Chapter 6 – Don't Get Lost in Techniques, Focus on Optimizing Your Solutions
1. Can a prototype be built with random data in corporate environments? (Yes | No)
The answer is yes and no. To start developing a prototype, using random data can help make sure that the basic algorithm works as planned.
However, once the prototype is advanced, it will be more reliable to use a well-designed dataset. Then, once the training has been successfully accomplished, random data can help again to see how your system behaves in all situations.
2. Do design matrices contain one example per matrix? (Yes | No)
The answer is no. A good design matrix contains one example in each row or each column depending on the shape you want it to have. But be careful; a design matrix that contains data that is too efficient might overfit. That means the learning algorithm will be efficient with that data but not adapt to new data. On the other hand, if the dataset contains too many errors, then the algorithm might underfit, meaning it won't learn correctly. A good design matrix should contain reliable data, some imprecise data, and some noise (some data that can influence the algorithm in unreliable ways).
3. Automatic Guided Vehicle will never be widespread. (Yes | No)
The answer is that AGVs will expand endlessly from now on: drones, cars, planes, warehouse vehicles, industrial vehicles, and more. AGVs, added to artificial intelligence and IoT, constitute the fourth industrial revolution.
4. Can k-means clustering be applied to drone traffic? (Yes | No)
The answer is yes. Seeing where traffic builds up will prevent drone jams (drones circling and waiting).
5. Can k-means clustering be applied to forecasting? (Yes | No)
The answer is yes and will be explained in Chapter 10, Applying Biomimicking to Artificial Intelligence; it'll be applied to all types of traffic.
6. Lloyd's algorithm is a two-step approach. (Yes | No)
Yes, Lloyd's algorithm first classifies each data point in the best cluster. Then, once that is done, it calculates the geometric center or centroid of that center. When no data point changes the cluster anymore, the algorithm has been trained.
7. Do hyperparameters control the behavior of the algorithm? (Yes | No)
The answer is yes. Hyperparameters determine the course of the computation: the number of clusters, features, batch sizes, and more.
8. Once a program works, the way it is presented does not matter. (Yes | No)
The answer is no. Without a clear presentation of the results, the whole training process is confusing at best and useless at worst.
9. K-means is only a classification algorithm. It's not a prediction algorithm. (Yes | No)
The answer is no. K-means clustering can be used as a prediction algorithm as well.