Add Presets

config/config.yaml → configs/best_quality.yaml

@@ -5,7 +5,6 @@ save_artifacts:
   enabled: False
   append_timestamp: True
   path: "./aga-artifacts"
-  # path: "s3://autogluon-assistant-agts/outputs/<run_id>/aga-artifacts/"
 feature_transformers:
   - _target_: autogluon_assistant.transformer.CAAFETransformer
     eval_model: lightgbm
@@ -25,19 +24,18 @@ autogluon:
   predictor_fit_kwargs:
     verbosity: 2
     presets: best_quality
-    time_limit: 60
+    time_limit: 14400
     dynamic_stacking: True
     num_bag_folds: 5
     num_stack_levels: 2
 llm:
   # Note: bedrock is only supported in limited AWS regions
   provider: bedrock
   api_key_location: BEDROCK_API_KEY
-  model: "anthropic.claude-3-5-sonnet-20241022-v2:0" #"anthropic.claude-3-5-sonnet-20240620-v1:0"
+  model: "anthropic.claude-3-5-sonnet-20241022-v2:0"
   # provider: openai
   # api_key_location: OPENAI_API_KEY
   # model: gpt-4o-2024-08-06
-  # model: gpt-3.5-turbo
   max_tokens: 512
   proxy_url: null
   temperature: 0

configs/high_quality.yaml

@@ -0,0 +1,29 @@
+infer_eval_metric: True
+detect_and_drop_id_column: False
+task_preprocessors_timeout: 3600
+save_artifacts:
+  enabled: False
+  append_timestamp: True
+  path: "./aga-artifacts"
+feature_transformers:
+autogluon:
+  predictor_init_kwargs: {}
+  predictor_fit_kwargs:
+    verbosity: 2
+    presets: high_quality
+    time_limit: 3600
+    dynamic_stacking: True
+    num_bag_folds: 5
+    num_stack_levels: 2
+llm:
+  # Note: bedrock is only supported in limited AWS regions
+  provider: bedrock
+  api_key_location: BEDROCK_API_KEY
+  model: "anthropic.claude-3-5-sonnet-20241022-v2:0"
+  # provider: openai
+  # api_key_location: OPENAI_API_KEY
+  # model: gpt-4o-2024-08-06
+  max_tokens: 512
+  proxy_url: null
+  temperature: 0
+  verbose: True

configs/medium_quality.yaml

@@ -0,0 +1,29 @@
+infer_eval_metric: True
+detect_and_drop_id_column: False
+task_preprocessors_timeout: 3600
+save_artifacts:
+  enabled: False
+  append_timestamp: True
+  path: "./aga-artifacts"
+feature_transformers:
+autogluon:
+  predictor_init_kwargs: {}
+  predictor_fit_kwargs:
+    verbosity: 2
+    presets: medium_quality
+    time_limit: 600
+    dynamic_stacking: True
+    num_bag_folds: 5
+    num_stack_levels: 2
+llm:
+  # Note: bedrock is only supported in limited AWS regions
+  provider: bedrock
+  api_key_location: BEDROCK_API_KEY
+  model: "anthropic.claude-3-5-sonnet-20241022-v2:0"
+  # provider: openai
+  # api_key_location: OPENAI_API_KEY
+  # model: gpt-4o-2024-08-06
+  max_tokens: 512
+  proxy_url: null
+  temperature: 0
+  verbose: True

src/autogluon_assistant/__init__.py

@@ -11,7 +11,7 @@
 from typing_extensions import Annotated
 
 from .assistant import TabularPredictionAssistant
-from .constants import NO_ID_COLUMN_IDENTIFIED
+from .constants import DEFAULT_QUALITY, NO_ID_COLUMN_IDENTIFIED, PRESETS
 from .task import TabularPredictionTask
 from .utils import load_config
 
@@ -74,6 +74,10 @@ def make_prediction_outputs(task: TabularPredictionTask, predictions: pd.DataFra
 
 def run_assistant(
     task_path: Annotated[str, typer.Argument(help="Directory where task files are included")],
+    presets: Annotated[
+        Optional[str],
+        typer.Option("--presets", "-p", help="Presets"),
+    ] = None,
     config_path: Annotated[
         Optional[str],
         typer.Option("--config-path", "-c", help="Path to the configuration file (config.yaml)"),
@@ -90,9 +94,15 @@ def run_assistant(
 ) -> str:
     logging.info("Starting AutoGluon-Assistant")
 
+    if presets is None or presets not in PRESETS:
+        logging.info(f"Presets is not provided or invalid: {presets}")
+        presets = DEFAULT_QUALITY
+        logging.info(f"Using default presets: {presets}")
+    logging.info(f"Presets: {presets}")
+
     # Load config with all overrides
     try:
-        config = load_config(config_path, config_overrides)
+        config = load_config(presets, config_path, config_overrides)
         logging.info("Successfully loaded config")
     except Exception as e:
         logging.error(f"Failed to load config: {e}")

src/autogluon_assistant/constants.py

@@ -17,6 +17,14 @@
 PROBLEM_TYPES = [BINARY, MULTICLASS, REGRESSION]
 CLASSIFICATION_PROBLEM_TYPES = [BINARY, MULTICLASS]
 
+# Presets/Configs
+CONFIGS = "configs"
+MEDIUM_QUALITY = "medium_quality"
+HIGH_QUALITY = "high_quality"
+BEST_QUALITY = "best_quality"
+DEFAULT_QUALITY = BEST_QUALITY
+PRESETS = [MEDIUM_QUALITY, HIGH_QUALITY, BEST_QUALITY]
+
 # Metrics
 ROC_AUC = "roc_auc"
 LOG_LOSS = "log_loss"

src/autogluon_assistant/utils/configs.py

@@ -4,14 +4,18 @@
 
 from omegaconf import OmegaConf
 
+from ..constants import CONFIGS
 
-def _get_default_config_path() -> Path:
+
+def _get_default_config_path(
+    presets: str,
+) -> Path:
     """
     Get default config folder under package root
     Returns Path to the config.yaml file
     """
     current_file = Path(__file__).parent.parent.parent.parent.absolute()
-    config_path = current_file / "config" / "config.yaml"
+    config_path = current_file / CONFIGS / f"{presets}.yaml"
 
     if not config_path.exists():
         raise ValueError(f"Config file not found at expected location: {config_path}")
@@ -77,7 +81,9 @@ def apply_overrides(config: Dict[str, Any], overrides: List[str]) -> Dict[str, A
     return OmegaConf.merge(config, override_conf)
 
 
-def load_config(config_path: Optional[str] = None, overrides: Optional[List[str]] = None) -> Dict[str, Any]:
+def load_config(
+    presets: str, config_path: Optional[str] = None, overrides: Optional[List[str]] = None
+) -> Dict[str, Any]:
     """
     Load configuration from yaml file, merging with default config and applying overrides
 
@@ -92,7 +98,7 @@ def load_config(config_path: Optional[str] = None, overrides: Optional[List[str]
         ValueError: If config file not found or invalid
     """
     # Load default config
-    default_config_path = _get_default_config_path()
+    default_config_path = _get_default_config_path(presets)
     logging.info(f"Loading default config from: {default_config_path}")
     config = OmegaConf.load(default_config_path)