autogluon.tabular 1.2.1b20250407__tar.gz → 1.2.1b20250409__tar.gz

{autogluon.tabular-1.2.1b20250407 → autogluon.tabular-1.2.1b20250409}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: autogluon.tabular
-Version: 1.2.1b20250407
+Version: 1.2.1b20250409
 Summary: Fast and Accurate ML in 3 Lines of Code
 Home-page: https://github.com/autogluon/autogluon
 Author: AutoGluon Community

{autogluon.tabular-1.2.1b20250407 → autogluon.tabular-1.2.1b20250409}/src/autogluon/tabular/register/_ag_model_register.py

@@ -28,7 +28,6 @@ from ..models import (
     XGBoostModel,
     XTModel,
 )
-from ..models.tab_transformer.tab_transformer_model import TabTransformerModel
 
 
 # When adding a new model officially to AutoGluon, the model class should be added to the bottom of this list.
@@ -42,7 +41,6 @@ REGISTERED_MODEL_CLS_LST = [
     TabularNeuralNetTorchModel,
     LinearModel,
     NNFastAiTabularModel,
-    TabTransformerModel,
     TextPredictorModel,
     ImagePredictorModel,
     MultiModalPredictorModel,

{autogluon.tabular-1.2.1b20250407 → autogluon.tabular-1.2.1b20250409}/src/autogluon/tabular/version.py

@@ -1,4 +1,4 @@
 """This is the autogluon version file."""
 
-__version__ = "1.2.1b20250407"
+__version__ = "1.2.1b20250409"
 __lite__ = False

{autogluon.tabular-1.2.1b20250407 → autogluon.tabular-1.2.1b20250409}/src/autogluon.tabular.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: autogluon.tabular
-Version: 1.2.1b20250407
+Version: 1.2.1b20250409
 Summary: Fast and Accurate ML in 3 Lines of Code
 Home-page: https://github.com/autogluon/autogluon
 Author: AutoGluon Community

{autogluon.tabular-1.2.1b20250407 → autogluon.tabular-1.2.1b20250409}/src/autogluon.tabular.egg-info/SOURCES.txt

@@ -83,17 +83,6 @@ src/autogluon/tabular/models/rf/rf_rapids_model.py
 src/autogluon/tabular/models/rf/compilers/__init__.py
 src/autogluon/tabular/models/rf/compilers/native.py
 src/autogluon/tabular/models/rf/compilers/onnx.py
-src/autogluon/tabular/models/tab_transformer/__init__.py
-src/autogluon/tabular/models/tab_transformer/modified_transformer.py
-src/autogluon/tabular/models/tab_transformer/pretexts.py
-src/autogluon/tabular/models/tab_transformer/tab_model_base.py
-src/autogluon/tabular/models/tab_transformer/tab_transformer.py
-src/autogluon/tabular/models/tab_transformer/tab_transformer_encoder.py
-src/autogluon/tabular/models/tab_transformer/tab_transformer_model.py
-src/autogluon/tabular/models/tab_transformer/utils.py
-src/autogluon/tabular/models/tab_transformer/hyperparameters/__init__.py
-src/autogluon/tabular/models/tab_transformer/hyperparameters/parameters.py
-src/autogluon/tabular/models/tab_transformer/hyperparameters/searchspaces.py
 src/autogluon/tabular/models/tabpfn/__init__.py
 src/autogluon/tabular/models/tabpfn/tabpfn_model.py
 src/autogluon/tabular/models/tabpfnmix/__init__.py

{autogluon.tabular-1.2.1b20250407 → autogluon.tabular-1.2.1b20250409}/src/autogluon.tabular.egg-info/requires.txt

@@ -3,20 +3,20 @@ scipy<1.16,>=1.5.4
 pandas<2.3.0,>=2.0.0
 scikit-learn<1.5.3,>=1.4.0
 networkx<4,>=3.0
-autogluon.core==1.2.1b20250407
-autogluon.features==1.2.1b20250407
+autogluon.core==1.2.1b20250409
+autogluon.features==1.2.1b20250409
 
 [all]
+numpy<2.0.0,>=1.25
+lightgbm<4.7,>=4.0
 catboost<1.3,>=1.2
+torch<2.7,>=2.2
 fastai<2.9,>=2.3.1
+autogluon.core[all]==1.2.1b20250409
+huggingface_hub[torch]
+xgboost<2.2,>=2.0
 einops<0.9,>=0.7
 spacy<3.8
-lightgbm<4.7,>=4.0
-torch<2.6,>=2.2
-xgboost<2.2,>=2.0
-huggingface_hub[torch]
-autogluon.core[all]==1.2.1b20250407
-numpy<2.0.0,>=1.25
 
 [catboost]
 numpy<2.0.0,>=1.25
@@ -24,7 +24,7 @@ catboost<1.3,>=1.2
 
 [fastai]
 spacy<3.8
-torch<2.6,>=2.2
+torch<2.7,>=2.2
 fastai<2.9,>=2.3.1
 
 [imodels]
@@ -34,7 +34,7 @@ imodels<1.4.0,>=1.3.10
 lightgbm<4.7,>=4.0
 
 [ray]
-autogluon.core[all]==1.2.1b20250407
+autogluon.core[all]==1.2.1b20250409
 
 [skex]
 scikit-learn-intelex<2025.1,>=2024.0
@@ -54,12 +54,12 @@ onnx<1.16.2,>=1.13.0
 tabpfn<0.2,>=0.1.11
 
 [tabpfnmix]
-torch<2.6,>=2.2
+torch<2.7,>=2.2
 huggingface_hub[torch]
 einops<0.9,>=0.7
 
