Create metadata/config.yaml

equitabpfn/metadata/config.yaml

@@ -0,0 +1,215 @@
+prior:
+  num_features: 100
+  n_samples: 1152
+  eval_positions_prop: 0.95
+  heterogeneous_batches: false
+  multiclass_loss_type: nono
+  boolean:
+    max_fraction_uninformative: 0.5
+    p_uninformative: 0.5
+  prior_type: prior_bag
+  prior_bag:
+    prior_bag_exp_weights_1:
+      distribution: uniform
+      min: 2.0
+      max: 10.0
+  mlp:
+    add_uninformative_features: false
+    pre_sample_causes: true
+    sampling: normal
+    prior_mlp_scale_weights_sqrt: true
+    random_feature_rotation: true
+    num_layers:
+      distribution: meta_gamma
+      max_alpha: 2
+      max_scale: 3
+      round: true
+      lower_bound: 2
+    prior_mlp_hidden_dim:
+      distribution: meta_gamma
+      max_alpha: 3
+      max_scale: 100
+      round: true
+      lower_bound: 4
+    prior_mlp_dropout_prob:
+      distribution: meta_beta
+      scale: 0.6
+      min: 0.1
+      max: 5.0
+    init_std:
+      distribution: log_uniform
+      min: 0.01
+      max: 12
+    noise_std:
+      distribution: log_uniform
+      min: 0.0001
+      max: 0.5
+    num_causes:
+      distribution: meta_gamma
+      max_alpha: 3
+      max_scale: 7
+      round: true
+      lower_bound: 2
+    is_causal:
+      distribution: meta_choice
+      choice_values:
+      - true
+      - false
+    pre_sample_weights:
+      distribution: meta_choice
+      choice_values:
+      - true
+      - false
+    y_is_effect:
+      distribution: meta_choice
+      choice_values:
+      - true
+      - false
+    prior_mlp_activations:
+      distribution: meta_choice
+      choice_values:
+      - torch.nn.Tanh
+      - torch.nn.Identity
+      - torch.nn.ReLU
+    block_wise_dropout:
+      distribution: meta_choice
+      choice_values:
+      - true
+      - false
+    sort_features:
+      distribution: meta_choice
+      choice_values:
+      - true
+      - false
+    in_clique:
+      distribution: meta_choice
+      choice_values:
+      - true
+      - false
+  gp:
+    outputscale:
+      distribution: log_uniform
+      min: 1.0e-05
+      max: 8
+    lengthscale:
+      distribution: log_uniform
+      min: 1.0e-05
+      max: 8
+    noise:
+      distribution: meta_choice
+      choice_values:
+      - 1.0e-05
+      - 0.0001
+      - 0.01
+    sampling: normal
+  classification:
+    max_num_classes: 10
+    num_classes:
+      distribution: uniform_int
+      min: 2
+      max: 10
+    num_features_used:
+      distribution: uniform_int
+      min: 1
+      max: 100
+    balanced: false
+    output_multiclass_ordered_p: 0.0
+    categorical_feature_p: 0.2
+    multiclass_max_steps: 10
+    multiclass_type: rank
+    nan_prob_unknown_reason_reason_prior: 0.5
+    nan_prob_a_reason: 0.0
+    nan_prob_no_reason: 0.0
+    nan_prob_unknown_reason: 0.0
+    set_value_to_nan: 0.1
+model:
+  decoder:
+    name: equitabpfn.models.decoders.KDEDecoder
+    kwargs:
+      bw: 1.0
+      kernel: gaussian
+      pointwise_mlp:
+        dim_feedforward: 512
+        with_layer_norm: true
+        layer_norm_eps: 1
+        activation: gelu
+        dropout: 0.0
+  y_encoder:
+    name: equitabpfn.models.encoders.EquiOneHotAndLinear
+    kwargs:
+      num_classes: 10
+  bkbn:
+    name: equitabpfn.models.equitabpfn.EquiTabPFN
+    kwargs:
+      emsize: 512
+      nlayers: 6
+      dropout: 0.0
+      nhead: 4
+      nhid_factor: 2
+      init_method: xavier-uniform
+      recompute_attn: true
+      pre_norm: false
+      efficient_eval_masking: true
+      input_normalization: false
+      tabpfn_zero_weights: false
+      output_features: all_features
+      equivariant_encoder: false
+      feature_mask_mode: Bq2Bk
+      compile_model: true
+      decoder_kwarg:
+        name: ktabpfn.models.decoders.KDEDecoder
+        kwargs:
+          bw: 1.0
+          kernel: gaussian
+          pointwise_mlp:
+            dim_feedforward: 512
+            with_layer_norm: true
+            layer_norm_eps: 1
+            activation: gelu
+            dropout: 0.0
+  logits: true
+seed: 0
+system:
+  device: 0
+  dtype: 32
+n_samples: 1152
+max_features: 100
+max_num_classes: 10
+data_path: data
+dataloader:
+  batch_size: 24
+  num_steps: 384
+  min_eval_pos: 2
+  max_eval_pos: 1000
+training:
+  aggregate_k_gradients: 3
+  epochs: 1200
+  train_mixed_precision: true
+  eval_freq: 10
+  ckpt_freq: 10
+  compile: false
+optimizer:
+  name: torch.optim.AdamW
+  kwargs:
+    lr: 0.0001
+    weight_decay: 0.0
+scheduler:
+  warmup_epoch: 10
+  first:
+    name: torch.optim.lr_scheduler.LinearLR
+    kwargs:
+      start_factor: 1.0e-10
+      end_factor: 1
+      total_iters: 10
+  second:
+    name: torch.optim.lr_scheduler.CosineAnnealingLR
+    kwargs:
+      T_max: 1190
+      eta_min: 1.0e-08
+load:
+  model_state_path: ''
+  load_model_strict: true
+  load_existing_cktp: true
+mode:
+  eval_mode: true
+  train_mode: true