Add v7 configs for models with better deduplication

configs/v7.0-large/dataset_train.yaml

@@ -0,0 +1,2 @@
+skeleton_pool: './skeleton_pool_train.yaml'
+padding: 'zero'

configs/v7.0-large/dataset_val.yaml

@@ -0,0 +1,2 @@
+skeleton_pool: './skeleton_pool_val.yaml'
+padding: 'zero'

configs/v7.0-large/evaluation.yaml

@@ -0,0 +1,14 @@
+n_support: 512
+beam_width: 32
+n_restarts: 8
+max_len: 32
+refiner_method: 'curve_fit_lm'
+numeric_head: False
+equivalence_pruning: True
+pointwise_close_criterion: 0.95
+pointwise_close_accuracy_rtol: 0.05
+pointwise_close_accuracy_atol: 0.001
+r2_close_criterion: 0.95
+refiner_p0_noise: 'uniform'
+refiner_p0_noise_kwargs: {'low': -5, 'high': 5}
+device: cuda

configs/v7.0-large/expression_space.yaml

@@ -0,0 +1,256 @@
+operators:
+  "+":
+    realization: "+"
+    alias: ["add", "plus"]
+    inverse: "-"
+    arity: 2
+    weight: 10
+    precedence: 1
+    commutative: true
+    symmetry: 0
+    positive: false
+    monotonicity: 0
+  "-":
+    realization: "-"
+    alias: ["sub", "minus"]
+    inverse: "+"
+    arity: 2
+    weight: 5
+    precedence: 1
+    commutative: false
+    symmetry: 0
+    positive: false
+    monotonicity: 0
+  neg:
+    realization: "nsrops.neg"
+    alias: ["negative"]
+    inverse: "neg"
+    arity: 1
+    weight: 5
+    precedence: 2.5
+    commutative: false
+    symmetry: -1 # anti-symmetric
+    positive: false
+    monotonicity: -1
+  "*":
+    realization: "*"
+    alias: ["mul", "times"]
+    inverse: "/"
+    arity: 2
+    weight: 10
+    precedence: 2
+    commutative: true
+    symmetry: 0
+    positive: false
+    monotonicity: 0
+  "/":
+    realization: "/"
+    alias: ["div", "divide"]
+    inverse: "*"
+    arity: 2
+    weight: 5
+    precedence: 2
+    commutative: false
+    symmetry: 0
+    positive: false
+    monotonicity: 0
+  abs:
+    realization: "abs"
+    alias: ["absolute"]
+    inverse: null
+    arity: 1
+    weight: 4
+    precedence: 3
+    commutative: false
+    symmetry: 1 # symmetric
+    positive: true
+    monotonicity: 0
+  inv:
+    realization: "nsrops.inv"
+    alias: ["inverse"]
+    inverse: "inv"
+    arity: 1
+    weight: 4
+    precedence: 4
+    commutative: false
+    symmetry: -1 # anti-symmetric
+    positive: false
+    monotonicity: -1
+  pow2:
+    realization: "nsrops.pow2"
+    alias: ["square"]
+    inverse: null
+    arity: 1
+    weight: 4
+    precedence: 3
+    commutative: false
+    symmetry: 1 # symmetric
+    positive: true
+    monotonicity: 0
+  pow3:
+    realization: "nsrops.pow3"
+    alias: ["cube"]
+    inverse: "pow1_3"
+    arity: 1
+    weight: 2
+    precedence: 3
+    commutative: false
+    symmetry: -1 # anti-symmetric
+    positive: false
+    monotonicity: 1
+  pow4:
+    realization: "nsrops.pow4"
+    alias: []
+    inverse: null
+    arity: 1
+    weight: 1
+    precedence: 3
+    commutative: false
+    symmetry: 1
+    positive: true
+    monotonicity: 0
+  pow5:
+    realization: "nsrops.pow5"
+    alias: []
+    inverse: "pow1_5"
+    arity: 1
+    weight: 1
+    precedence: 3
+    commutative: false
+    symmetry: -1 # anti-symmetric
+    positive: false
+    monotonicity: 1
+  pow1_2:
+    realization: "nsrops.pow1_2"
+    alias: ["sqrt"]
+    inverse: null
+    arity: 1
+    weight: 4
+    precedence: 3
+    commutative: false
+    symmetry: 0
+    positive: true
+    monotonicity: 1
+  pow1_3:
+    realization: "nsrops.pow1_3"
+    alias: []
+    inverse: null
+    arity: 1
+    weight: 2
+    precedence: 3
+    commutative: false
+    symmetry: -1 # anti-symmetric
+    positive: false
+    monotonicity: 1
+  pow1_4:
+    realization: "nsrops.pow1_4"
+    alias: []
+    inverse: null
+    arity: 1
+    weight: 1
+    precedence: 3
+    commutative: false
+    symmetry: 0
+    positive: true
+    monotonicity: 1
+  pow1_5:
+    realization: "nsrops.pow1_5"
+    alias: []
+    inverse: null
+    arity: 1
+    weight: 1
+    precedence: 3
+    commutative: false
+    symmetry: -1 # anti-symmetric
+    positive: false
+    monotonicity: 1
+  sin:
+    realization: "numpy.sin"
+    alias: []
+    inverse: "asin"
+    arity: 1
+    weight: 4
+    precedence: 2
+    commutative: false
+    symmetry: -1 # anti-symmetric
+    positive: false
+    monotonicity: 0
+  cos:
+    realization: "numpy.cos"
+    alias: []
+    inverse: "acos"
+    arity: 1
+    weight: 4
+    precedence: 2
+    commutative: false
+    symmetry: 1 # symmetric
+    positive: false
+    monotonicity: 0
+  tan:
+    realization: "numpy.tan"
+    alias: []
+    inverse: "atan"
+    arity: 1
+    weight: 4
+    precedence: 2
+    commutative: false
+    symmetry: -1 # anti-symmetric
+    positive: false
+    monotonicity: 0
+  asin:
+    realization: "numpy.arcsin"
+    alias: ["arcsin"]
+    inverse: "sin"
+    arity: 1
+    weight: 2
+    precedence: 2
+    commutative: false
+    symmetry: -1 # anti-symmetric
+    positive: false
+    monotonicity: 1
+  acos:
+    realization: "numpy.arccos"
+    alias: ["arccos"]
+    inverse: "cos"
+    arity: 1
+    weight: 2
+    precedence: 2
+    commutative: false
+    symmetry: 0
+    positive: true
+    monotonicity: 1
+  atan:
+    realization: "numpy.arctan"
+    alias: ["arctan"]
+    inverse: "tan"
+    arity: 1
+    weight: 2
+    precedence: 2
+    commutative: false
+    symmetry: -1 # anti-symmetric
+    positive: false
+    monotonicity: 1
+  exp:
+    realization: "numpy.exp"
+    alias: []
+    inverse: "log"
+    arity: 1
+    weight: 4
+    precedence: 3
+    commutative: false
+    symmetry: 0
+    positive: true
+    monotonicity: 1
+  log:
+    realization: "numpy.log"
+    alias: ["ln"]
+    inverse: "exp"
+    arity: 1
+    weight: 4
+    precedence: 2
+    commutative: false
+    symmetry: 0
+    positive: false
+    monotonicity: 1
+
+variables: 3

