Applications of AutoML
What is AutoML?
The full-form of AutoML is Automated Machine Learning. With AutoML, we can automate the machine learning tasks and reduce human intervention, to lessen human efforts.
A Machine Learning pipeline consists of several steps, from Data Acquisition to Making Predictions on a new dataset.
Figure: ML Pipeline
From this complete pipeline, we can automate the two steps Model Training / Model Selection and Hyper-parameter optimization by using AutoML.
Today, many libraries are available to implement Automated Machine Learning. According to the use-case, the user can choose the library that best suits their needs.
Let’s see some of the widely used AutoML libraries.
Auto-sklearn is designed specifically for Machine Learning models. Auto-sklearn works smoothly with all the sklearn libraries. It automatically selects the best performing algorithm for a dataset and even changes the hyperparameters to give the best results.
Auto-sklearn library is easy to install with just one statement.
pip install auto-sklearn
The two primary classes in the Auto-sklearn module are AutoSklearnClassifier and AutoSklearnRegressor; we use one for the Classification based Machine Learning tasks while the other for generating numerical predictions.
The official documentation of Auto-Sklearn can be found here.
The motive behind the development of AutoKeras was to make computer vision tasks simpler such that it is accessible to everyone. We use AutoKeras for Deep-Learning based tasks. Neural Networks and computer vision tasks are more complex, so we use AuoKeras to make it easier to implement.
AutoKeras can generate the best neural network architectures based on the task. In Neural Networks, selecting the number of neurons, the number of hidden layers, or setting the parameters is tedious and takes up huge amounts of human effort. Fortunately, AutoKeras considers all the possibilities and outputs the best model for the dataset.
We can easily install the AutoKeras library in our python environment.
pip install autokeras
Here is a small demonstration of importing AutoKeras in the python script and using it for Image Classification.
import autokeras as autok classifier = autok.ImageClassifier() classifier.fit(train_data, train_labels) result = classifier.predict(test_data)
To read the official documentation of AutoKeras, visit here.
We use TPOT for Automating certain stages in the Machine Learning pipeline. If data preparation is the most critical objective in our task, we can use TPOT as the data preparation, and modeling algorithms stages are emphasized more with this library.
The stages that are automated with TPOT are:
- Feature Selection,
- Feature Processing,
- Feature Construction,
- Model Selection, and
- Parameter Optimization
Like most AutoML libraries, TPOT can take hours to days to produce good quality results.
We can install TPOT using the pip command.
pip install tpot
Figure: TPOT Pipeline
Here is a simple illustration to use TPOT to classify digits in the sklearn digits dataset.
from tpot import TPOTClassifier from sklearn.datasets import load_digits from sklearn.model_selection import train_test_split digits = load_digits() train_data, test_data, train_labels, test_labels = train_test_split(digits.data, digits.target, train_size=0.75, test_size=0.25) classifier = TPOTClassifier(generations=5, population_size=50, verbosity=2, n_jobs=-1) classifier.fit(train_data, train_labels) classifier.score(test_data, test_labels)
Google Cloud AutoML
Google Cloud offers its own paid AutoML solution. VertexAI in the Google Cloud platform combines AI platform and AutoML. In GCloud, we use the AI Platform to run jupyter notebooks and train custom models (after all the preprocessing is done).
Combining the AI Platform with AutoML does wonders because now we can use AutoML to select the best model & hyperparameters, and we can make some tweaks to the model using the AI Platform according to our needs.
VertexAI has pre-trained APIs for computer vision, natural language processing, translation, and speech recognition applications. So, even a beginner who doesn’t have much knowledge of Deep Learning can develop good models. Labeling the dataset is another problem in Deep Learning which requires a lot of time and human effort; we can even use VertexAI to label the unlabelled dataset.
What is the score threshold in VertexAI?
Suppose we have some multi-class classification-based tasks. To train the model more effectively, the VertexAI will select some random image and pass it through the model; the model will output the probabilities of that image belonging to the respective classes. For instance, if the probability is highest for class A, the image belongs to class A.
The score threshold is the class’s probability of getting classified. For example, suppose our model has classified an 0.85 probability that an image belongs to class A. If the score-threshold is <0.85, we’ll classify it as class A whereas if the score-threshold is >0.85; we’ll not classify the image as class A.
Frequently Asked Questions
Q1. What are the disadvantages of using AutoML?
Ans. One of the most significant disadvantages of AutoML is that it takes a lot of time to train the models; high computational resources and power is required to train a model on large datasets. Apart from that, the AutoML is still in the development stage, so there is a high-risk factor with it while using AutoML in production applications.
Q2. How does VertexAI evaluate the performance of a model?
Ans. While evaluating the model, the VertexAI takes many parameters into consideration, such as:
- Model’s Output
- Classification Report (True Positives, True Negatives, False Positives & False Negatives)
- Precision and Recall of the Model
- The score threshold
- Precision and Recall Curves
Q3. What are libraries that we can use to implement AutoML in projects?
Ans. Nowadays, there are many AutoML solutions available online. We can choose any framework based on the scale of the project and use case. Some of the widely used AutoML frameworks are:
In today’s scenario, more & more industries are adapting to AutoML applications in their products; with this rise, it has become clear that AutoML can be the next boon in the technology.
Check out this link if you are a Machine Learning enthusiast or want to brush up your knowledge with ML blogs.
If you are preparing for the upcoming Campus Placements, don’t worry. Coding Ninjas has your back. Visit this link for a carefully crafted and designed course on-campus placements and interview preparation.