π If you're new to AI and eager to learn, PyTorch
is an excellent library to start with. For beginners, "Deep Learning for Coders with Fastai and PyTorch, 1st Edition", along with the Practical Deep Learning for Coders YouTube playlist and its website
, provide a basic introduction to AI topics.
βοΈ While Fastai
simplifies PyTorch
for newbies, it may not be suited for real-world projects due to maintenance problems between versions 1 and 2.
π₯οΈ To create an efficient environment for practicing the content of "Deep Learning with PyTorch, 1st Edition", consider utilizing a remote server. This technique avoids the complications of local machine setup and potential package conflicts, which are typical in resource-intensive activities like deep learning algorithms.
π¦ The second part of the book entails downloading 66.7 gigabytes of data (Luna16 Dataset), which requires around 220 gigabytes of storage for manipulating medical images.
π§ To solve these obstacles, lightning.ai
offers free remote servers with suitable processing power and storage. lightning.ai
prevents the establishment of new Conda environments within their studios, however Docker containers give a rapid alternative to this issue.
π During my study on the PyTorch
medical imaging project, "Docker: Up & Running: Shipping Reliable Containers in Production, 3rd Edition" proved vital. Docker's isolation feature allows you to treat running containers as full OS environments, easing the installation and upgrading of programs and libraries.
- Click on the badge (you need to create an account in
lightning.ai
). - Clone my repository:
git clone https://github.com/AliSerwat/PyTorch-DeepLearning-EasySetup.git
- Change
create_docker_image.sh
's mode:
chmod +x ~/PyTorch-DeepLearning-EasySetup/create_docker_image.sh
- Execute
create_docker_image.sh
:
~/PyTorch-DeepLearning-EasySetup/create_docker_image.sh
After Creating and Executing the Container, something like this should be displayed in your terminal
(base) root@aeee2fcb741b:~#
- Activate
pytorch_env
environment:
conda activate pytorch_env
It must look something like this:
(pytorch_env) root@aeee2fcb741b:~#
git clone https://github.com/deep-learning-with-pytorch/dlwpt-code.git
-
Fix a minor issue in
~/dlwpt-code/util/disk.py
:-
Open the file in your editor:
code ~/dlwpt-code/util/disk.py
-
Replace the following lines:
import gzip from diskcache import FanoutCache, Disk from diskcache.core import BytesType, MODE_BINARY, BytesIO from util.logconf import logging log = logging.getLogger(__name__) # log.setLevel(logging.WARN) log.setLevel(logging.INFO) # log.setLevel(logging.DEBUG)
with:
from util.logconf import logging import io import gzip from diskcache import FanoutCache, Disk # delete BytesType and BytesIO declarations from diskcache.core import MODE_BINARY BytesType = bytes # Import them by ourselves BytesIO = io.BytesIO log = logging.getLogger(__name__) # log.setLevel(logging.WARN) log.setLevel(logging.INFO) # log.setLevel(logging.DEBUG)
-
Due to internet connectivity restrictions in Iran, including filtering and sanctions, I want to share how to overcome these challenges.
β‘ AI has brought tremendous potential to our lives, requiring minimal resources compared to other emerging fields. However, conducting original research projects can be challenging due to budget and resource scarcity in many fields.
Since the infrastructure for AI development is accessible remotely and freely, I want to emphasize this opportunity. It allows your talents and potential to flourish despite circumstances beyond your control, such as where you were born.
- π VPN for accessing unlimited connection: Purchase VPN plans tailored to your needs from
this Telegram bot
, which offers helpful tutorials (I typically use a Great Britain IP address to bypass restrictions). - π Purchase a single-use virtual phone number from
numberland.ir
. I bought an inexpensive England number to verify mylightning.ai account
.β οΈ Ensure your VPN's IP address and the number's origin country match to avoid discrepancies.