Sprint Challenge

This sprint challenge will test your understanding of major neural network architectures covered in this sprint. You'll explore some of the cutting edge of Deep Learning by working with recurrent neural networks (RNNs), long short-term memory (LSTMs), convolutional neural networks (CNNs), and Autoencoders.

The challenge consists of two main parts:

1. LSTM Classification Model: Train an LSTM network to classify Reuters news articles into topics.
2. Pre-trained CNN for Object Detection: Use ResNet50v2 to detect frogs in images.

What You Need to Know

To successfully complete this Sprint Challenge, you should be familiar with:

• LSTM architecture and implementation in Keras
• Multi-class classification with neural networks
• Pre-trained CNN models, particularly ResNet50v2
• Image preprocessing techniques for neural networks
• Transfer learning concepts

Key Concepts to Review

• Text preprocessing for LSTM models
• Sequence padding and embedding layers
• How to use pre-trained models for inference
• Interpretation of neural network outputs
• Model evaluation for classification tasks

The challenge uses two datasets:

1. Reuters news articles dataset for the LSTM classification task
2. A collection of images for the CNN object detection task

Access the Challenge Notebook

You can access the Sprint Challenge notebook here:

Before submitting your notebook, follow these steps:

1. Restart your notebook's Kernel
2. Run all cells sequentially, from top to bottom
3. Delete or comment out any gridsearch code
4. For Part 2, comment out the code where you:
   • Load the images
   • Make the predictions
   • Display any plots or images
5. Verify that your notebook runs completely without errors

Part 1: LSTM Classification Model

In this section, you will:

• Use Keras to fit a predictive model classifying Reuters news articles into topics
• Name your model as model
• Use a single hidden layer
• Use sparse_categorical_crossentropy as your loss function
• Use accuracy as your metric
• Set epochs=1 due to CodeGrade resource constraints
• Report your overall score and accuracy

Note: Focus on getting a running model, not on maximizing accuracy with extreme data size or epoch numbers.

Part 2: Pre-trained CNN for Image Classification

For this part, you will:

• Use a pre-trained ResNet50v2 model to detect frogs in images
• Instantiate the model and save it to a variable named resnet_model
• Resize images to the appropriate dimensions for the model
• Apply proper preprocessing techniques
• Make predictions on each image to identify which ones contain frogs
• Decode the predictions (Note: ResNet50v2 labels frogs as "bullfrog", "tree frog", or "tailed frog")

Submission Requirements

Before submitting your completed notebook, ensure that you:

1. Restart the kernel
2. Run all cells sequentially, from top to bottom
3. Check that all cells have executed without errors
4. Remove or comment out any code that might cause CodeGrade to crash (gridsearch, image loading, etc.)
5. Verify that your notebook runs completely before submission