contribs.torchsummary.summary¶

Function · Source

summary_str, params_info = mdnc.contribs.torchsummary.summary_str(
    model, input_size, batch_size=-1, device='cuda:0', dtypes=None
)

Iterate the whole pytorch model and summarize the infomation as a Keras-style text report. The output would be store in a str.

Arguments¶

Requries

Argument	Type	Description
`model`	`nn.Module`	The pyTorch network module instance. It is to be analyzed.
`input_size`	`(seq/int, )`	A sequence (`list/tuple`) or a sequence of sequnces, indicating the size of the each model input variable.
`batch_size`	`int`	The batch size used for testing and displaying the results.
`device`	`str` or `torch.device`	Should be set according to the deployed device of the argument `model`.
`dtypes`	`(torch.dtype, )`	A sequence of torch data type for each input variable. If set `None`, would use float type for all variables.

Returns

Argument	Description
`summary_str`	The summary text report.
`params_info`	A `tuple` of two values. The first value is the total parameter numbers. The second value is the trainable parameter numbers.

Tip

This function could be used for generating the text log file:

...
with open('my_module.log', 'w') as f:
    report, _ = mdnc.contribs.torchsummary.summary_string(model, ...)
    f.write(report)

Last update: March 14, 2021