elasticai.creator.nn.fixed_point.conv1d.testbench

Module Contents

Classes

Conv1dDesignProtocol
Conv1dTestbench

API

class elasticai.creator.nn.fixed_point.conv1d.testbench.Conv1dDesignProtocol

Bases: typing.Protocol

abstract property name: str

abstract property input_signal_length: int

abstract property port: elasticai.creator.vhdl.design.ports.Port

abstract property kernel_size: int

abstract property in_channels: int

abstract property out_channels: int

class elasticai.creator.nn.fixed_point.conv1d.testbench.Conv1dTestbench(name: str, uut: elasticai.creator.nn.fixed_point.conv1d.testbench.Conv1dDesignProtocol, fxp_params: elasticai.creator.nn.fixed_point.number_converter.FXPParams)

Bases: elasticai.creator.vhdl.simulated_layer.Testbench

Initialization

save_to(destination: elasticai.creator.file_generation.savable.Path)

property name: str

prepare_inputs(*inputs) → list[dict]

parse_reported_content(content: list[str]) → list[list[list[float]]]

This function parses the reported content, which is just a list of strings. All lines starting with ‘output_text:’ are considered as a result of the testbench. These results will be stacked for each batch. So you get a list[list[list[float]]] which is similar to batch[out channels[output neurons[float]]]. For each item reported it is checked if the string starts with ‘result: ‘. If so the remaining part will be split by ‘,’. The first number gives the batch. The second the result. The channels are greedy guessed. We do so by taking the first X-values, where X is the number of values per channel. After we have enough values for the channel, we increase the channel number. If you report not enough values per channel, this will look like the last channel has not reported enough values.