Mercurial > repos > goeckslab > multimodal_learner
changeset 12:dbf05afb8e6c draft default tip
planemo upload for repository https://github.com/goeckslab/gleam.git commit f84ea3cb29d6cd8f9e50df78048f71d9274efc47
| author | goeckslab |
|---|---|
| date | Wed, 11 Mar 2026 20:55:56 +0000 |
| parents | 499a039871d0 |
| children | |
| files | multimodal_learner.xml training_pipeline.py |
| diffstat | 2 files changed, 168 insertions(+), 51 deletions(-) [+] |
line wrap: on
line diff
--- a/multimodal_learner.xml Thu Feb 26 17:11:59 2026 +0000 +++ b/multimodal_learner.xml Wed Mar 11 20:55:56 2026 +0000 @@ -1,4 +1,4 @@ -<tool id="multimodal_learner" name="Multimodal Learner" version="0.1.7" profile="22.01"> +<tool id="multimodal_learner" name="Multimodal Learner" version="0.1.8" profile="22.01"> <description>Train and evaluate an AutoGluon Multimodal model (tabular + image + text)</description> <requirements> @@ -318,11 +318,11 @@ <data name="output_json" format="json" label="Multimodal Learner metric results on data ${on_string}"/> </outputs> - <tests> - <!-- Basic run with images + external test set --> - <test expect_num_outputs="3"> - <param name="input_csv" value="train.csv"/> - <param name="target_column" value="7"/> + <tests> + <!-- Basic run with images + external test set --> + <test expect_num_outputs="3"> + <param name="input_csv" value="train.csv"/> + <param name="target_column" value="7"/> <param name="test_dataset_conditional|has_test_dataset" value="yes"/> <param name="test_dataset_conditional|input_test" value="test.csv"/> <param name="use_images_conditional|use_images" value="yes"/> @@ -332,14 +332,14 @@ <output name="output_html"> <assert_contents> <has_text text="Model Performance Summary"/> - </assert_contents> - </output> - </test> + </assert_contents> + </output> + </test> - <!-- Custom threshold --> - <test expect_num_outputs="3"> - <param name="input_csv" value="train.csv"/> - <param name="target_column" value="7"/> + <!-- Custom threshold --> + <test expect_num_outputs="3"> + <param name="input_csv" value="train.csv"/> + <param name="target_column" value="7"/> <param name="test_dataset_conditional|has_test_dataset" value="yes"/> <param name="test_dataset_conditional|input_test" value="test.csv"/> <param name="use_images_conditional|use_images" value="yes"/> @@ -349,27 +349,27 @@ <output name="output_json"> <assert_contents> <has_text text=""threshold": 0.4"/> - </assert_contents> - </output> - </test> + </assert_contents> + </output> + </test> - <!-- No external test set; internal split --> - <test expect_num_outputs="3"> - <param name="input_csv" value="train.csv"/> - <param name="target_column" value="7"/> + <!-- No external test set; internal split --> + <test expect_num_outputs="3"> + <param name="input_csv" value="train.csv"/> + <param name="target_column" value="7"/> <param name="test_dataset_conditional|has_test_dataset" value="no"/> <param name="use_images_conditional|use_images" value="yes"/> <param name="use_images_conditional|images_zip_repeat_0|images_zip" value="images.zip"/> <output name="output_json"> <assert_contents> <has_text text=""val""/> - </assert_contents> - </output> - </test> + </assert_contents> + </output> + </test> - <!-- Text/tabular only (ignore images) --> - <test expect_num_outputs="3"> - <param name="input_csv" value="train.csv"/> + <!-- Text/tabular only (ignore images) --> + <test expect_num_outputs="3"> + <param name="input_csv" value="train.csv"/> <param name="target_column" value="7"/> <param name="test_dataset_conditional|has_test_dataset" value="yes"/> <param name="test_dataset_conditional|input_test" value="test.csv"/>
