Word Embedding for predictive predictive maintenance in industry based on historic daily reports, this word embedding use the GloVe algorithm, the Visualization of the Learnings of the GloVe Algorithm are made with TensorBoard and standar vis techniques. The information about how to run this is into the NLP.ipynb.
Each word is influenced by its neighbors and this influence reflects the degree of correlation that exists between them, for example the word "falla" will be surrounded by words that could cause a failure of some kind.
Example of PCA with tensorboard output of the GloVe embedding
The goal of this project is to create a word map that describes the influence of different aspects in an industrial installation whether motor, boards, electric machines, protection systems, etc. described in the daily reports by the workers of a factory X over the years. This is done using a GloVe word embedding in this daily reports data.
Fitting embeddings to T-SNE and a random sample of it.
This project and his data set is released under the Apache 2.0 license.
- Python 3.5
- TensorFlow >= 1.4
- NLTK
- This is a work in progres
- The data was extracted from historic excel tables as a process of data mining, code will be added soon.
- Trained word embedding and model checkpoints for tensorboard visualization will be added soon.