Table of Contents
The process of creating a machine learning model is complex and requires more than just theoretical knowledge; it also calls for a practical grasp of machine learning tools and fundamental mathematics. An outline of the essential instruments and mathematical ideas you’ll encounter when creating machine learning models is given in this article.
Consider it an early look at what lies ahead: as we move forward on this trip, we will delve deeper into these subjects, particularly when we begin to examine the roadmap and learn machine learning step-by-step. If you have not read the previous three articles – “What is machine learning”, “Machine learning problems”, “Machine learning process” I highly encourage you read them first to get the whole idea of this one.
In order for you to choose the best resources and methods for your projects, it is intended that you become familiar with the world of machine learning tools and mathematics. Together, we will begin to investigate the environment that underpins contemporary machine learning.
Libraries and Code space
NumPy
Numerical Python or NumPy is a main library for numerical computing in python. This is like a foundation tool in machine learning which will be used by more advance libraries below.
Pandas
Pandas is also based on python and pandas is used for data analysis and data manipulation and handle structured data efficiently by providing Data Frames.
Matplotlib
Matplotlib is one of the most widely used tool in the machine learning eco system and it helps to visualize datasets while providing insights into patterns and trends before choosing machine learning model.
Jupyter Notebook
Jupyter Notebook is an open-source web application that allows users to create and share documents containing live code, visualizations, and text in machine learning eco system.
Scikit-Learn
Scikit-Learn is an open-source python based extensive machine learning library with features for preprocessing data, modeling data and evaluating models. Can be used for classification, regression, dimensionality reduction, clustering and many more.
PyTorch
Pytorch is an opensource deep learning library with capabilities for preprocessing data, modeling data and serving models. It is also used for model changes and debugging in real-time. Pytorch lightning is a simplified but a powerful version of PyTorch
TensorFlow
TensorFlow is an open source deep learning ML framework with capabilities from server to embedded devices. TensorFlow.js is used for deep learning in the web, TensorFlow Lite is used for on-device inference for mobile and IOT. Keras is a high-level API now integrated into TensorFlow 2.0
ONNX
Open Neural Network Exchange designed to have interoperability between different neural network frameworks. Both python based and C++ based versions available. For example, you can export PyTorch model and run it on the same hardware used to run a TensorFlow model.
Pretrained models
These tools accelerate the deployment and mostly used for transfer learning.
- TensorFlow Hub – Offers pertained models for tasks such as classification, embedding and more.
- Pytorch Hub – This simplifies transfer learning and commonly used in natural language processing (NLP) and computer vision
- HuggingFace Transformers – Leader of NLP offers the services for models like GPT, BERT and T5.
- Detectron2 – A segmentation from facebook.AI and offers pre-trained models for computer vision
Tools for Experiment Tracking
These tools help to compare and analyze the experiments done in machine learning
TensorBoard
Tool to track and visualize metrics, view model graphs, look at images, text and audio data. And TensorBoard is integrated with TensorFlow and PyTorch.
Dashboard
This tool is provided by Weights & Biases and this is a great platform for tracking the machine learning experiments by using minimum code lines.
neptune.ai
This tool is used to log the experiments done in machine learning, experiment the model performance, data versions, notebooks changes and more
Data and model tracking tools
These tools help you to understand what changes we have made to the data and helps us in finding how did they effect the models.
Artefacts
This is another great tool provided by Weights and Biases. Artefacts helps you to identify version of datasets, track your different machine learning pipelines reproduce your previous datasets.
DVC
Data Version Control is an open source version control just like Git specifically used to track the models and data sets of machine learning.
Cloud compute services
To run these machine learning models, we need a high computer power sometimes it costs us over 1000$. Because of this we use scalable online cloud resources for training and deploying machine learning models.
- Google Colab – This is a free jupyternotebook environment supports GPU and recommend for beginners who starts to deploy machine learning models
- AWS SageMaker – Provided by Amazon web services
- AI platform – provided by google cloud platform (GCP)
- Azure ML – provided by Microsoft Azure cloud services
If the developers want to run their models from local environment with high performance they must use a high GPU machine which accelerates deep learning and need to built custom deep learning PC s for ML workloads.
AutoML and hyperparameter tuning tools
AutoML tools helps building machine learning models automatically based on your dataset and hyperparameter tuning tools helps to find the best parameters and tools.
- TPot – Python automated machine learning tool that optimizes machine learning pipelines using genetic programming.
- Google Cloud AutoML – works phenomenal, downside is G (usually) can’t download the model, so you need to run API calls to Google for inference.
- Microsoft Automated Machine Learning
- Sweeps by Weights & Biases – trial and track a range of hyperparameter experiments and see which ones work best.
- Keras Tuner- Hyperparameter tuning library for Keras models.
Explainability Tools
These tools help users to understand why the model did this or that making it more transparent and ensure fairness
- What-if tool – compare different machine learning models? visualize inference results, change data points and see how the model reacts. 00
- SHAP values – use game theory to explain the outputs of your machine learning models.
Machine Learning Lifecycle Tools
These technologies facilitate the development of user interfaces and oversee the machine learning lifecycle, from data collection to the deployment of a machine learning model.
Streamlit – create stunning data-driven user interfaces with so you can showcase your work.
MLflow – an open-source platform that tracks experiments, packages code, deploys, and stores models.
Kubernetes – platform for creating contained applications, and Kubeflow offers a framework for implementing machine learning processes on it.
Seldon – serves as a link between machine learning and DevOps, which is the process of ensuring that a software application is deployed. builds an MLOps framework.
Machine learning Mathematics
Mathematics is the backbone of machine learning. It is so true to say that without proper knowledge about mathematics machine learning can be difficult to understand. Therefore, there some key aspects that you needed to know about mathematics before jumping into the roadmap and start learning. We will now discuss the basic idea of each aspect and find some valuable resources to learn mathematics in the next article.
Linear algebra
Creating objects and a set of rules to manipulate these objects). E.g. x^2, x is the object, and ^2 is manipulating that object. Machine learning is about finding the right set of objects and right set of rules to model a dataset.
Matrix manipulation
In machine learning data (all kinds of it) often gets turned into rows, columns and features (features is the 3rd dimension, which can actually be many dimensions) of numbers. These collections of numbers are often referred to as matrices or tensors.
Multivariate calculus
Foundation for optimizing a function (for example cost function) with respect to multiple parameters (the patterns a machine learning model learns)
Probability and Probability distribution
Probability means the study of uncertainty. A collection of probabilities, including, a sample space, a series of possible events, the probability of an event and the (unpredictable) random event.
Discrete Mathematics
Discreet Mathematics focuses on insights into data structures and algorithms used in machine learning
Optimization
If machine learning is about finding the most ideal patterns which describe a dataset, how do you co optimize a model to do so?
The Chain Rule
Basis of backpropagation, how neural networks improve themselves.
Conclusion
Gaining an understanding of mathematics and machine learning tools is similar to mastering a car’s mechanics before operating a car. Even if you don’t have to know everything at once, having a solid foundation aids in problem-solving and decision-making. We’ll go over these tools and ideas again in more detail as we go along, connecting them to real-world machine learning applications.
The most important component of the puzzle, Machine Learning Resources and the Roadmap will be our final article in this series of exploring machine learning field. From introductory courses to more complex subjects, we’ll go over how to methodically learn machine learning so that you have a clear plan to follow. Here is where theory, tools, and application really come together, so stay tuned!