--- a/training_pipeline.py Thu Feb 26 17:11:59 2026 +0000 +++ b/training_pipeline.py Wed Mar 11 20:55:56 2026 +0000 @@ -45,6 +45,26 @@ "model.hf_text.checkpoint_name", "model.timm_image.checkpoint_name", ] +_AG_OPTIM_NAMESPACE_CUTOFF = Version("1.4.0") + + +def _detect_autogluon_version() -> Optional[Version]: + try: + ag_mod = importlib.import_module("autogluon") + ag_ver = getattr(ag_mod, "__version__", None) + if not ag_ver: + return None + return Version(str(ag_ver)) + except Exception: + return None + + +def _is_ag_14_or_newer(ag_version: Optional[Version]) -> bool: + # Default to modern namespace to avoid hard failures on AG 1.4+ when version + # detection is unavailable. + if ag_version is None: + return True + return ag_version >= _AG_OPTIM_NAMESPACE_CUTOFF def _get_env_int(keys: List[str]) -> Optional[int]: @@ -311,6 +331,94 @@ return synced +def _normalize_hparams_for_ag_version(hp: dict, ag_version: Optional[Version]) -> dict: + """ + Normalize hyperparameters to avoid deprecated/broken override keys. + For AutoGluon >=1.4, always emit canonical ``optim`` keys. + """ + normalized = _apply_dotted_overrides_to_nested(hp or {}) + + if not _is_ag_14_or_newer(ag_version): + return _synchronize_hparam_aliases(normalized) + + optim_cfg = normalized.get("optim") + if not isinstance(optim_cfg, dict): + optim_cfg = {} + + legacy_optim_cfg = normalized.get("optimization") + if isinstance(legacy_optim_cfg, dict): + for key, value in legacy_optim_cfg.items(): + if key not in optim_cfg: + optim_cfg[key] = copy.deepcopy(value) + + for key, value in list(normalized.items()): + if not isinstance(key, str) or "." not in key: + continue + prefix, leaf_key = key.split(".", 1) + if prefix not in {"optim", "optimization"}: + continue + if leaf_key not in optim_cfg: + optim_cfg[leaf_key] = copy.deepcopy(value) + + if "max_epochs" not in optim_cfg and "epochs" in optim_cfg: + optim_cfg["max_epochs"] = copy.deepcopy(optim_cfg["epochs"]) + if "lr" not in optim_cfg and "learning_rate" in optim_cfg: + optim_cfg["lr"] = copy.deepcopy(optim_cfg["learning_rate"]) + + batch_value = _MISSING + for batch_key in ("per_device_train_batch_size", "batch_size", "train_batch_size"): + if batch_key in optim_cfg and optim_cfg[batch_key] is not None: + batch_value = optim_cfg[batch_key] + break + if batch_value is not _MISSING: + optim_cfg.setdefault("per_device_train_batch_size", copy.deepcopy(batch_value)) + optim_cfg.setdefault("batch_size", copy.deepcopy(batch_value)) + + if "max_epochs" in optim_cfg: + try: + optim_cfg["max_epochs"] = int(optim_cfg["max_epochs"]) + except (TypeError, ValueError): + logger.warning("Ignoring non-integer optim.max_epochs=%s", optim_cfg["max_epochs"]) + optim_cfg.pop("max_epochs", None) + + if "lr" in optim_cfg: + try: + optim_cfg["lr"] = float(optim_cfg["lr"]) + except (TypeError, ValueError): + logger.warning("Ignoring non-numeric optim.lr=%s", optim_cfg["lr"]) + optim_cfg.pop("lr", None) + + for batch_key in ("per_device_train_batch_size", "batch_size"): + if batch_key not in optim_cfg: + continue + try: + optim_cfg[batch_key] = int(optim_cfg[batch_key]) + except (TypeError, ValueError): + logger.warning("Ignoring non-integer optim.%s=%s", batch_key, optim_cfg[batch_key]) + optim_cfg.pop(batch_key, None) + + for stale_key in ("epochs", "learning_rate", "train_batch_size"): + optim_cfg.pop(stale_key, None) + + normalized["optim"] = optim_cfg + normalized.pop("optimization", None) + + for key in list(normalized.keys()): + if not isinstance(key, str): + continue + if key.startswith("optimization."): + normalized.pop(key, None) + continue + if key in {"optim.epochs", "optim.learning_rate", "optim.train_batch_size"}: + normalized.pop(key, None) + + for key in ("max_epochs", "lr", "per_device_train_batch_size", "batch_size"): + if key in optim_cfg: + normalized[f"optim.