6 Engaging Deep Learning Applications for Beginner Python Developers

Introduction

Deep learning has emerged as one of the most exciting and transformative areas of artificial intelligence in recent years. By training artificial neural networks on massive amounts of data, deep learning enables computers to learn and extract intricate patterns to make decisions and predictions. This powerful approach has led to remarkable breakthroughs in fields like computer vision, natural language processing, and robotics.

While cutting-edge deep learning research can get complex quickly, the good news is that you don‘t need a PhD to start using this technology for your own projects. With the help of open source frameworks and cloud APIs, any developer can now easily leverage pre-trained deep learning models to build their own intelligent applications. Python, with its simplicity and extensive library ecosystem, has become the language of choice for deep learning.

In this article, we‘ll walk through 6 fun and engaging deep learning projects that are perfect for Python beginners to sink their teeth into. For each project, we‘ll use either an existing cloud API or open source code to demonstrate the key concepts. Let‘s jump in and see the fascinating applications that are possible with deep learning!

Deep Learning APIs

One of the easiest ways to harness the power of deep learning is through web APIs. A number of companies like Google, Microsoft, IBM, Amazon, and startups offer APIs that provide access to pre-trained deep learning models. These APIs allow you to perform complex tasks like image classification, sentiment analysis, language translation and more with just a few lines of code.

Advantages of Deep Learning APIs

– No need to train your own models, which can be time and resource intensive
– Simpler to use and integrate into applications
– Continuously improved models without extra effort on your part
– Can scale to handle production workloads

Disadvantages of Deep Learning APIs

– Less flexibility and control compared to training your own models
– Can be expensive, especially at high volumes
– Requires internet connectivity to work
– Dependent on the API provider, risking loss of functionality if API shuts down

Now let‘s check out 3 cool deep learning APIs you can use for your Python projects.

1. Colorize Photos with DeOldify

Have you ever wondered what historical black and white photos would look like in color? Or wanted to colorize your own monochrome images? DeOldify is a deep learning model that does exactly that. DeOldify‘s API, powered by Algorithmia, couldn‘t be simpler to use.

Here are the steps:

1. Sign up for a free API key at Algorithmia.com
2. Install the Algorithmia Python client:

   pip install algorithmia
   

3. Use the code below to colorize an image:

import Algorithmia
input = {
"image": "data://username/folder/image.jpg",
}
client = Algorithmia.client(‘YOUR_API_KEY‘)
algo = client.algo(‘deeplearning/ColorfulImageColorization/1.1.13‘)
result = algo.pipe(input).result

Replace the image path with your own image URL on Algorithmia, paste in your API key, and run the script. The original and colorized images will be available in your Algorithmia hosted data collection. It‘s amazing to see old photos come to life with vibrant colors using this API.

2. Build Your Own Chatbot with IBM Watson

Chatbots have become ubiquitous these days, helping businesses automate customer support and interact with users in engaging ways. With IBM‘s Watson Assistant API, you can build your own AI chatbot without any prior machine learning experience.

Follow these steps to create your Python chatbot:

1. Sign up for an IBM Cloud account and provision a Watson Assistant service
2. Obtain the service credentials and API key
3. Install the Watson Developer Cloud Python library:

   pip install --upgrade watson-developer-cloud
   

4. Train your chatbot with some sample dialogs in the Watson Assistant UI
5. Use this code to start conversing with your bot:

from watson_developer_cloud import AssistantV1
assistant = AssistantV1(
version=‘2018-02-16‘,
username=‘YOUR SERVICE USERNAME‘,
password=‘YOUR SERVICE PASSWORD‘
)
response = assistant.message(
workspace_id=‘YOUR WORKSPACE ID‘,
input={
‘text‘: ‘Hello‘
}
).get_result()
print(json.dumps(response, indent=2))

Paste in your service credentials and workspace ID from step 2. Run this script and chat with your bot from the console! You‘ll be amazed at how it can understand intents, extract entities, and carry on a multi-turn dialog.

3. Perform Sentiment Analysis on News with Aylien

Sentiment analysis is a powerful natural language processing technique that determines the emotion and opinion expressed in a piece of text. With the Aylien News API, you can easily perform sentiment analysis on news articles to understand the overall positive or negative tone.

Here‘s how to build a news sentiment analyzer in Python:

1. Sign up for an Aylien developer account and get your API key
2. Install the Aylien News API Python client:

   pip install aylien-news-api
   

3. Use the following code to retrieve news articles and analyze their sentiment:

from aylienapiclient import textapi
client = textapi.Client("YOUR_APP_ID", "YOUR_APP_KEY")
opts = {
‘title‘: ‘trump‘,
‘sentiment_title_polarity‘: ‘positive‘,
‘language‘: [‘en‘],
‘published_at_start‘: ‘NOW-7DAYS‘,
‘published_at_end‘: ‘NOW‘
}
stories = client.Stories(**opts)
for story in stories.stories:
print(story.title, story.sentiment.polarity)

Enter your app ID and key, and run the script. It will retrieve English news stories from the past week with "trump" in the title and a positive sentiment, and print each story‘s title and sentiment polarity score. You could build a news app that only shows users positive stories based on this API.

