concept cleaning in category machine learning

This is an excerpt from Manning's book Machine Learning with TensorFlow, Second Edition.
Now that you can predict the sentiment of unseen text with your logistic-regression classifier, a good way to measure its general effectiveness is to test it out at scale, on lots of unseen text. You used 25,000 IMDB movie reviews to train it, so you’ll use 25,000 of those reviews that you held out to test it on. When you trained your classifier using Kaggle’s TSV file, you were using a joined version of the raw IMDB review data. You don’t always have that benefit, and sometimes you need to grab the actual raw data and preprocess it. To make sure you can handle both approaches for data preparation and cleaning, you’ll use the original raw aclImdb_v1.tar.gz file from https://www.dropbox.com/s/3ahq4z9376xu7yc/aclImdb_v1.tar.gz?dl=0 and prepare it for testing.
Unfortunately, the dataset isn’t directly usable in the deep speech model because the model expects Windows Audio Video interleaved (.wav) file audio format instead of the Free Lossless Audio Codec (.flac) file format that LibriSpeech comes in. So as usual, your first step for machine learning is going to involve—you guessed it—Time for some data preparation and cleaning!