Tutorial

Getting Started

Install the ‘bom_analyzer’ package from PYPI using:

pip install bom_analyzer

After installing, import the library into your file using:

import bom_analyzer as ba

Using the Library

Note

The input dataset must be a csv with the following attributes:

SERNUM	Column denoting the serial number of the product
HWRMA	Column denoting whether a product has a known error with value True or False
CPN_#	Every component must have an associated CPN number

To begin, call run_sentence_transform with the path of the product dataset. To use hardware acceleration, make sure to specify the “device” attribute. It is also recommended, that an archive path is specified to store the output.

embeddings = ba.run_sentence_transform('dataset.csv', device='cuda', archive_path='embeddings.npy')

Next, with the output embeddings, call run_optimizer to determine the hyperparameters. that groups the data most effectively (based on DBCV score). The number of trials, random seed, and a location to archive the output can all be specified as shown below.

params = ba.run_optimizer(embeddings, seed=42, trials=50, archive_path='params.json')

Now that you have the embeddings and the ideal hyperparameters, call run_dimension_reduction to convert the embeddings into two-dimensional data.

This function accepts a path to, or dataframe of the original dataset and appends the result in the columns DATA_X and DATA_Y.

table = ba.run_dimension_reduction(table, embeddings, params, seed=42, archive_path='dataset.csv')

With the dimension reduced data now accessible, pass the updated table into run_clustering to get the cluster labels associated with each product.

As before, this function accepts a path to, or dataframe of the original dataset with DATA_X and DATA_Y already defined. It then appends the result in the column CLUSTERS.

table = ba.run_clustering(table, params, archive_path='dataset.csv')

Now, we can determine the density of errors in these groups with the label_outliers function.

As before, this function accepts a path to, or dataframe of the original dataset with DATA_X, DATA_Y, and CLUSTERS already defined. It then appends the result in the column OUTLIER_DENSITY.

table = ba.label_outliers(table, archive_path='dataset.csv')

Last, we can find the components associated with these errors using the report_suspect_components function.

As before, this function accepts a path to, or dataframe of the original dataset with DATA_X, DATA_Y, CLUSTERS, and OUTLIER_DENSITY already defined.

The function returns a dataframe of components unique to the ‘n’ clusters with the highest density of errors. This dataframe will contain columns CPN, DateCode, LOTCODE, MPN, and RD.

components = ba.report_suspect_components(table, num_clusters=n, archive_path='components.csv')