econml/solutions/causal_analysis/_causal_analysis.py (11 lines):
	- line 30: # TODO: this utility is documented but internal; reimplement?
	- line 32: # TODO: this utility is even less public...
	- line 360: # TODO: we can't currently handle unseen values of the feature column when getting the effect;
	- line 370: # TODO: consider addding an API to DML that allows for better understanding of how the nuisance inputs are
	- line 659: # TODO: check compatibility of X and Y lengths
	- line 709: # TODO: bail out also if categorical columns, classification, random_state changed?
	- line 713: # TODO: should we also train a new model_y under any circumstances when warm_start is True?
	- line 1002: # TODO: enrich outcome logic for multi-class classification when that is supported
	- line 1066: # TODO: Note that there's no column metadata for the sample number - should there be?
	- line 1443: # TODO: it seems like it would be better to just return the tree itself rather than plot it;
	- line 1535: # TODO: it seems like it would be better to just return the tree itself rather than plot it;


econml/iv/nnet/_deepiv.py (8 lines):
	- line 19: # TODO: make sure to use random seeds wherever necessary
	- line 20: # TODO: make sure that the public API consistently uses "T" instead of "P" for the treatment
	- line 96: # TODO: does the numeric stability actually make any difference?
	- line 340: # TODO: is there a more robust way to do this?
	- line 350: # TODO: do we need to give the user more control over other arguments to fit?
	- line 365: # TODO: do we need to give the user more control over other arguments to fit?
	- line 370: # TODO: it seems like we need to sum over the batch because we can only apply gradient to a scalar,
	- line 436: # TODO: any way to get this to work on batches of arbitrary size?


econml/utilities.py (4 lines):
	- line 167: # TODO: any way to avoid creating a copy if the array was already dense?
	- line 752: # TODO: might be faster to break into connected components first
	- line 757: # TODO: Consider investigating other performance ideas for these cases
	- line 822: # TODO: would using einsum's paths to optimize the order of merging help?


econml/iv/dr/_dr.py (3 lines):
	- line 95: # TODO: prel_model_effect could allow sample_var and freq_weight?
	- line 1539: # TODO: support freq_weight and sample_var in debiased lasso
	- line 2511: # TODO: do correct adjustment for sample_var


econml/dml/dml.py (3 lines):
	- line 419: # TODO: consider whether we need more care around stateful featurizers,
	- line 857: # TODO: support freq_weight and sample_var in debiased lasso
	- line 1116: # TODO: consider whether we need more care around stateful featurizers,


econml/sklearn_extensions/linear_model.py (2 lines):
	- line 27: # TODO: consider working around relying on sklearn implementation details
	- line 1771: # TODO: Add other types of covariance estimation (e.g. Newey-West (HAC), HC2, HC3)


econml/_cate_estimator.py (2 lines):
	- line 588: # TODO: what if input is sparse? - there's no equivalent to einsum,
	- line 1140: # TODO Share some logic with non-discrete version


econml/orf/_ortho_forest.py (2 lines):
	- line 46: # TODO: consider working around relying on sklearn implementation details
	- line 318: # TODO: Check performance


econml/inference/_bootstrap.py (2 lines):
	- line 70: # TODO: Add a __dir__ implementation?
	- line 177: # TODO: studentized bootstrap? this would be more accurate in most cases but can we avoid


econml/sklearn_extensions/model_selection.py (1 line):
	- line 16: # TODO: conisder working around relying on sklearn implementation details


prototypes/orthogonal_forests/causal_tree.py (1 line):
	- line 65: rho = - moment / grad  # TODO: Watch out for division by zero!


econml/grf/_base_grf.py (1 line):
	- line 216: # TODO: support freq_weight and sample_var


econml/iv/sieve/_tsls.py (1 line):
	- line 256: # TODO: is it right that the effective number of intruments is the


econml/dr/_drlearner.py (1 line):
	- line 1175: # TODO: support freq_weight and sample_var in debiased lasso


econml/dynamic/dml/_dml.py (1 line):
	- line 151: # TODO: update docs


econml/_ortho_learner.py (1 line):
	- line 747: # TODO: ideally, we'd also infer whether we need a GroupKFold (if groups are passed)


econml/dml/causal_forest.py (1 line):
	- line 515: # TODO: consider whether we need more care around stateful featurizers,