[TASK] Implement PredictNext, PredictLast & PredictRandom

[marcromeyn]

We need a bunch of augmentations to support session-based recsys. In the examples below we have a sequence: ABCDE that we augment in following ways:

Prediction-next

Inputs	A	B	C	D
Targets	B	C	D	E

class PredictNext(DataAugmentation):
	def __init__(
		self, 
		schema: Schema
		target: Union[str, Tag, ColSchema],
	):
		...

Predict-last

Inputs	A	B	C	D
Targets	E

class PredictLast(DataAugmentation):
	def __init__(
		self, 
		schema: Schema
		target: Union[str, Tag, ColSchema],
	):
		...

Predict-random

We grow the batch with extra samples where we sample a row_index, use the previous timesteps as features, the next as target.

Inputs	A	B	C	D
Targets	E
Inputs	A
Targets	B
Inputs	A	B
Targets	C

class PredictRandom(DataAugmentation):
	def __init__(
		self, 
		schema: Schema
		target: Union[str, Tag, ColSchema],
		# causal will control whether or not the elements in the sequence after the radomly picket element will be added to the inputs
		causal=True,
		# When this is bigger than 1, we will grow the batch with more samples
		num_per_batch=1
	):
		...