static_reals Parameter bug

[Arjein]

static_reals Parameter bug
The TimeSeriesDataset requires a static_reals parameter, which is described in the documentation as accepting a List of variables. However, when I pass a List to this parameter, I encounter an error. Interestingly, it works correctly when I provide a single String instead of a List.

# create dataset and dataloaders
max_encoder_length = 36
max_prediction_length = 6
training_cutoff = df["time_idx"].max()  - max_prediction_length

print(f'Training Cutoff: {training_cutoff}')

# Target Labels
time_idx = 'time_idx'
target_labels = 'close_price'

group_ids = ['asset_id']

static_reals=['asset_id'],

time_varying_known_reals = ["time_idx", "month", "day", "day_of_week", "hour", "minute"]


time_varying_unknown_reals = [
    "open_price", 
    "high_price", 
    "low_price", 
    "close_price", 
    "volume", 
    "quote_asset_volume", 
    "number_of_trades", 
    "taker_buy_base_asset_volume", 
    "taker_buy_quote_asset_volume"
]
# 
training = TimeSeriesDataSet(
    df[lambda x: x['time_idx'] <= training_cutoff],
    time_idx=time_idx,  
    
    target=target_labels, # For single Data Prediction
    
    group_ids=group_ids,
    time_varying_known_reals=time_varying_known_reals,
    static_reals=static_reals,
    time_varying_unknown_reals=time_varying_unknown_reals,
    
    min_encoder_length=max_encoder_length // 2,  # keep encoder length long (as it is in the validation set)
    max_encoder_length=max_encoder_length,
    max_prediction_length=max_prediction_length,
)

TypeError

TypeError                                 Traceback (most recent call last)
Cell In[47], [line 31](vscode-notebook-cell:?execution_count=47&line=31)
     [19](vscode-notebook-cell:?execution_count=47&line=19) time_varying_unknown_reals = [
     [20](vscode-notebook-cell:?execution_count=47&line=20)     "open_price", 
     [21](vscode-notebook-cell:?execution_count=47&line=21)     "high_price", 
   (...)
     [28](vscode-notebook-cell:?execution_count=47&line=28)     "taker_buy_quote_asset_volume"
     [29](vscode-notebook-cell:?execution_count=47&line=29) ]
     [30](vscode-notebook-cell:?execution_count=47&line=30) # 
---> [31](vscode-notebook-cell:?execution_count=47&line=31) training = TimeSeriesDataSet(
     [32](vscode-notebook-cell:?execution_count=47&line=32)     df[lambda x: x['time_idx'] <= training_cutoff],
     [33](vscode-notebook-cell:?execution_count=47&line=33)     time_idx=time_idx,  
     [34](vscode-notebook-cell:?execution_count=47&line=34)     
     [35](vscode-notebook-cell:?execution_count=47&line=35)     target=target_labels, # For single Data Prediction
     [36](vscode-notebook-cell:?execution_count=47&line=36)     
     [37](vscode-notebook-cell:?execution_count=47&line=37)     group_ids=group_ids,
     [38](vscode-notebook-cell:?execution_count=47&line=38)     time_varying_known_reals=time_varying_known_reals,
     [39](vscode-notebook-cell:?execution_count=47&line=39)     static_reals=static_reals,
     [40](vscode-notebook-cell:?execution_count=47&line=40)     time_varying_unknown_reals=time_varying_unknown_reals,
     [41](vscode-notebook-cell:?execution_count=47&line=41)     
     [42](vscode-notebook-cell:?execution_count=47&line=42)     min_encoder_length=max_encoder_length // 2,  # keep encoder length long (as it is in the validation set)
     [43](vscode-notebook-cell:?execution_count=47&line=43)     max_encoder_length=max_encoder_length,
     [44](vscode-notebook-cell:?execution_count=47&line=44)     max_prediction_length=max_prediction_length,
     [45](vscode-notebook-cell:?execution_count=47&line=45) 
...
    [833](https://file+.vscode-resource.vscode-cdn.net/Users/arjein/Documents/GitHub/AI/coin_pred/~/Documents/GitHub/AI/coin_pred/.venv/lib/python3.12/site-packages/pytorch_forecasting/data/timeseries.py:833)         # lagged variables are only transformed - not fitted
    [834](https://file+.vscode-resource.vscode-cdn.net/Users/arjein/Documents/GitHub/AI/coin_pred/~/Documents/GitHub/AI/coin_pred/.venv/lib/python3.12/site-packages/pytorch_forecasting/data/timeseries.py:834)         continue
    [835](https://file+.vscode-resource.vscode-cdn.net/Users/arjein/Documents/GitHub/AI/coin_pred/~/Documents/GitHub/AI/coin_pred/.venv/lib/python3.12/site-packages/pytorch_forecasting/data/timeseries.py:835)     elif name not in self._scalers:

