Beginning to learn a new subject can be intimidating. This is particularly true when that new subject is part of a science and technology field. This may make many of the beginners wonder what should they do, how should they proceed, and what tools they need. The primary tool for data science is -obviously- a computer; and if you are a new student or come from a substantially different background, there is a high probability you will get a little anxious about whether your old personal computer is fit for the task. If you’re wondering whether you must buy a new laptop for your data science course, you need to read this beforehand.
There’s no denying that training large machine learning models needs a lot of computational power. Particularly deep neural networks, which most of the time are run on monstrous GPUs. Industrial models may need workstations or supercomputers to run. However, most of us can’t don’t have supercomputers at home, and for some, purchasing a new upper-grade laptop may be a financial challenge. The good news is you don’t need to afford any of them to begin learning data science. Small datasets still are easy to work with, even with your old laptop, and Transfer learning lets you use pre-trained networks. You can use one of the multiple available free cloud platforms, or pay a more affordable fee to use premium cloud computing power. In the table below, we’ve compared these two options.
Why should I use cloud platforms?
Technically, cloud platforms may be better options than buying your own hardware. When using a cloud platform, your cloud projects are available anytime, anywhere with Internet access; while a powerful-enough PC or laptop is too heavy and has a limited battery, which makes it much less portable than your normal notebook. At the same time, however powerful your laptop is, it won’t be able to process really large datasets. For large projects, you will probably be forced to buy more and more expensive GPUs and other -also expensive- hardware parts that can work with them.
You will need to maintain your hardware yourself: take care of its environment’s temperature, clean it, and update it. You will always have to take care of your development environment: manage your libraries and packages, update them, and take care of any problems that arise when changing your settings. At the same time, buying hardware gives you a fully private development environment. You will own your data and can access it even when you’re offline. You can do any experiment, and have as many databases as you want.
Cloud vs. Desktop: which one is better for machine learning?
|Buying new hardware||Renting cloud services|
|More powerful laptops and PCs are heavier and have less battery life, so you don’t have much mobility.||Your cloud projects are available anytime, anywhere with internet access.|
|It works well with small and medium datasets but still isn’t enough for larger datasets.||You can rent as much computational power as you need, but you will be charged for any computation further than free service limits.|
|You must do hardware maintenance yourself, and take care of things like cooling and power. If your device gets damaged, you must buy new ones.||You don’t need to worry about hardware maintenance.|
|You must manage and update your development environment, including compilers and installed packages. In case of any inconsistencies, you must solve the problem yourself.||The platform and packages are updated by the service provider.|
|You have full control over your development environment, and do not need to worry about charges for extra hours or computation.||Your control over the work environment is limited. The platform will charge you for every extra computation.|
|You can have as many databases as you want without further charge.||In some cases -like Google Colaboratory- you don’t have databases. In other platforms, having several databases means getting charged for each one.|
|You can work offline.||No Internet means no access to your work environment.|
|Your devices will be obsolete in a few years.||Cloud environments are getting cheaper and more powerful every day.|
|Your program can run for as long as you want.||There may be some limitations on how long your programs can run without pause.|
It is possible to do data science on a budget. If you’re at the beginning of your path as a data scient, the only thing you need to worry about is learning. You don’t need to buy a high-end gaming laptop or expensive GPUs for educational projects. In the future, you may choose from multiple cloud options or buy your own equipment for the big things you will hopefully do.