toolkits

DL Toolkits

?‍  Analytical tools for deep learning experiments  ?‍

 

Whenever I analyzed the results of the DL experiment, I had to re-implement the analysis function every time. So, I implement frequently used functions in this repository. New features continue to be implemented, and simple examples of function usages can be found in the examples directory.

Installation

You can install the package with pip command. Python>=3 are supported.

pip install dl-toolkits

You can check the version of the package using the following commands.

import toolkits
print(toolkits.__version__)

Modules

Visualization

• viz.tsne: t-SNE plot

Clustering quality

• cluster.sse: Sum of squared error(SSE) ¹
• cluster.batch_sse: Sum of squared error(SSE) for batch input
• cluster.nsse: SSE normalized by the squared distance to the nearest interfering centroid(nSSE) ¹
• cluster.batch_nsse: SSE normalized by the squared distance to the nearest interfering centroid(nSSE) for batch input
• cluster.nearc: Top N nearest interfering centroid
• cluster.rfc: Feature space clustering quality(R_fc) ²

Linear algebra

• linalg.get_singular_values: Get singular values per class
• linalg.get_sum_of_singular_values: Get sum of singular values per class
• linalg.get_average_sum_of_singular_values: Get average of sum of singular values per class ³

Pretty print

• pprint.pred_summary: Simple print for predictions and true labels

Log parser

• parse.between_lines: Extract the log between the two input sentences
• parse.between_lines_on_file: Extract the log between the two input sentences on the target file
• parse.between_lines_on_dir: Extract the log between the two input sentences on the target directory

PyTorch helper function

• torch_helper.freeze_selected_param: Freeze the weights with the selected name
• torch_helper.get_important_param_idx: Get important parameters indices⁴

References

Pedregosa, Fabian, et al. "Scikit-learn: Machine learning in Python." the Journal of machine Learning research 12 (2011): 2825-2830.

Footnotes

1. Yoon, Sung Whan, et al. "Xtarnet: Learning to extract task-adaptive representation for incremental few-shot learning." International Conference on Machine Learning. PMLR, 2020. ↩ ↩²

2. Goldblum, Micah, et al. "Unraveling meta-learning: Understanding feature representations for few-shot tasks." International Conference on Machine Learning. PMLR, 2020. ↩

3. Verma, Vikas, et al. "Manifold mixup: Better representations by interpolating hidden states." International Conference on Machine Learning. PMLR, 2019. ↩

4. Mazumder, Pratik, Pravendra Singh, and Piyush Rai. "Few-Shot Lifelong Learning." Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 35. No. 3. 2021. ↩