model.predict() should work with a df with just k rows where k = encoder_length

[pchalasani]

• PyTorch-Forecasting version: 0.10.1
• PyTorch version: 1.11.1
• Python version: 3.8.10
• Operating System: MacOS

Expected behavior

My TimeSeriesDataSet was created via

   dataset = TimeSeriesDataSet(
        df_all,
        group_ids = ['symbol'],
        target = 'return', # predict 5-day forward return
        time_idx='time_idx',
        min_encoder_length=5,
        max_encoder_length=5,
        min_prediction_length=5,
        max_prediction_length=5,
        time_varying_unknown_reals=['return'],
        scalers=[]
    )

My understanding is that this means the model will use the past 5 time steps and predict the next 5 steps. So if we pass in a dataframe (with just one group_id) with 5 rows, predict() should work. But it gives an error, which goes away if I pass in a df with 10 rows (the sum of encoder_len + prediction_len).