ax/models/torch/fully_bayesian.py (9 lines): - line 11: TODO: move some of this into botorch. - line 102: # TODO: Allow Posterior to (optionally) return the full covariance matrix - line 118: TODO: use gpytorch `Distance` module. This will require some care to make sure - line 206: # TODO: test alternative outputscale priors - line 532: # use_saas is deprecated. TODO: remove - line 575: # use_saas is deprecated. TODO: remove - line 624: # TODO: Remove best_point_recommender for botorch_moo. Used in modelbridge._gen. - line 635: # use_saas is deprecated. TODO: remove - line 642: # use_saas is deprecated. TODO: remove ax/models/torch/botorch_modular/surrogate.py (6 lines): - line 88: # TODO: make optional when BoTorch model factory is checked in. - line 181: # TODO: We currently only pass in `covar_module` and `likelihood` if they are - line 255: # TODO: Create a `fit_botorch_model` method that handles the fitting. - line 294: # TODO: When we move `botorch_modular` directory to OSS, we will extend - line 325: # TODO (ref: https://fburl.com/diff/uneqb3n9) - line 338: acqf = Acquisition( # TODO: For multi-fidelity, might need diff. class. ax/models/torch/botorch_defaults.py (5 lines): - line 100: # TODO: Better logic for deciding when to use a ModelListGP. Currently the - line 130: # TODO: Is this equivalent an "else:" here? - line 172: # TODO: Add bounds for optimization stability - requires revamp upstream - line 482: # TODO: update optimizers to handle inequality_constraints - line 545: # TODO (jej): Replace with inferred noise before making perf judgements. ax/service/utils/report_utils.py (4 lines): - line 75: # TODO: implement `_get_hypervolume_trace()` - line 231: - TODO: multi-objective optimization - line 232: - TODO: ChoiceParameter plots - line 294: # TODO: Check if model can predict in favor of try/catch. ax/benchmark/benchmark_result.py (3 lines): - line 93: # TODO: If `evaluate_suggested` is True on the problem - line 105: # TODO: remove rows from [method] of length different - line 347: # TODO: Currently, the timestamps generated below must exactly match the ax/early_stopping/utils.py (3 lines): - line 22: # TODO: Allow normalizing progr_key (e.g. subtract min time stamp) - line 85: # TODO: Allow passing of additional kwargs to `interpolate` - line 86: # TODO: Allow using an arbitrary prediction model for this instead ax/models/torch/botorch.py (2 lines): - line 223: optimization problems. % TODO: refer to an example. - line 364: # TODO: Remove once https://github.com/pytorch/pytorch/issues/41489 is resolved. ax/models/torch/utils.py (2 lines): - line 328: # TODO: move this reuseable function and its equivalent reverse functions - line 520: # TODO: Allow Posterior to (optionally) return the full covariance matrix ax/plot/pareto_utils.py (2 lines): - line 373: # TODO: Verify whether 0, 1 weights cause problems because of subset_model. - line 388: # TODO: (jej) T64002590 Let this serve as a starting point for optimization. ax/storage/sqa_store/decoder.py (2 lines): - line 232: # TODO: Use metrics-like Data type field in Data instead. - line 978: # TODO: extract data type from SQAData after DataRegistry added. ax/models/torch/botorch_kg.py (2 lines): - line 151: # TODO: update optimizers to handle inequality_constraints - line 288: # Extract acquisition value (TODO: Make this less painful and repetitive) ax/utils/testing/backend_simulator.py (2 lines): - line 278: # TODO: Allow failure behavior based on a survival rate - line 442: # TODO: Improve performance / make less ad hoc by using a priority queue ax/core/search_space.py (2 lines): - line 605: # TODO: In the future, do not need to fail here; can add a "unifying" root - line 628: return visited # TODO: Should there be other validation? ax/modelbridge/transforms/percentile_y.py (1 line): - line 58: # TODO (jej): Transform covariances. ax/plot/base.py (1 line): - line 89: # @TODO T40555279: metric --> metric_name everywhere in plotting ax/benchmark/botorch_modular/standard_methods.py (1 line): - line 101: # TODO: Add commented out methods when they are brought back to modular BotAx ax/modelbridge/transforms/int_to_float.py (1 line): - line 120: obsf.parameters[p_name] = int(round(param)) # TODO: T41938776 ax/modelbridge/array.py (1 line): - line 276: # TODO: pass array_model_gen_args to _model_gen ax/storage/json_store/encoder.py (1 line): - line 166: # TODO: check size and add warning for large tensors: T69137799 setup.py (1 line): - line 11: # TODO: read pinned Botorch version from a shared source ax/storage/sqa_store/json.py (1 line): - line 41: try: # TODO T61331534: revert this; just a hotfix for AutoML ax/models/torch/botorch_moo_defaults.py (1 line): - line 248: # TODO (jej): rewrite optimize_acqf wrappers to avoid duplicate code. ax/metrics/sklearn.py (1 line): - line 133: # TODO: Consider parallelizing evaluation of large batches ax/exceptions/core.py (1 line): - line 40: It should not be used for TODO (another common use case of NIE). ax/models/torch/botorch_modular/utils.py (1 line): - line 174: # TODO: Add this if all acq. functions accept the `subset_model` ax/storage/transform_registry.py (1 line): - line 31: # TODO: Annotate and add `register_transform` ax/plot/scatter.py (1 line): - line 365: # TODO: Figure out if there's a better way to color code out-of-sample points ax/utils/testing/backend_scheduler.py (1 line): - line 124: # TODO: The status on the experiment does not distinguish between ax/models/torch/botorch_modular/list_surrogate.py (1 line): - line 52: # TODO: Allow passing down `covar_module_class`, `covar_module_options`, ax/modelbridge/transforms/inverse_gaussian_cdf_y.py (1 line): - line 49: # TODO (jej): Transform covariances. ax/service/scheduler.py (1 line): - line 997: # TODO: Add optional timeout between retries of `run_trial(s)`. ax/plot/contour.py (1 line): - line 381: # TODO T38563759: Sort parameters by feature importances ax/models/torch/botorch_modular/acquisition.py (1 line): - line 323: # TODO: It will be more memory-efficient to do this filtering before ax/modelbridge/transforms/log_y.py (1 line): - line 106: # TODO: Support covariances for a subset of observations ax/utils/measurement/synthetic_functions.py (1 line): - line 185: # TODO: support batch evaluation ax/modelbridge/modelbridge_utils.py (1 line): - line 267: # TODO: use some container down the road (similar to ax/core/experiment.py (1 line): - line 177: # TODO: maybe return a copy here to guard against implicit changes ax/plot/pareto_frontier.py (1 line): - line 511: # TODO (jej): replace dropdown with two dropdowns, one for x one for y. ax/utils/testing/core_stubs.py (1 line): - line 1161: # TODO replace with sobol ax/models/torch/botorch_moo.py (1 line): - line 202: # TODO: Remove best_point_recommender for botorch_moo. Used in modelbridge._gen. ax/benchmark/utils.py (1 line): - line 26: raise NotImplementedError # TODO (done in D18009570)