Open Source Deep Learning

While APIs are convenient, you‘ll often want more control and flexibility over your deep learning models. That‘s where open source frameworks like TensorFlow, PyTorch, and Keras come in. These libraries allow you to build and train your own custom models on your own datasets.

Advantages of Open Source Deep Learning

– Complete flexibility to customize models and training pipelines
– Ability to train on your own private data, keeping it secure
– Can run models offline without depending on APIs
– No cost beyond your own development and compute resources

Disadvantages of Open Source Deep Learning

– Steeper learning curve to understand the frameworks and techniques
– Building effective models requires significant time and effort
– Need access to large amounts of training data and compute power
– Have to maintain and update your own models over time

Here are 3 awesome open source deep learning projects you can try out.

4. Fix Grammatical Errors with Deep Text Corrector

Most spell checkers today can catch typos and misspellings, but fixing grammar and word usage errors is much harder. Deep Text Corrector is an open source project that uses neural machine translation to correct various types of errors in sentences.

To try it out:

1. Clone the GitHub repo:

   git clone https://github.com/atpaino/deep-text-corrector.git
   

2. Download and extract the pre-trained model here
3. Install the Python dependencies:

   pip install -r requirements.txt  
   

4. Run the model to correct an input sentence:

from correct_text import correct_text
corrected_text = correct_text(
"she was unaware that he stole candy from the baby.", 
model_path=‘path/to/downloaded/model/‘
)
print(corrected_text)  

The corrected sentence will be printed out. The model works reasonably well, catching errors like "unaware" to "aware", "stole" to "had stolen", etc. Of course, no automated grammar tool is perfect, but this project shows the potential of deep learning for natural language correction.

5. Transform Male Portraits to Female with CycleGAN

Want to see what the Mona Lisa would look like as a man? With generative adversarial networks (GANs), we can train models to transform images in remarkable ways, like converting between male and female faces.

The CycleGAN project enables this by training on a dataset of male and female portraits. To run it:

1. Clone the repo:

   git clone https://github.com/junyanz/CycleGAN.git
   

2. Download the pre-trained model
3. Install PyTorch and the other Python requirements
4. Convert a male portrait to female with:

python test.py 
  --dataroot datasets/male2female/testA
  --name male2female_pretrained
  --model test 
  --no_dropout

Wait a few minutes and check the results folder – the male portraits will be converted to females! GANs are capable of learning the key facial features that distinguish genders and transferring them to the input images. The results are scarily good.

6. Train an AI to Play Flappy Bird

Remember that addictive mobile game Flappy Bird? Turns out you can teach an AI bot to play it better than any human using deep reinforcement learning. This project uses Python and Keras to train an agent to flap through the pipes.

Here‘s how to see it in action:

1. Install the Python dependencies:

   pip install keras pygame
   

2. Clone the repo:

   git clone https://github.com/yanpanlau/Keras-FlappyBird.git
   

3. Run the pre-trained agent:

   python run_pretrained_model.py
   

Watch in awe as the bot perfectly times its flaps to navigate through the pipes! The agent learned through trial-and-error – by initially choosing random actions, observing the rewards (points), and gradually learning to choose better actions to maximize its score. After a few hundred games, it discovered an optimal policy to play.

With a few tweaks to the game parameters and training code, you can build an AI to play just about any simple arcade game. Deep reinforcement learning is a hot area of research that‘s led to computers mastering complex games like Go, Dota and Starcraft.

Additional Resources

I hope these 6 projects gave you a taste of the exciting applications you can build with deep learning. There are endless more projects to try – I‘ll leave you with a few:

• Fast Style Transfer: Render your images in the style of famous artworks
• Facial Emotion Recognition: Detect emotions in facial expressions
• Neural Story Generation: Generate creative stories with deep learning
• Gesture Controlled Games: Play Pac-man with your hand gestures!

To learn more, check out these excellent resources:

• Machine Learning Crash Course (Google)
• Practical Deep Learning for Coders (Fast.ai)
• Deep Learning Specialization (Coursera)
• TensorFlow and Keras documentation

Conclusion

We‘re living in an incredible time when anyone can leverage the power of deep learning to build meaningful and impactful applications. The 6 projects we covered here barely scratch the surface of what‘s possible. As the field advances and more pre-trained models and APIs become available, the barriers to building AI applications will only get lower.

At the same time, deep learning is a complex and rapidly evolving field. Beginners can easily get started with APIs and simple models, but achieving state-of-the-art results requires advanced math, algorithms and a deep (no pun intended) understanding of the technology.

I encourage you to not just use deep learning as a black box, but to dive in and learn the core concepts. Our world is being transformed by AI, and it‘s up to us as developers and citizens to steer that transformation in a positive direction. The more people who deeply understand this technology, the better equipped humanity will be for the future.

So get out there and start tinkering with deep learning! Happy hacking!