configs/v7.0-large/nsr.yaml

@@ -0,0 +1,25 @@
+encoder_max_n_variables: 4  # includes the dependent variable
+size: 1024
+
+pre_encoder_input_type: "ieee-754"
+pre_encoder_support_nan: False
+
+encoder: "SetTransformer"
+encoder_kwargs:
+  hidden_size: 1024
+  n_enc_isab: 12
+  n_dec_sab: 6
+  n_induce: 384
+  n_heads: 16
+  layer_norm: False
+  n_seeds: 384
+
+decoder_n_heads: 16
+decoder_ff_size: 1024
+decoder_dropout: 0.1
+decoder_n_layers: 12
+
+learnable_positional_embeddings: false
+max_input_length: null
+
+expression_space: './expression_space.yaml'

configs/v7.0-large/skeleton_pool_train.yaml

@@ -0,0 +1,37 @@
+expression_space: './expression_space.yaml'
+holdout_pools: [
+  "{{ROOT}}/data/ansr-data/v7.0-large/skeleton_pool_val/",
+  "{{ROOT}}/data/ansr-data/test_set/soose_nc/skeleton_pool/",
+  "{{ROOT}}/data/ansr-data/test_set/feynman/skeleton_pool/",
+  "{{ROOT}}/data/ansr-data/test_set/nguyen/skeleton_pool/",
+  "{{ROOT}}/data/ansr-data/test_set/pool_15/skeleton_pool/"
+]
+
+sample_strategy:
+  n_operator_distribution: "length_proportional"
+  min_operators: 0
+  max_operators: 10
+  power: 1
+  max_length: 21
+  max_tries: 1
+  independent_dimensions: True
+
+allow_nan: False
+simplify: True
+
+literal_prior: 'uniform'
+literal_prior_kwargs:
+  low: -5
+  high: 5
+
+support_prior: "uniform_intervals"
+support_prior_kwargs:
+  low: -10
+  high: 10
+
+n_support_prior: "uniform"
+n_support_prior_kwargs:
+  low: 16
+  high: 512
+  min_value: 16
+  max_value: 512

configs/v7.0-large/skeleton_pool_val.yaml

@@ -0,0 +1,36 @@
+expression_space: './expression_space.yaml'
+holdout_pools: [
+  "{{ROOT}}/data/ansr-data/test_set/soose_nc/skeleton_pool/",
+  "{{ROOT}}/data/ansr-data/test_set/feynman/skeleton_pool/",
+  "{{ROOT}}/data/ansr-data/test_set/nguyen/skeleton_pool/",
+  "{{ROOT}}/data/ansr-data/test_set/pool_15/skeleton_pool/"
+]
+
+sample_strategy:
+  n_operator_distribution: "length_proportional"
+  min_operators: 0
+  max_operators: 10
+  power: 1
+  max_length: 21
+  max_tries: 1
+  independent_dimensions: True
+
+allow_nan: False
+simplify: True
+
+literal_prior: 'uniform'
+literal_prior_kwargs:
+  low: -5
+  high: 5
+
+support_prior: "uniform_intervals"
+support_prior_kwargs:
+  low: -10
+  high: 10
+
+n_support_prior: "uniform"
+n_support_prior_kwargs:
+  low: 16
+  high: 512
+  min_value: 16
+  max_value: 512