 [tests]
-torch<2.6,>=2.2
+torch<2.7,>=2.2
 huggingface_hub[torch]
 einops<0.9,>=0.7
 imodels<1.4.0,>=1.3.10

autogluon.tabular-1.2.1b20250407/src/autogluon/tabular/models/tab_transformer/hyperparameters/__init__.py

@@ -1 +0,0 @@
-

autogluon.tabular-1.2.1b20250407/src/autogluon/tabular/models/tab_transformer/hyperparameters/parameters.py

@@ -1,66 +0,0 @@
-def get_fixed_params():
-    """Parameters that currently cannot be searched during HPO"""
-    fixed_params = {
-        "batch_size": 512,  # The size of example chunks to predict on.
-        "n_cont_embeddings": 0,  # How many continuous feature embeddings to use.
-        "norm_class_name": "LayerNorm",  # What kind of normalization to use on continuous features.
-        "column_embedding": True,  # If True, 1/(n_shared_embs)th of every embedding will be reserved for a learned parameter that's common to all embeddings.
-        #'shared_embedding': False,
-        #'n_shared_embs': 8,
-        "orig_emb_resid": False,  # If True, concatenate the original embeddings on top of the feature embeddings in the Transformer layers.
-        "one_hot_embeddings": False,  # If True, one-hot encode variables whose cardinality is < max_emb_dim.
-        "drop_whole_embeddings": False,  # If True, dropout pretends the embedding was a missing value. If false, dropout sets embed features to 0
-        "max_emb_dim": 8,  # Maximum allowable amount of embeddings.
-        "base_exp_decay": 0.95,  # Rate of exponential decay for learning rate, used during finetuning.
-        "encoders": {
-            "CATEGORICAL": "CategoricalOrdinalEnc",  # How to "encode"(vectorize) each column type.
-            "DATETIME": "DatetimeOrdinalEnc",
-            "LATLONG": "LatLongQuantileOrdinalEnc",
-            "SCALAR": "ScalarQuantileOrdinalEnc",
-            "TEXT": "TextSummaryScalarEnc",
-        },
-        "aug_mask_prob": 0.4,  # What percentage of values to apply augmentation to.
-        "num_augs": 0,  # Number of augmentations to add.
-        "pretext": "BERTPretext",  # What pretext to use when performing pretraining/semi-supervised learning.
-        "n_cont_features": 8,  # How many continuous features to concatenate onto the categorical features
-        "fix_attention": False,  # If True, use the categorical embeddings in the transformer architecture.
-        "epochs": 200,  # How many epochs to train on with labeled data.
-        "pretrain_epochs": 200,  # How many epochs to pretrain on with unlabeled data.
-        "epochs_wo_improve": 30,  # How many epochs to continue running without improving on metric. aka "Early Stopping Patience"
-        "num_workers": 16,  # How many workers to use for torch DataLoader.
-        "max_columns": 500,  # Maximum number of columns TabTransformer will accept as input. This is to combat huge memory requirements/errors.
-        "tab_readout": "none",  # What sort of readout from the transformer. Options: ['readout_emb', 'mean', 'concat_pool', 'concat_pool_all', 'concat_pool_add', 'all_feat_embs', 'mean_feat_embs', 'none']
-    }
-
-    return fixed_params
-
-
-def get_hyper_params():
-    """Parameters that currently can be tuned during HPO"""
-    hyper_params = {
-        "lr": 3.6e-3,  # Learning rate
-        # Options: Real(5e-5, 5e-3)
-        "weight_decay": 1e-6,  # Rate of linear weight decay for learning rate
-        # Options: Real(1e-6, 5e-2)
-        "p_dropout": 0,  # dropout probability, 0 turns off Dropout.
-        # Options: Categorical(0, 0.1, 0.2, 0.3, 0.4, 0.5)
-        "n_heads": 4,  # Number of attention heads
-        # Options: Categorical(2, 4, 8)
-        "hidden_dim": 128,  # hidden dimension size
-        # Options: Categorical(32, 64, 128, 256)
-        "n_layers": 2,  # Number of Tab Transformer encoder layers,
-        # Options: Categorical(1, 2, 3, 4, 5)
-        "feature_dim": 64,  # Size of fully connected layer in TabNet.
-        # Options: Int(8, 128)
-        "num_output_layers": 1,  # How many fully-connected layers on top of transformer to produce predictions. Minimum 1 layer.
-        # Options: Categorical(1, 2, 3)
-    }
-
-    return hyper_params
-
-
-def get_default_param():
-    params = get_fixed_params()
-    params.update(get_hyper_params())
-
-    return params

autogluon.tabular-1.2.1b20250407/src/autogluon/tabular/models/tab_transformer/hyperparameters/searchspaces.py

@@ -1,17 +0,0 @@
-from autogluon.common import space
-
-
-# TODO: May have to split search space's by problem type. Not necessary right now.
-def get_default_searchspace():
-    params = {
-        "lr": space.Real(5e-5, 5e-3, default=1e-3, log=True),
-        "weight_decay": space.Real(1e-6, 5e-2, default=1e-6, log=True),
-        "p_dropout": space.Categorical(0.1, 0, 0.5),
-        "n_heads": space.Categorical(8, 4),
-        "hidden_dim": space.Categorical(128, 32, 64, 256),
-        "n_layers": space.Categorical(2, 1, 3, 4, 5),
-        "feature_dim": space.Int(8, 128, default=64),
-        "num_output_layers": space.Categorical(1, 2),
-    }
-
-    return params.copy()