Hands-On Deep Learning for IoT

Unsupervised deep feature synthesis

We will see how the entity set structures can contribute to improve the predictive accuracy. We will build custom primitives using time-series functions from the tsfresh library. Before that, we will make cut-off times by selecting a random one from the life of each engine. We are going to make five sets of cut-off times to use for cross-validation:

from tqdm import tqdmsplits = 10cutoff_time_list = []for i in tqdm(range(splits)):    cutoff_time_list.append(utils.make_cutoff_times(data))cutoff_time_list[0].head()

The preceding code block should show the cut-off time and the RUL values for five engines, as shown here:

Then, we will use an unsupervised way of generating the entity set. As we can see, the values ...