airpack.pytorch.model

Contains neural network models for signal classification implemented in the PyTorch framework.

When getting started with this package, you can use these models as-is: the default sizes and parameter values are reasonable for the bundled datasets. Next, you can explore how tuning the values provided as arguments to these models (“model hyperparameters”) affects their performance and accuracy. Finally, these models are each standalone functions, so you can copy them into your own code to add layers, change activation functions, and experiment with different model structures entirely!

Module Contents

class airpack.pytorch.model.default_network(input_len, output_len, num_filt=16, strides=(1, 1), filt_size=(32, 2), pool_size=(16, 1), fc_num_nodes=[128, 64, 32], padding=(0, 0))

Bases: torch.nn.Module

Creates a convolutional signal classifier model for chunks of signal data that are a specified number of samples long.

Required arguments are the input length (number of samples in the input) and the output length number of classes of signal in the classifier. Other optional arguments are tunable hyperparameters to optimize the model for specific datasets and/or types of signals. When the optional arguments are set to None, default values will be used that will work reasonably well.

This model object is created with default-initialized weights and must be first trained before being used for inference.

Parameters

• input_len – Number of complex samples to be input to the model

• output_len – Number of output classes for this classifier

• num_filt – Number of output filters in the convolution

• strides – Stride within convolutional layers

• filt_size – Kernel/window size of convolutional layers, typically (X, 2) for some X

• pool_size – Kernel/window size of downsampling layers, typically (X, 1) for some X

• fc_num_nodes – List of sizes of fully-connected (FC) layers

• padding – Padding dimensions

forward(self, x_real)

Feed forward of the neural network

Note

Data needs to be converted from complex tensor prior to this

Parameters

x_real (torch.float32) – non complex tensor

Returns

output

Return type

torch.Tensor

get_dummy_data(self, batch_size)

Get a tensor (containing random data) of the shape and data type expected by this model. This dummy data tensor is used to print a summary of the model as well as during tracing for ONNX model export and in test cases.

Note

You can always test that the model is sized correctly to pass data by running model.forward(model.get_dummy_data(batch_size))

Parameters

batch_size (int) – batch size

Returns

input_data

Return type

torch.Tensor

airpack.pytorch.model.calc_output_dims(h_in, w_in, c, kernel_size, stride=None, padding=None, dilation=None)

Calculate torch.nn.Conv2d or torch.nn.MaxPool2d output dimensions

Note

Ref: https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html https://pytorch.org/docs/stable/generated/torch.nn.MaxPool2d.html

Parameters

• h_in (int) – height dimensions of layer input

• w_in (int) – width dimension of layer input

• c (int) – channel dimension of layer output

• kernel_size (tuple) – kernel dimensions

• stride (Optional[tuple]) – stride dimensions

• padding (Optional[tuple]) – padding dimensions

• dilation (Optional[tuple]) – dilation dimensions

Returns

output dimensions of layer (C, H, W)

Return type

Tuple[int, int, int]

airpack.pytorch.model.MODEL