ColumnSpec#

class hybrid_aif.train.ColumnSpec[source]#

Bases: NamedTuple

One task-specific column of an Episode.

shape: tuple[int, ...]#

The column’s per-step trailing shape.

dtype: str#

The column’s dtype, by name.

pad: str#

How fit_length() pads the column: "zero" or "hold" (repeat the last value).

static __new__(_cls, shape=(), dtype='float32', pad='zero')#

Create new instance of ColumnSpec(shape, dtype, pad)