TypeError: unhashable type: 'list'

Expected behavior
And it works when I pass a string variable.

max_encoder_length = 36
max_prediction_length = 6
training_cutoff = df["time_idx"].max()  - max_prediction_length

print(f'Training Cutoff: {training_cutoff}')

# Target Labels
time_idx = 'time_idx'
target_labels = 'close_price'

group_ids = ['asset_id']

static_reals='asset_id',

time_varying_known_reals = ["time_idx", "month", "day", "day_of_week", "hour", "minute"]


time_varying_unknown_reals = [
    "open_price", 
    "high_price", 
    "low_price", 
    "close_price", 
    "volume", 
    "quote_asset_volume", 
    "number_of_trades", 
    "taker_buy_base_asset_volume", 
    "taker_buy_quote_asset_volume"
]
# 
training = TimeSeriesDataSet(
    df[lambda x: x['time_idx'] <= training_cutoff],
    time_idx=time_idx,  
    
    target=target_labels, # For single Data Prediction
    
    group_ids=group_ids,
    time_varying_known_reals=time_varying_known_reals,
    static_reals=static_reals,
    time_varying_unknown_reals=time_varying_unknown_reals,
    
    min_encoder_length=max_encoder_length // 2,  # keep encoder length long (as it is in the validation set)
    max_encoder_length=max_encoder_length,
    max_prediction_length=max_prediction_length,

)

OUTPUT

Training Cutoff: 49993
Training Dataset Length: 49989 | Validation Dataset Length: 1
Train DataLoader Length: 1562 | Validation DataLoader Length: 1
Max time_idx: 49999, Training Cutoff: 49993
Data points in training: 49994
Data points in validation: 6

Additional context
It also does not work when I pass a string directly in the intialzer.

# create dataset and dataloaders
max_encoder_length = 36
max_prediction_length = 6
training_cutoff = df["time_idx"].max()  - max_prediction_length

print(f'Training Cutoff: {training_cutoff}')

# Target Labels
time_idx = 'time_idx'
target_labels = 'close_price'

group_ids = ['asset_id']

#static_reals='asset_id',

time_varying_known_reals = ["time_idx", "month", "day", "day_of_week", "hour", "minute"]


time_varying_unknown_reals = [
    "open_price", 
    "high_price", 
    "low_price", 
    "close_price", 
    "volume", 
    "quote_asset_volume", 
    "number_of_trades", 
    "taker_buy_base_asset_volume", 
    "taker_buy_quote_asset_volume"
]
# 
training = TimeSeriesDataSet(
    df[lambda x: x['time_idx'] <= training_cutoff],
    time_idx=time_idx,  
    
    target=target_labels, # For single Data Prediction
    
    group_ids=group_ids,
    time_varying_known_reals=time_varying_known_reals,
    static_reals='asset_id',
    time_varying_unknown_reals=time_varying_unknown_reals,
    
    min_encoder_length=max_encoder_length // 2,  # keep encoder length long (as it is in the validation set)
    max_encoder_length=max_encoder_length,
    max_prediction_length=max_prediction_length,

)

