize_eo_data:
      n_workers: 30
    match_gbif_pfts:
      n_workers: 80
      n_partitions: 80
    build_gbif_maps:
      n_workers: 50
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 60
      dask:
        n_workers: 30
        memory_limit: 40GB
    featurize_train:
      n_chunks: 1
      n_workers: 30
      memory_limit: null
      threads_per_worker: 5
    build_predict:
      n_chunks: 1
      n_workers: null
      memory_limit: 100GB
      impute_chunks: 1
    calc_spatial_autocorr:
      n_workers: 20
      n_workers_variogram: 5
      n_chunks: 4
    skcv_splits:
      n_workers: 30
    aoa:
      device_ids: [0, 1]
      predict_sample: 1
      splot:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
geos:
  "1km":
    harmonize_eo_data:
      n_workers: 48
    match_gbif_pfts:
      n_workers: 60
      n_partitions: 80
    build_gbif_maps:
      n_workers: 60 # 80 was throwing too many files open error...
      threads_per_worker: 5
      memory_limit: 15GB
    build_splot_maps:
      npartitions: 60  
      dask:
        n_workers: 40
        threads_per_worker: 5
        memory_limit: 40GB
    featurize_train:
      n_chunks: 4
      n_workers: 8
      memory_limit: 100GB
      threads_per_worker: 2
    build_predict:
      n_chunks: 3
      n_workers: 80
      memory_limit: 400GB
      impute_chunks: 2
    calc_spatial_autocorr:
      n_workers: 1
      n_workers_variogram: 20  # For WGS84 coords only
      n_chunks: 12 # x direction only
      gpu_ids: [0, 1]
    skcv_splits:
      n_workers: 50
      threads_per_worker: 1
    aoa:
      device_ids: [0, 1]
      predict_sample: 0.5
      splot:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: true
        train_sample: 0.5
        avg_dist_batch_size: 5000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
  "001":
    harmonize_eo_data:
      n_workers: 48
    match_gbif_pfts:
      n_workers: 60
      n_partitions: 80
    build_gbif_maps:
      n_workers: 50
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 60
      dask:
        n_workers: 60
        memory_limit: 40GB
    featurize_train:
      n_chunks: 2
      n_workers: 100
      memory_limit: 150GB
      threads_per_worker: 5
    build_predict:
      n_chunks: 3
      n_workers: 100
      memory_limit: 500GB
      impute_chunks: 2
    calc_spatial_autocorr:
      n_workers: 100
      n_workers_variogram: 20
      n_chunks: 4
    skcv_splits:
      n_workers: 80
    aoa:
      device_ids: [0, 1, 2, 3]
      predict_sample: 0.5
      splot:
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: true
        train_sample: 0.5
        avg_dist_batch_size: 5000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
  "22km":
    harmonize_eo_data:
      n_workers: 20
    match_gbif_pfts:
      n_workers: 60
      n_partitions: 80
    build_gbif_maps:
      n_partitions: null
      n_workers: 40 # 80 was throwing too many files open error...
      threads_per_worker: 4
      memory_limit: 40GB
    build_splot_maps:
      npartitions: 40
      dask:
        n_workers: 40
        threads_per_worker: 4
        memory_limit: 40GB
    featurize_train:
      n_chunks: 10
      n_workers: 8
      memory_limit: 150GB
      threads_per_worker: 1
    build_predict:
      n_chunks: 3
      n_workers: 80
      memory_limit: 400GB
      impute_chunks: 1
    calc_spatial_autocorr:
      n_workers: 3
      n_workers_variogram: 20  # For WGS84 coords only
      n_chunks: 16 # x direction only
    skcv_splits:
      n_workers: 50
      threads_per_worker: 1
    aoa:
      device_ids: [1, 2, 3]
      predict_sample: 1
      splot:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: true
        train_sample: 1
        avg_dist_batch_size: 5000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
  "02":
    harmonize_eo_data:
      n_workers: 48
    match_gbif_pfts:
      n_workers: 60
      n_partitions: 80
    build_gbif_maps:
      n_workers: 50
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 80
      dask:
        n_workers: 80
        memory_limit: 40GB
    featurize_train:
      n_chunks: 1
      n_workers: 100
      memory_limit: 150GB
      threads_per_worker: 5
    build_predict:
      n_chunks: 1
      n_workers: 100
      memory_limit: 500GB
      impute_chunks: 1
    calc_spatial_autocorr:
      n_workers: 100
      n_workers_variogram: 20
      n_chunks: 4
    skcv_splits:
      n_workers: 80
    aoa:
      device_ids: [0, 1, 2, 3]
      predict_sample: 0.5
      splot:
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: false
        train_sample: 0.5
        avg_dist_batch_size: 5000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
  "55km":
    harmonize_eo_data:
      n_workers: 20
