Why Data Scientists Should Learn Swift
shared by pshravani727 on Feb 28
0
views
0
faves
0
comments
I've learned tidbits of about a dozen languages this way in the past ten years. I never received formal instruction in software engineering except for a C++-taught introductory computer science cou...
rse and a Java-based database course in graduate school. Coding allowed me to complete my assignments, examine data to find an answer to a topic, or materialize an idea in my thoughts.
This occasionally required me to become familiar with the specifics of algorithms or data structures, but I never found myself coding just for the purpose of it. I don't have a position on generics. I believe most data scientists and machine learning engineers fit this description. When selecting tools, usability, and efficiency are frequently prioritized over software foundations in the context of the issues we hope to tackle.
In 2018, the machine learning and data science communities appeared to have chosen Python. In addition to being a fantastic scripting language and having simple syntax, C also allows you to interact with lower-level libraries to increase performance.
The fact that Python is a "good enough" language for creating comprehensive systems from top to bottom is what appeals to me the most about it. Community support exists for scientific computing products like Numpy, Pandas, Matplotlib, and Jupyter notebooks. Yet when the time comes to create an application based on your work, frameworks like Flask and Django can scale to hundreds of millions of people. Using just one programming language, I am able to create an entire system.
Want to build Python projects for data science? Learnbay has the best data science courses, where you can develop multiple domain-specific projects with the help of industry experts.
I've used Python with satisfaction for almost a decade. I don't believe I will continue to use it for ten years. I will probably use Swift.
Google's Chris Lattner revealed that TensorFlow would soon support Swift during the 2018 TensorFlow Dev Conference. TensorFlow is not the same as Swift, which is just a basic iOS wrapper for TensorFlow. Much more is involved. This project aims to alter the industry-wide ecosystem for machine learning and data science's default tools.
Two more technology trends were slowly brewing as I studied the Python scientific stack: the comeback of AI through neural networks and deep learning and the transition towards mobile-first applications that run on billions of smartphones and IoT devices. Python is only somewhat fit for the high-performance computing needs of either of these technologies.
Deep learning requires running large data sets through numerous long chains of tensor operations, which is computationally expensive. Software targeting specialized processors with thousands of threads and cores must be developed to carry out these calculations rapidly. These issues are amplified in the setting of mobile devices, where power and heat are serious considerations. Application optimization for slower, more effective CPUs with less memory is complicated. Python hasn't provided much assistance so far.
This presents a minor issue for academics studying machine learning and data scientists. We frequently struggle with mobile app development and must revert to hack-like methods to interface with GPUs. The costs of switching languages are substantial, yet it is not impossible. Go no further than JavaScript initiatives like Node.js and cross-platform abstractions like React Native to see just how high.
I had intended to hold on to my numpy arrays in perpetuity, but it's becoming increasingly difficult to finish tasks within the Python ecosystem. It's no longer an adequate fix. Swift for TensorFlow was developed in response to Python's limitations as an end-to-end language in a world driven by machine learning and edge computing.
Note: Do check out the trending data science course online, to familiarize yourself with the newest data science technologies.
Chris Lattner argues that Python, with its dynamic typing and interpreter, cannot advance our cause. He claimed that engineers require a language that views machine learning as a "first-class citizen." However, the most convincing parts of his case center on the experience of the programmers themselves, even while he lays out several highly technical reasons why a new method of compiler analysis is required to alter how TensorFlow programs are created and run.
Any wish list of characteristics a programming language could have to make it simpler to *do* machine learning would consist of the following:
Understandable and effective syntax
Skills with scripts
UIs reminiscent of notebooks
Third-party libraries with a sizable and vibrant community, a tidy, automated method of assembling code for specialized hardware, such as TPUs or mobile chips
Mobile native execution
Performance more similar to Using Swift for TensorFlow, C Lattner and his team check each box. The syntax is nearly as attractive as Python. It includes a scripting and notebook interpreter. It's quick, too. Additionally, since Swift is now the standard for developing iOS apps, deploying to mobile is made more accessible by the ability to run any Python code, which is provided to help with easy migration.
Swift's internal static typing and open-source compiler enable build-time targeting of specific AI chipsets. Lattner may be biased because he was one of the original developers of Swift, but he has persuaded me that he is knowledgeable about the machine learning procedure. Visit Learnbay offers the best data science courses in India, for aspiring professionals in different domains.
Source
https://www....s-in-indiaCategory
EducationGet a Quote
