Error with TimeSeriesDataSet: Addition/subtraction is no longer supported

[vpozdnyakov]

Environment Details

• PyTorch-Forecasting version: 0.8.5
• PyTorch version: 1.8.1+cu101
• Python version: 3.7.10
• Operating System: Google Colab web platform

Expected behavior

TimeSeriesDataSet object is successfully created

Actual behavior

I try to create a TimeSeriesDataSet object as described in README.md, but I get an error

TypeError: Addition/subtraction of integers and integer-arrays with Timestamp is no longer supported.  Instead of adding/subtracting `n`, use `n * obj.freq`

Code to reproduce the problem

Link to notebook: https://colab.research.google.com/drive/14rlslq0O1hVnQiW-tSV1lA6hX9iw0iBl?usp=sharing

!pip install pytorch-forecasting -q

from pytorch_forecasting import TimeSeriesDataSet
import pandas as pd

aus_retail = pd.read_csv('https://raw.githubusercontent.com/vpozdnyakov/probabilistic_forecasting/main/datasets/aus_retail.csv', index_col=0)
aus_retail.index = pd.to_datetime(aus_retail.index)

data = []
for col in aus_retail.iloc[:, :4].columns:
    ts = aus_retail.loc[:, [col]]
    ts['ts_id'] = col
    ts = ts.rename(columns={col: 'value'})
    data.append(ts)
data = pd.concat(data)
data.index.name = 'time_id'
data = data.reset_index()

training = TimeSeriesDataSet(
    data,
    time_idx='time_id',
    target='value',
    group_ids=['ts_id'],
    max_encoder_length=60,
    max_prediction_length=60)

Here is a traceback

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-37-8deed329ccbc> in <module>()
     23     group_ids=['ts_id'],
     24     max_encoder_length=60,
---> 25     max_prediction_length=60)

3 frames
/usr/local/lib/python3.7/dist-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, lags, add_relative_time_idx, add_target_scales, add_encoder_length, target_normalizer, categorical_encoders, scalers, randomize_length, predict_mode)
    427             # filtering for min_prediction_idx will be done on subsequence level ensuring
    428             # minimal decoder index is always >= min_prediction_idx
--> 429             data = data[lambda x: x[self.time_idx] >= self.min_prediction_idx - self.max_encoder_length - self.max_lag]
    430         data = data.sort_values(self.group_ids + [self.time_idx])
    431 

/usr/local/lib/python3.7/dist-packages/pandas/core/frame.py in __getitem__(self, key)
   2869     def __getitem__(self, key):
   2870         key = lib.item_from_zerodim(key)
-> 2871         key = com.apply_if_callable(key, self)
   2872 
   2873         if is_hashable(key):

/usr/local/lib/python3.7/dist-packages/pandas/core/common.py in apply_if_callable(maybe_callable, obj, **kwargs)
    339     """
    340     if callable(maybe_callable):
--> 341         return maybe_callable(obj, **kwargs)
    342 
    343     return maybe_callable

/usr/local/lib/python3.7/dist-packages/pytorch_forecasting/data/timeseries.py in <lambda>(x)
    427             # filtering for min_prediction_idx will be done on subsequence level ensuring
    428             # minimal decoder index is always >= min_prediction_idx
--> 429             data = data[lambda x: x[self.time_idx] >= self.min_prediction_idx - self.max_encoder_length - self.max_lag]
    430         data = data.sort_values(self.group_ids + [self.time_idx])
    431 

pandas/_libs/tslibs/timestamps.pyx in pandas._libs.tslibs.timestamps._Timestamp.__sub__()

pandas/_libs/tslibs/timestamps.pyx in pandas._libs.tslibs.timestamps._Timestamp.__add__()

TypeError: Addition/subtraction of integers and integer-arrays with Timestamp is no longer supported.  Instead of adding/subtracting `n`, use `n * obj.freq`

[vpozdnyakov]

Sorry, I missed that the time_idx column should be integer