Error: TimeSeriesDataSet with target list

[robertclf]

• PyTorch-Forecasting version: 0.7.0
• PyTorch version: 1.7.0
• Python version: 3.8.5
• Operating System: Ubuntu 18.04

Hello,

I'm facing an error using TimeSeriesDataSet when trying to setup the "target" parameter with a list:

max_prediction_length = 3  # 3 months
max_encoder_length = 12

training_cutoff = data["time_idx"].max() - max_prediction_length

training = TimeSeriesDataSet(
    data[lambda x: x.time_idx <= training_cutoff],
    time_idx="time_idx",    
    # target="purchase_item_A",  # No error, works great!
    target=["purchase_item_A", "purchase_item_B"],  # Error
    group_ids=["client_id"],

    min_encoder_length=3,
    max_encoder_length=max_encoder_length,
    min_prediction_length=1,
    max_prediction_length=max_prediction_length,

    static_categoricals=["client_id"],
    time_varying_known_reals=["time_idx"],
    time_varying_unknown_reals=["purchase_item_A", "purchase_item_B"],

    target_normalizer=GroupNormalizer(
        groups=["client_id"], transformation="softplus"
    ),

    add_relative_time_idx=True,
    add_target_scales=True,
    add_encoder_length=True,

    allow_missings=True,
)

It works fine setting the target with just one column name, but when using a 2-element list it throws:

KeyError: '__target__'

During handling of the above exception, another exception occurred:

ValueError                                Traceback (most recent call last)
<ipython-input-63-932023d62ce0> in <module>
      4 training_cutoff = data["time_idx"].max() - max_prediction_length
      5 
----> 6 training = TimeSeriesDataSet(
      7     data[lambda x: x.time_idx <= training_cutoff],
      8     time_idx="time_idx",

~/anaconda3/envs/ai4/lib/python3.8/site-packages/pytorch_forecasting/data/timeseries.py in __init__(self, data, time_idx, target, group_ids, weight, max_encoder_length, min_encoder_length, min_prediction_idx, min_prediction_length, max_prediction_length, static_categoricals, static_reals, time_varying_known_categoricals, time_varying_known_reals, time_varying_unknown_categoricals, time_varying_unknown_reals, variable_groups, dropout_categoricals, constant_fill_strategy, allow_missings, add_relative_time_idx, add_target_scales, add_encoder_length, target_normalizer, categorical_encoders, scalers, randomize_length, predict_mode)
    322 
    323         # preprocess data
--> 324         data = self._preprocess_data(data)
    325 
    326         # create index

~/anaconda3/envs/ai4/lib/python3.8/site-packages/pytorch_forecasting/data/timeseries.py in _preprocess_data(self, data)
    481         data["__time_idx__"] = data[self.time_idx]  # save unscaled
    482         assert "__target__" not in data.columns, "__target__ is a protected column and must not be present in data"
--> 483         data["__target__"] = data[self.target]
    484         if self.weight is not None:
    485             data["__weight__"] = data[self.weight]

~/anaconda3/envs/ai4/lib/python3.8/site-packages/pandas/core/frame.py in __setitem__(self, key, value)
   3042         else:
   3043             # set column
-> 3044             self._set_item(key, value)
   3045 
   3046     def _setitem_slice(self, key: slice, value):

~/anaconda3/envs/ai4/lib/python3.8/site-packages/pandas/core/frame.py in _set_item(self, key, value)
   3119         self._ensure_valid_index(value)
   3120         value = self._sanitize_column(key, value)
-> 3121         NDFrame._set_item(self, key, value)
   3122 
   3123         # check if we are modifying a copy

~/anaconda3/envs/ai4/lib/python3.8/site-packages/pandas/core/generic.py in _set_item(self, key, value)
   3575         except KeyError:
   3576             # This item wasn't present, just insert at end
-> 3577             self._mgr.insert(len(self._info_axis), key, value)
   3578             return
   3579 

~/anaconda3/envs/ai4/lib/python3.8/site-packages/pandas/core/internals/managers.py in insert(self, loc, item, value, allow_duplicates)
   1187             value = _safe_reshape(value, (1,) + value.shape)
   1188 
-> 1189         block = make_block(values=value, ndim=self.ndim, placement=slice(loc, loc + 1))
   1190 
   1191         for blkno, count in _fast_count_smallints(self.blknos[loc:]):

