timeserio.keras.timing module

A set of tools for determining output latency (delay and stride).

For temporal NN models.

timeserio.keras.timing.absolute_latencies(relative_latencies)

Compute absolute delay + stride for a each in a stack of layers.

relative_latencies: list of tuples

each tuple is (rel_delay, rel_stride)

absolute_latencies: list of tuples

each tuple is (abs_stride, abs_stride)

timeserio.keras.timing.layer_latency(layer)

Get latency for a given keras layer.

layer: Layer

a Keras layer

delay: float stride: int

timeserio.keras.timing.model_latency(model)

Get latency for a given keras temporal model.

model: keras.Model

a Keras multi-layer temporal model

delay: float stride: int

timeserio.keras.timing.model_output_length(model, input_shape)

Get number of output timesteps for keras temporal model.

model: keras.Model

a Keras multi-layer temporal model

input_shape: tuple

shape of a single input

Return type

int

timeserio.keras.timing.total_latency(relative_latencies)

Compute absolute delay + stride for last layer in a stack.

relative_latencies: list of tuples

each tuple is (rel_delay, rel_stride)

delay: float stride: int