Advanced Features

Temporal Enrichment

Add temporal features automatically:

d1_dataset = MultiSourceTSDataSet(
    file_paths=['data.csv'],
    target_cols=['value'],
    enrich_cat=['hour', 'dow', 'month', 'minute']
)

Available temporal features:

• hour: Hour of day (0-23)

• dow: Day of week (0-6)

• month: Month (1-12)

• minute: Minute (0-59)

Global vs Local Forecasting

Local Forecasting (default):

Each time series is treated independently.

Global Forecasting:

Train a single model across multiple time series.

d1_dataset = MultiSourceTSDataSet(
    file_paths=['group_a.csv', 'group_b.csv'],
    group_cols=['group_id'],
    global_forecasting=True
)

Custom Data Splits

Split by groups:

d2_dataset = EncoderDecoder(
    d1_dataset=d1_dataset,
    split_ratio=(0.7, 0.15, 0.15),
    split_group_config=([0, 1], [2])  # Train on groups 0,1; test on 2
)

Scaling Strategies

DSIPTS supports multiple scaling methods:

d2_dataset = EncoderDecoder(
    d1_dataset=d1_dataset,
    scaling_method='standard',  # or 'minmax'
    target_scaling_method='standard'  # Scale targets separately
)

Memory Efficiency

For large datasets:

d1_dataset = MultiSourceTSDataSet(
    file_paths=['large_data.csv'],
    memory_efficient=True  # Load data on-demand
)

d2_dataset = EncoderDecoder(
    d1_dataset=d1_dataset,
    memory_efficient=True  # Transform on-the-fly
)

Quantile Forecasting

Predict multiple quantiles:

model = ITransformer(
    out_channels=3,
    past_steps=96,
    future_steps=24,
    quantiles=[0.1, 0.5, 0.9]
)

Custom Models

Extend the base model class:

from dsipts.models.base import BaseModel

class MyCustomModel(BaseModel):
    def __init__(self, **kwargs):
        super().__init__(**kwargs)
        # Your architecture here

    def forward(self, batch):
        # Your forward pass
        pass