~/anaconda3/envs/ai4/lib/python3.8/site-packages/pandas/core/internals/blocks.py in make_block(values, placement, klass, ndim, dtype)
   2720         values = DatetimeArray._simple_new(values, dtype=dtype)
   2721 
-> 2722     return klass(values, ndim=ndim, placement=placement)
   2723 
   2724 

~/anaconda3/envs/ai4/lib/python3.8/site-packages/pandas/core/internals/blocks.py in __init__(self, values, placement, ndim)
    128 
    129         if self._validate_ndim and self.ndim and len(self.mgr_locs) != len(self.values):
--> 130             raise ValueError(
    131                 f"Wrong number of items passed {len(self.values)}, "
    132                 f"placement implies {len(self.mgr_locs)}"

ValueError: Wrong number of items passed 2, placement implies 1

I read that it supports a list, so I'm wondering what I could do to fix this or workaround this problem, please.

Thanks in advance!

[jdb78]

Multiple targets are currently not implemented but it is definitely on the roadmap.

[robertclf]

oh! ok.

Thank you.

[turmeric-blend]

looking forward to this feature !! @jdb78 :)

[jdb78]

Already implemented :) But some issues with the predict method that will be resolved shortly.

[turmeric-blend]

Hmmm weird, from the demo, when I changed to input target=["volume", "avg_volume_by_sku"] like below,

training = TimeSeriesDataSet(
    data[lambda x: x.time_idx <= training_cutoff],
    time_idx="time_idx",
    target=["volume", "avg_volume_by_sku"],
    group_ids=["agency", "sku"],
    min_encoder_length=max_encoder_length // 2,  # keep encoder length long (as it is in the validation set)
    max_encoder_length=max_encoder_length,
    min_prediction_length=1,
    max_prediction_length=max_prediction_length,
    static_categoricals=["agency", "sku"],
    static_reals=["avg_population_2017", "avg_yearly_household_income_2017"],
    time_varying_known_categoricals=["special_days", "month"],
    variable_groups={"special_days": special_days},  # group of categorical variables can be treated as one variable
    time_varying_known_reals=["time_idx", "price_regular", "discount_in_percent"],
    time_varying_unknown_categoricals=[],
    time_varying_unknown_reals=[
        "volume",
        "log_volume",
        "industry_volume",
        "soda_volume",
        "avg_max_temp",
        "avg_volume_by_agency",
        "avg_volume_by_sku",
    ],
    target_normalizer=GroupNormalizer(
        groups=["agency", "sku"], transformation="softplus"
    ),  # use softplus and normalize by group
    add_relative_time_idx=True,
    add_target_scales=True,
    add_encoder_length=True,
)

I am getting this error:

ValueError: Wrong number of items passed 2, placement implies 1

@jdb78

[jdb78]

You might want to update the package.

[turmeric-blend]

I downloaded the master branch directly from zip

[jdb78]

You are using the GroupNormalizer which can handle only one target. I suggest to use the MultiNormalizer or leave the argument empty to determine it automaticaly.

[turmeric-blend]

that did the trick, thanks

[turmeric-blend]

Hi @jdb78, just to confirm, is this

Already implemented :) But some issues with the predict method that will be resolved shortly.

related to this error?:

AssertionError: Prediction should only have one extra dimension

which occurs in to_prediction() of metrics.py when I run this in a multi-target setting:

# find optimal learning rate
res = trainer.tuner.lr_find(
    tft,
    train_dataloader=train_dataloader,
    val_dataloaders=val_dataloader,
    max_lr=10.0,
    min_lr=1e-6,
)

[jdb78]

Are you having the same issue with 0.8.3?

[turmeric-blend]

I'm sorry which one is 0.8.3 this referring to? I think don't worry about it I will just wait for you to finalize this first and then I will try again. Thanks.

Already implemented :) But some issues with the predict method that will be resolved shortly.

[jdb78]

0.8.3 refers to the latest release. I think you might have something else misspecified such as no MultiLoss or MultiNormalizer. The new version will flag this.