    match_gbif_pfts:
      n_workers: 60
      n_partitions: 80
    build_gbif_maps:
      n_partitions: null
      n_workers: 40 # 80 was throwing too many files open error...
      threads_per_worker: 4
      memory_limit: 40GB
    build_splot_maps:
      npartitions: 40
      dask:
        n_workers: 40
        threads_per_worker: 4
        memory_limit: 40GB
    featurize_train:
      n_chunks: 10
      n_workers: 8
      memory_limit: 150GB
      threads_per_worker: 1
    build_predict:
      n_chunks: 3
      n_workers: 80
      memory_limit: 400GB
      impute_chunks: 1
    calc_spatial_autocorr:
      n_workers: 3
      n_workers_variogram: 20  # For WGS84 coords only
      n_chunks: 16 # x direction only
    skcv_splits:
      n_workers: 50
      threads_per_worker: 1
    aoa:
      device_ids: [1, 2, 3]
      predict_sample: 1
      splot:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: true
        train_sample: 1
        avg_dist_batch_size: 5000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
  "05":
    harmonize_eo_data:
      n_workers: 48
    match_gbif_pfts:
      n_workers: 60
      n_partitions: 80
    build_gbif_maps:
      n_workers: 50
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 80
      dask:
        n_workers: 80
        memory_limit: 40GB
    featurize_train:
      n_chunks: 1
      n_workers: 100
      memory_limit: 150GB
      threads_per_worker: 5
    build_predict:
      n_chunks: 1
      n_workers: 100
      memory_limit: 500GB
      impute_chunks: 1
    calc_spatial_autocorr:
      n_workers: 100
      n_workers_variogram: 20
      n_chunks: 4
    skcv_splits:
      n_workers: 80
    aoa:
      device_ids: [0, 1, 2, 3]
      predict_sample: 0.5
      splot:
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: false
        train_sample: 0.5
        avg_dist_batch_size: 5000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
  "1":
    harmonize_eo_data:
      n_workers: 80
    match_gbif_pfts:
      n_workers: 60
      n_partitions: 80
    build_gbif_maps:
      n_workers: 50
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 60  
      dask:
        n_workers: 40
        threads_per_worker: 5
        memory_limit: 40GB
    featurize_train:
      n_chunks: 1
      n_workers: 100
      memory_limit: 150GB
      threads_per_worker: 5
    build_predict:
      n_chunks: 1
      n_workers: 100
      memory_limit: 500GB
      impute_chunks: 1
    calc_spatial_autocorr:
      n_workers: 100
      n_workers_variogram: 40
      n_chunks: 4
    skcv_splits:
      n_workers: 80
    aoa:
      device_ids: [0, 1, 2, 3]
      predict_sample: 0.5
      splot:
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
  "2":
    harmonize_eo_data:
      n_workers: 80
    match_gbif_pfts:
      n_workers: 60
      n_partitions: 80
    build_gbif_maps:
      n_workers: 50
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 80
      dask:
        n_workers: 80
        memory_limit: 40GB
    featurize_train:
      n_chunks: 1
      n_workers: 100
      memory_limit: 150GB
      threads_per_worker: 5
    build_predict:
      n_chunks: 1
      n_workers: 100
      memory_limit: 500GB
      impute_chunks: 1
    calc_spatial_autocorr:
      n_workers: 100
      n_workers_variogram: 40
      n_chunks: 4
    skcv_splits:
      n_workers: 80
    aoa:
      device_ids: [0, 1, 2, 3]
      predict_sample: 0.5
      splot:
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
      splot_gbif:
        chunked_dist: false
        train_sample: 1
        avg_dist_batch_size: 10000
        predict_partitions: null
    build_final_product:
      n_workers: 5
      threads_per_worker: 1
nemo2:
  "1km":
    match_gbif_pfts:
      n_workers: 60
    build_gbif_maps:
      n_workers: 32
      threads_per_worker: 5
    build_splot_maps:
      npartitions: 80  
      dask:
        n_workers: 80
        threads_per_worker: 5
        memory_limit: 40GB
    featurize_train:
      n_chunks: 4
      n_workers: 8
      memory_limit: 100GB
      threads_per_worker: 2
    calc_spatial_autocorr:
      n_workers: 1
      n_workers_variogram: 20  # For WGS84 coords only
      n_chunks: 16 # x direction only
      gpu_ids: [0, 1, 2, 3]
  "55km":
    match_gbif_pfts:
      n_workers: 60
      n_partitions: 80
    build_gbif_maps:
      n_partitions: null
      n_workers: 40 # 80 was throwing too many files open error...
      threads_per_worker: 4
      memory_limit: 40GB
    build_splot_maps:
      npartitions: 40
      dask:
        n_workers: 40
        threads_per_worker: 4
        memory_limit: 40GB
    featurize_train:
      n_chunks: 10
      n_workers: 8
      memory_limit: 150GB
      threads_per_worker: 1

