This document describes the structure and configuration of the YAML file that defines the search space for supported layers and their parameters. To extend the search space, follow the instructions outlined in the README.md file.
✅ The YAML configuration defines the available layers and their parameters.
✅ Parameters using lists follow the [start, stop, step] format.
✅ Rule sets ensure valid model architectures.
✅ Extend the search space by adding layers in layer_definitions.py and updating the YAML.
To add a new layer to the search space:
- Implement the layer in
neural_architecture_search/src/layer_definitions.py. - If needed, place custom operations in
neural_architecture_search/src/search_space_modules/. - Add the layer configuration to the
gene_poolsection of the YAML file. - Update the
rule_setto define layer connectivity.
The gene_pool section defines available layers and their parameters for feature extraction, global pooling, preprocessing, and dense layers. All list-based parameters follow the [start, stop, step] format.
| Layer Type | Function Name | Filters/Kernel/Pool Size | Strides | Padding | Activation |
|---|---|---|---|---|---|
C_1D |
Conv1D |
Filters: [4, 32, 1], Kernel: [1, 5, 1] | [1, 2, 1] | same |
relu |
DC_1D |
DepthwiseConv1D |
Kernel: [1, 5, 1] | [1, 2, 1] | same |
relu |
| Layer Type | Function Name | Filters/Kernel/Pool Size | Strides | Padding | Activation |
|---|---|---|---|---|---|
C_2D |
Conv2D |
Filters: [4, 32, 1], Kernel: [1, 5, 1] | [1, 2, 1] | same |
relu |
DC_2D |
DepthwiseConv2D |
Kernel: [1, 5, 1] | [1, 2, 1] | same |
relu |
| Layer Type | Function Name |
|---|---|
GAP_1D |
GlobalAveragePooling1D() |
GMP_1D |
GlobalMaxPooling1D() |
| Layer Type | Function Name | Parameters |
|---|---|---|
STFT_2D |
STFT |
n_fft: [64, 1024, 16], hop_length: [64, 396, 16] |
| Layer Type | Function Name | Parameters |
|---|---|---|
D |
Dense |
Units: [8, 128, 8], Activation: [relu] |
The rule_set defines valid layer connectivity.
| Layer | Next Possible Layers |
|---|---|
C_1D |
AP_1D, MP_1D, BN_1D, DC_1D, C_1D, STFT_2D |
| Layer | Next Possible Layers |
|---|---|
STFT_2D |
MAG_2D |