Melanoma Prediction Using CNN Methods. Part 5
Jupiter notebook cloud environments
Jupiter notebook (jupyter.org, 2022) is one of the main instruments for machine learning algorithms developing and testing. Also, Jupiter notebook is one of the most popular development environments for deep machine learning architectures training and validation.
Jupiter notebook can be installed locally, but deep learning architectures requires a big volume of RAM and fast and CPU, sometimes GPU or TPU. That is why, I am going to use cloud platforms for my CNN training. There are deep learning platforms as Google Colab (colab.research.google.com, 2022), Kaggle (kaggle.com, 2022) and Gradient (Gradient.run, 2022). Moreover, Kaggle and Google Colab are products of Google. The papers provide comparison of deep learning cloud providers by available CPUs, GPUs, prices, cost to train (Hale, J., 2018).
Understanding the cases of using GPUs in deep learning helps to choose when to use CPU or GPU. (Dettmers, T., 2020) helps with analysis of making “cost-effective choice”, “how to think about deep learning performance”.
Google Colab platform provides free CPU, but Google Colab can be used only for prototyping, training on small datasets. However, Google Colab platform provides different subscriptions (colab.research.google.com subscription, 2022) with different processors and amount of memory.
The paper (Tan, E., 2022) provides comparison of Gradient, Colab and Kaggle platforms. With reference to the paper (Tan, E., 2022), Colab suggests the max execution time per session. Also, with reference to (Tan, E., 2022), private notebooks are in Colab and Kaggle platforms.
Moreover, I compare Google Colab and Google Colab Pro. Regarding (Homes, J., 2022), Colab Pro provides faster GPUs, more memory, longer runtimes. At the same time, Colab Pro is not free, but I can use it by buying subscription.
The paper (Hale, J., 2019) compares Kaggle cloud environment and Google Colab. Regarding the research in the paper (Hale., J, 2019), Google Colab provides more flexibility in settings batch sizes. At the same time, both of environments are suitable for deep learning (Hale, J., 2019).
However, Google Colab is not available around the World, only users is some countries as Unites States, Canada, Japan, Brazil, Germany, France, India, United Kingdom, and Thailand can use Colab Pro (Droste, B., 2021).
References
jupyter.org, 2022, Jupiter. Available at: https://docs.jupyter.org/en/latest/index.html
[Accessed 09 July 2022]
kaggle.com, 2022, Kaggle. Available at: https://www.kaggle.com/
[Accessed 09 July 2022]
Gradient.run, 2022, Gradient. Available at: https://gradient.run/
[Accessed 09 July 2022]
Hale, J., 2018, List of Deep Learning Cloud Service Providers
Available at: https://towardsdatascience.com/list-of-deep-learning-cloud-service-providers-579f2c769ed6
[Accessed 09 July 2022]
Dettmers, T., 2020, Which GPU(s) to Get for Deep Learning: My Experience and Advice for Using GPUs in Deep Learning
Available at: https://timdettmers.com/2020/09/07/which-gpu-for-deep-learning/
[Accessed 09 July 2022]
colab.research.google.com subscription, 2022, Google Colab Subscription. Available at: https://colab.research.google.com/signup
[Accessed 09 July 2022]
Tan, E., 2022, Free GPUs for Training Your Deep Learning Models
Available at: https://towardsdatascience.com/free-gpus-for-training-your-deep-learning-models-c1ce47863350
[Accessed 09 July 2022]
Homes, J., 2022, Comparison of Basic Deep Learning Cloud Platforms
Available at: https://medium.com/geekculture/comparison-of-basic-deep-learning-cloud-platforms-337657edb710
[Accessed 09 July 2022]
Hale, J., 2019, Kaggle vs. Colab Faceoff — Which Free GPU Provider is Tops?
Available at: https://towardsdatascience.com/kaggle-vs-colab-faceoff-which-free-gpu-provider-is-tops-d4f0cd625029
[Accessed 09 July 2022]
Droste, B., 2021, Google Colab Pro+: Is it worth $49.99?
Available at: https://towardsdatascience.com/google-colab-pro-is-it-worth-49-99-c542770b8e56
[Accessed 09 July 2022]