mask:
  path: data/raw/esa_worldcover_v100_1km/esa_worldcover_v100_1km.tif
  keep_classes:
    - 10 # Tree cover
    - 20 # Shrubland
    - 30 # Grassland
    - 40 # Cropland
    - 60 # Bare/sparse vegetation
    - 70 # Snow and ice
    - 90 # Herbaceous wetland
    - 95 # Mangroves
    - 100 # Moss and lichen

datasets:
  X:
    canopy_height: ETH_GlobalCanopyHeight_2020_v1_1km
    modis: modis_sur_refl_monthly_avg_1km
    soilgrids: soilgrids_v2-0_1km
    vodca: vodca_mean-p5-p95_1km
    worldclim: wc2-1_30s_bio
  Y:
    splot: splot4-0
    gbif: all_tracheophyta_non-cult_2024-04-10
    fd_mode: false
    trait_stats: ["mean", "std", "median", "q05", "q95", "count"]  # non-FD metrics
    # trait_stats: ["f_ric", "f_eve", "f_div", "sp_ric"]  # FD metrics
    trait_stat: 1 # (bands start at 1, FD metrics contain only (metric, 'count')
    correlation_fn: splot_gbif_correlation.csv
    traits:
      - 4 # Stem specific density (SSD, stem dry mass per stem fresh volume) or wood density
      - 6 # Root rooting depth
      - 13 # Leaf carbon (C) content per leaf dry mass
      - 14 # Leaf nitrogen (N) content per leaf dry mass
      - 15 # Leaf phosphorus (P) content per leaf dry mass
      - 21 # Stem diameter
      - 26 # Seed dry mass
      - 27 # Seed length
      - 46 # Leaf thickness
      - 47 # Leaf dry mass per leaf fresh mass (leaf dry matter content, LDMC)
      - 50 # Leaf nitrogen (N) content per leaf area
      - 55 # Leaf dry mass (single leaf)
      - 78 # Leaf nitrogen (N) isotope signature (delta 15N)
      - 95 # Seed germination rate (germination efficiency)
      - 138 # Seed number per reproducton unit
      - 144 # Leaf length
      - 145 # Leaf width
      - 146 # Leaf carbon/nitrogen (C/N) ratio
      - 163 # Leaf fresh mass
      - 169 # Stem conduit density (vessels and tracheids)
      # - 223 # Species genotype: chromosome number
      # - 224 # Species genotype: chromosome cDNA content
      - 237 # Dispersal unit length
      - 281 # Stem conduit diameter (vessels, tracheids)
      - 282 # Wood vessel element length; stem conduit (vessel and tracheids) element length
      - 289 # Wood fiber lengths
      - 297 # Wood rays per millimetre (wood ray density)
      # - 351 # Seed number per dispersal unit
      - 614 # Fine root length per fine root dry mass (specific fine root length, SRL)
      - 1080 # Root length per root dry mass (specific root length, SRL)
      - 3106 # Plant height vegetative
      # - 3107 # Plant height generative
      # - 3112 # Leaf area (in case of compound leaves: leaf, undefined if petiole in- or excluded)
      - 3113 # Leaf area (in case of compound leaves: leaflet, undefined if petiole is in- or excluded)
      # - 3114 # Leaf area (in case of compound leaves undefined if leaf or leaflet, undefined if petiole is in- or e
      - 3117 # Leaf area per leaf dry mass (specific leaf area, SLA or 1/LMA): undefined if petiole is in- or exclu
      - 3120 # Leaf water content per leaf dry mass (not saturated)

eo_data:
  interim:
    dir: eo_data
  predict:
    dir: predict
    mask_fn: eo_predict_mask.parquet
    imputed_fn: eo_predict_imputed.parquet

