timecast package

Module contents

timecast: a library for online time series analysis

class timecast.Module(*args, **kwargs)[source]

Bases: object

Core module class

add_module(module, name=None)[source]

Add module outside attributes

add_param(param, name)[source]

Add parameter outside attributes

get_param_tree()[source]

Return recursed parameter tree

set_param_tree(tree)[source]

Apply parameter tree

class timecast.experiment(argnames: Union[List[str], str], arglists: List[Any])[source]

Bases: object

Class decorator to run experiments

run(processes=1, chunksize=1, tqdm=None)[source]

Execute the experiment

timecast.tscan(X: Union[numpy.ndarray, Tuple[numpy.ndarray, ...]], Y: Union[numpy.ndarray, Tuple[numpy.ndarray, ...]], optimizer: flax.optim.base.Optimizer, loss_fn: Callable[[numpy.ndarray, numpy.ndarray], numpy.ndarray] = <function <lambda>>, state: flax.nn.base.Collection = None, objective: Callable[[numpy.ndarray, numpy.ndarray, Callable[[numpy.ndarray, numpy.ndarray], numpy.ndarray], flax.nn.base.Model], Tuple[numpy.ndarray, numpy.ndarray]] = None)[source]

Take gradients steps performantly on one data item at a time

Parameters
  • X – np.ndarray or tuple of np.ndarray of inputs

  • Y – np.ndarray or tuple of np.ndarray of outputs

  • optimizer – initialized optimizer

  • loss_fn – loss function to compose where first arg is true value and

  • is pred (second) –

  • state – state required by flax

  • objective – function composing loss functions

Returns

result

Return type

np.ndarray