Summary: 125 instances, 106 unique

Text	Count
input_size=1,  # TODO fix	1
# TODO hyperparameter override for hyper parameter optimization	1
# TODO: is this the right way to do things?	1
# TODO: shouldn't we sum and divide by the number of observed values	1
# FIXME: mx.gluon.SymbolBlock cannot infer float_type and uses default np.float32	1
# TODO: add support for kernel size=1	1
# TODO: add temporal attention option	1
# FIXME: validate list elements	2
# TODO: remove	1
# TODO: cover the multivariate case here too	1
# TODO: Fix for missing values	2
# TODO check if anything can be optimized here	1
# TODO: update to mxnet cumsum when it supports axis=-1	2
# TODO: computations on knot spacings could be avoided here	1
# TODO: crps	1
# TODO: should have a "nanmean" here	1
# TODO: think about maybe only using past_ts_data[- max(5*season_length, 2*prediction_length):] for speedup	1
elif offset.name == "B":  # TODO: check this case	1
# TODO: implement custom data serializer and deserializer: convert between gluonts dataset and bytes	1
# r: growth rate shape (num_ts, ) TODO: Learn	1
# TODO add support for HPO	1
# TODO filter out time series with target shorter than prediction length	2
# TODO: support parameters	1
# TODO: on windows, operators below may produce NaN values	1
# k: carrying capacity shape (num_ts, ) TODO: Learn	1
# FIXME: we persist input/output formats of hybrid blocks as mxnet does not	1
# TODO: Allow the local model to be defined as an arbitrary local model, e.g. DF-GP and DF-LDS	1
kvstore="device",  # FIXME: initialize properly	1
# TODO optimize	2
# TODO Add parameter allowing for rolling of other arrays	1
# TODO: find out whether this is a duplicate	1
),  # TODO: pass the actual observed here	1
# FIXME: also needs to serialize the output_transform	2
# TODO validation of prediction_length and freq could also	1
# TODO: add support for static input at some point	1
# TODO: GRUCell activation is fixed to tanh	1
# TODO: find a fast way to assert absence of nans.	1
# TODO: Do we really need this check?	1
) -> None:  # TODO: we may want to return some training information here	1
# TODO be bubbled-up here from subclasses classes	1
# TODO: Define scaling for the frequency	1
!= "ignore"  # TODO: Figure out how to include 'auto' with no feat_static_cat in this check	1
TODO: This could be converted to a RecipeDataset to avoid code duplication.	2
# FIXME: https://github.com/apache/incubator-mxnet/issues/17488	3
# TODO: implement model_fn, input_fn, predict_fn, and output_fn !!	1
# TODO: segment script for readability	1
# TODO: This assumes that future_feat_dynamic has no missing values	1
# FIXME number 2	1
# FIXME number 1	1
# TODO: use event_shape	1
# TODO:	2
# TODO: not used	1
# TODO: handle conversion from image name to params, once default	1
# TODO: when we have upgraded this will give notebook progress bars	1
# TODO: Metadata and pass it instead of freq.	1
# TODO: Based on form of the prior decide to do either filtering	1
# TODO: the following are for backward compatibility	1
# TODO 1 pass inputs in proj args	1
# TODO: holidays is type List[datetime.date]	2
# TODO this is temporary, we should make the callable object serializable in the first place	1
# TODO: handle error	1
# TODO: Implement it for the general case: `rank` > 0	1
# TODO: fix handling of static features	1
# TODO: reorganize modules to avoid circular dependency	2
# TODO or fix the evaluator so it supports missing values instead (all	2
# TODO implement local mode support	1
# TODO: error checking	1
# TODO: fix mutable arguments	1
)  # TODO: handle possible exception	1
# FIXME: https://github.com/apache/incubator-mxnet/issues/11849	1
# TODO 2 concatenate inputs features to x, better names would be great	1
# TODO: optimize what we pass to the decoder for the prediction case,	1
# TODO: DONT FORGET TO PARSE ANY ADDITIONAL ARGUMENTS YOU SPECIFIED, FOR EXAMPLE THE INPUTS	1
# TODO implement proper logic handling images when none are provided by user	1
# TODO: eventually change for	1
# TODO: information about required features.	1
# TODO: Add support for symbolic case: Cannot use < operator with symbolic variables	1
# TODO does it make sense to have this then?	1
# TODO: in the extension?	1
# TODO the test set may be gone otherwise with such a filtering)	2
# TODO: here?	1
# TODO: add support for static variables at some point	1
# TODO: think about using shared context NDArrays	1
# TODO make timeout configurable	1
# TODO: Consider using shared memory for the data transfer.	1
# TODO param disable_default_callbacks to get backwards compatibility	1
# TODO: and substitute Any with Dataset	2
# TODO: WRITE YOUR CUSTOM CODE HERE	1
# TODO: given that we only support json, should we also filter json	1
# TODO: handle case with all components of the same type more efficiently when sampling	1
# TODO: In addition to passing freq, the descriptor should be carry	1
F.ones_like(past_target),  # TODO: pass the actual observed here	2
),  # TODO: use load	2
# TODO: Output the scale as well to be used by the decoder	1
# TODO: create a FormatDescriptor object that can be derived from a	1
# FIXME: try to retrieve input/output format	1
# TODO: Add support for symbolic case: Cannot use <= operator with symbolic variables	1
# TODO the following is done for backwards compatibility. For future	1
#             ),  # TODO: use load	2
# TODO clean up	2
# TODO: is the following restriction needed?	1
except Exception:  # TODO: this looks weird	1
# FIXME: prevents mxnet from failing with empty saved parameters list	1
# TODO: there used to be "nansum" here, to be fully equivalent we	1
# TODO: when mxnet has searchsorted replace this	1
# TODO switch to click	1