worldclim:
  bio_vars:
    - "1" # Annual Mean Temperature
    - "4" # Temperature Seasonality (standard deviation ×100)
    - "7" # Temperature Annual Range (BIO5-BIO6)
    - "12" # Annual Precipitation
    - "13-14" # Annual precipitation range (BIO13-BIO14)
    - "15" # Precipitation Seasonality (Coefficient of Variation)

gbif:
  raw:
    dir: all_tracheophyta_non-cult_2024-04-10
  interim:
    dir: gbif
    matched: gbif_pfts.parquet
    subsampled: gbif_subsampled.parquet
    subsample_binsize: 4
    subsample_n_max: 10000
    min_count: 10
    max_count: 500
    traits: trait_maps

splot:
  raw:
    dir: splot4-0
  interim:
    dir: splot
    extracted: extracted
    traits: trait_maps

trydb:
  raw:
    try6:
      dir: TRY_6_gapfilled_for_distribution
      zip: TRY6_gapfilled_for_distribution.zip
      zipfile_csv: TRY6_gapfilled_for_distribution/TRY6_gapfilled_filtered_2.csv.zip
    try5:
      dir: TRY_5_GapFilledData_2020
      zip: TRY_5_GapFilledData_2020.zip
      zipfile_csv: TRY_50_2020_01/gapfilled_data/mean_gap_filled_back_transformed_incl_species_names.csv
    pfts: try_pft_v2.parquet
  interim:
    dir: try
    quantile_range: null  # Use [0.005, 0.995] for TRY5. No filtering needed for TRY6
    transform: power # "log", "power" (yeo-johnson), null (no transformation)
    already_norm: null
    filtered: traits.parquet
    transformer_fn: power_transformer.pkl
    perform_pca: false # This will reduce the number of traits by projecting them onto the principal components
    pca_n_components: 4 # Fraction of variance to keep. If integer, this is the number of components.
    pca_fn: no_pca.pkl # Set to no_pca.pkl if perform_pca is false

biomes: # Used for model analysis by biome
  reclassification:
    1: 1  # Tropical/Subtropical Forests
    2: 1
    3: 1
    4: 2  # Temperate/Boreal Forests
    5: 2
    6: 2
    7: 3  # Grasslands and Savannas
    8: 3
    12: 4 # Mediterranean
    13: 5  # Deserts
    9: 6  # Wetlands
    14: 6  
    98: 6
    10: 7  # Alpine/Polar
    11: 7
    99: 7
  raw_path: wwf_terr_biomes/wwf_terr_biomes.tif # Relative to raw_dir
  interim_path: biomes/biomes.tif # Relative to interim_dir
  
calc_spatial_autocorr:
  # Use existing spatial autocorrelation data from this model_res. To recompute, set to false.
  use_existing: false

train:
  dir: data/features
  Y:
    fn: Y.parquet
  missing_val_thresh: 0.4 # Drop features with more than this fraction of missing values
  trait_sets: ["splot", "gbif", "splot_gbif"]
  spatial_autocorr: spatial_autocorr.parquet
  cv_splits:
    range_stat: "mean"
    n_splits: 5
    n_sims: 200
    dir: skcv_splits
  weights:
    fn: feature_weights.parquet
    method: auto # "auto", "manual" -- "manual" will use below weights
    splot: 1.0
    gbif: 0.08661
  arch: "autogluon"
  eval_results: evaluation_results.csv
  feature_importance: feature_importance.csv

processed:
  dir: data/processed
  predict_dir: predict
  aoa_dir: aoa
  cov_dir: cov
  splot_corr: splot_correlation.csv

autogluon:
  included_model_types: ["GBM"]
  presets: "high"
  save_bag_folds: true
  refit_full: false
  set_best_to_refit_full: false
  cv_fit_time_limit: 7200
  full_fit_time_limit: 12600 # 3.5 hours per trait
  num_gpus: 2
  num_cpus: 74
  feature_importance: true
  FI_time_limit: 1800
  FI_num_shuffle_sets: 10

models:
  dir: models

aoa:
  dir: aoa
  
predict:
  dir: predict
  geos:
    n_workers: 1
    batches: 8
  pylos:
    n_workers: 1
    batches: 24

cov:
  dir: cov
  
public:
  destination: "both"
  local_dir: final
  sftp_dir: PANOPS/cit-sci-traits/trait_maps

analysis:
  dir: results
  multires_results_fn: all_results.parquet
  multires_fi_fn: all_fi.parquet