KeyError

KeyError                                  Traceback (most recent call last)
Cell In[56], [line 31](vscode-notebook-cell:?execution_count=56&line=31)
     [19](vscode-notebook-cell:?execution_count=56&line=19) time_varying_unknown_reals = [
     [20](vscode-notebook-cell:?execution_count=56&line=20)     "open_price", 
     [21](vscode-notebook-cell:?execution_count=56&line=21)     "high_price", 
   (...)
     [28](vscode-notebook-cell:?execution_count=56&line=28)     "taker_buy_quote_asset_volume"
     [29](vscode-notebook-cell:?execution_count=56&line=29) ]
     [30](vscode-notebook-cell:?execution_count=56&line=30) # 
---> [31](vscode-notebook-cell:?execution_count=56&line=31) training = TimeSeriesDataSet(
     [32](vscode-notebook-cell:?execution_count=56&line=32)     df[lambda x: x['time_idx'] <= training_cutoff],
     [33](vscode-notebook-cell:?execution_count=56&line=33)     time_idx=time_idx,  
     [34](vscode-notebook-cell:?execution_count=56&line=34)     
     [35](vscode-notebook-cell:?execution_count=56&line=35)     target=target_labels, # For single Data Prediction
     [36](vscode-notebook-cell:?execution_count=56&line=36)     
     [37](vscode-notebook-cell:?execution_count=56&line=37)     group_ids=group_ids,
     [38](vscode-notebook-cell:?execution_count=56&line=38)     time_varying_known_reals=time_varying_known_reals,
     [39](vscode-notebook-cell:?execution_count=56&line=39)     static_reals='asset_id',
     [40](vscode-notebook-cell:?execution_count=56&line=40)     time_varying_unknown_reals=time_varying_unknown_reals,
     [41](vscode-notebook-cell:?execution_count=56&line=41)     
     [42](vscode-notebook-cell:?execution_count=56&line=42)     min_encoder_length=max_encoder_length // 2,  # keep encoder length long (as it is in the validation set)
     [43](vscode-notebook-cell:?execution_count=56&line=43)     max_encoder_length=max_encoder_length,
     [44](vscode-notebook-cell:?execution_count=56&line=44)     max_prediction_length=max_prediction_length,
     [45](vscode-notebook-cell:?execution_count=56&line=45) 
...
-> [6249](https://file+.vscode-resource.vscode-cdn.net/Users/arjein/Documents/GitHub/AI/coin_pred/~/Documents/GitHub/AI/coin_pred/.venv/lib/python3.12/site-packages/pandas/core/indexes/base.py:6249)         raise KeyError(f"None of [{key}] are in the [{axis_name}]")
   [6251](https://file+.vscode-resource.vscode-cdn.net/Users/arjein/Documents/GitHub/AI/coin_pred/~/Documents/GitHub/AI/coin_pred/.venv/lib/python3.12/site-packages/pandas/core/indexes/base.py:6251)     not_found = list(ensure_index(key)[missing_mask.nonzero()[0]].unique())
   [6252](https://file+.vscode-resource.vscode-cdn.net/Users/arjein/Documents/GitHub/AI/coin_pred/~/Documents/GitHub/AI/coin_pred/.venv/lib/python3.12/site-packages/pandas/core/indexes/base.py:6252)     raise KeyError(f"{not_found} not in index")

KeyError: "None of [Index(['a'], dtype='object')] are in the [columns]"

and these are the columns:

df.columns

Index(['open_price', 'high_price', 'low_price', 'close_price', 'volume',
       'quote_asset_volume', 'number_of_trades', 'taker_buy_base_asset_volume',
       'taker_buy_quote_asset_volume', 'time_idx', 'asset_id', 'year', 'month',
       'day', 'day_of_week', 'hour', 'minute'],
      dtype='object')

Versions

System:
python: 3.12.8 (main, Dec 3 2024, 18:42:41) [Clang 16.0.0 (clang-1600.0.26.4)]
executable: /Users/arjein/Documents/GitHub/AI/coin_pred/.venv/bin/python
machine: macOS-15.1.1-arm64-arm-64bit

Python dependencies:
pip: 24.3.1
sktime: 0.35.0
sklearn: 1.5.2
skbase: 0.12.0
numpy: 2.1.3
scipy: 1.15.1
pandas: 2.2.3
matplotlib: 3.10.0
joblib: 1.4.2
numba: None
statsmodels: 0.14.4
pmdarima: None
statsforecast: None
tsfresh: None
tslearn: None
torch: 2.5.1
tensorflow: None

[Sohaib-Ahmed21]

@Arjein according to my understanding so far, for single static_reals , pass it as string and for multiple, pass it as a list as passed in this tutorial.