{key}"] = copy.deepcopy(optim_cfg[key]) + + return _synchronize_hparam_aliases(normalized) + + @contextlib.contextmanager def suppress_stdout_stderr(): """Silence noisy prints from AG internals (fit_summary).""" @@ -353,14 +461,7 @@ ] text_cols = inferred_text_cols - ag_version = None - try: - ag_mod = importlib.import_module("autogluon") - ag_ver = getattr(ag_mod, "__version__", None) - if ag_ver: - ag_version = Version(str(ag_ver)) - except Exception: - ag_mod = None + ag_version = _detect_autogluon_version() def _log_missing_support(key: str) -> None: logger.info( @@ -399,6 +500,7 @@ user_hp = args.hyperparameters if isinstance(args.hyperparameters, dict) else load_user_hparams(args.hyperparameters) if user_hp and _is_valid_hp_dict(user_hp): user_hp = _apply_dotted_overrides_to_nested(user_hp) + user_hp = _normalize_hparams_for_ag_version(user_hp, ag_version) else: user_hp = {} @@ -489,7 +591,7 @@ if ag_version: logger.info(f"Detected AutoGluon version: {ag_version}; applied robust hyperparameter mappings.") - return _synchronize_hparam_aliases(hp) + return _normalize_hparams_for_ag_version(hp, ag_version) def train_predictor( @@ -693,8 +795,11 @@ cleaned[k] = v return cleaned + ag_version = _detect_autogluon_version() + use_modern_optim_namespace = _is_ag_14_or_newer(ag_version) + # Base hyperparameters following the structure described in the AutoGluon - # customization guide (env / optimization / model). + # customization guide (env / optim / model). env_cfg = {} if random_seed is not None: env_cfg["seed"] = int(random_seed) @@ -740,11 +845,17 @@ if epochs is not None: optim_cfg["max_epochs"] = int(epochs) if learning_rate is not None: - optim_cfg["learning_rate"] = float(learning_rate) + lr_val = float(learning_rate) + if use_modern_optim_namespace: + optim_cfg["lr"] = lr_val + else: + optim_cfg["learning_rate"] = lr_val if batch_size is not None: bs = int(batch_size) optim_cfg["per_device_train_batch_size"] = bs - optim_cfg["train_batch_size"] = bs + optim_cfg["batch_size"] = bs + if not use_modern_optim_namespace: + optim_cfg["train_batch_size"] = bs model_cfg = {} if eval_metric: @@ -754,29 +865,34 @@ if backbone_text: model_cfg.setdefault("hf_text", {})["checkpoint_name"] = str(backbone_text) - hp = { - "env": env_cfg, - "optimization": optim_cfg, - "model": model_cfg, - } + hp = {"env": env_cfg, "model": model_cfg} + if use_modern_optim_namespace: + hp["optim"] = copy.deepcopy(optim_cfg) + else: + hp["optimization"] = copy.deepcopy(optim_cfg) + hp["optim"] = copy.deepcopy(optim_cfg) # Also expose the most common dotted aliases for robustness across AG versions. if epochs is not None: - hp["optimization.max_epochs"] = int(epochs) hp["optim.max_epochs"] = int(epochs) + if not use_modern_optim_namespace: + hp["optimization.max_epochs"] = int(epochs) if learning_rate is not None: lr_val = float(learning_rate) - hp["optimization.learning_rate"] = lr_val - hp["optimization.lr"] = lr_val - hp["optim.learning_rate"] = lr_val hp["optim.lr"] = lr_val + if not use_modern_optim_namespace: + hp["optimization.learning_rate"] = lr_val + hp["optimization.lr"] = lr_val + hp["optim.learning_rate"] = lr_val if batch_size is not None: bs_val = int(batch_size) - hp["optimization.per_device_train_batch_size"] = bs_val - hp["optimization.batch_size"] = bs_val hp["optim.per_device_train_batch_size"] = bs_val hp["optim.batch_size"] = bs_val hp["env.per_gpu_batch_size"] = bs_val + if not use_modern_optim_namespace: + hp["optimization.per_device_train_batch_size"] = bs_val + hp["optimization.batch_size"] = bs_val + hp["optimization.train_batch_size"] = bs_val if resolved_num_workers is not None: hp["env.num_workers"] = int(resolved_num_workers) if resolved_num_workers_inference is not None: @@ -792,10 +908,11 @@ user_hp = hyperparameters else: user_hp = load_user_hparams(hyperparameters) + user_hp = _normalize_hparams_for_ag_version(user_hp, ag_version) hp = deep_update(hp, _apply_dotted_overrides_to_nested(user_hp)) - hp = _synchronize_hparam_aliases(hp) + hp = _normalize_hparams_for_ag_version(hp, ag_version) hp = _enforce_cpu_gpu_safety(hp) - hp = _synchronize_hparam_aliases(hp) + hp = _normalize_hparams_for_ag_version(hp, ag_version) hp = _prune_empty(hp) fit_cfg = {}