AI-FDU

Project1: SVM Protein Classification

Environment Setup:

1. Python Environment: Ensure you have Python installed (preferably Python 3.x).
2. Dependencies: Install required dependencies using pip:

pip install numpy pandas scikit-learn biopython

Code Flow:

1. Data Preprocessing:
   • The code preprocesses the data, loading protein structure diagrams and sequence information.
   • If the --ent flag is provided, the data is loaded from a file using a feature engineering function feature_extraction() from fea.py. Otherwise, it loads from pre-existing files.
   • The code reads a CAST file containing protein sequence information and a Numpy array containing diagrams.
2. Model Initialization:
   • Three types of models are supported and need implementing: Support Vector Machine (SVM), Linear SVM, and Logistic Regression (LR).
   • You can specify the model type using the --model_type argument. Options are 'svm', 'linear_svm', and 'lr'.
   • For SVM models, you can choose the kernel type (--kernel) from 'linear', 'poly', 'rbf', or 'sigmoid'.
   • Regularization parameter C can be set using the --C argument.
3. Training and Evaluation:
   • The code trains the selected model on the training data and evaluates its performance on both training and test datasets.
   • It partitions the dataset into training and testing sets for each task.
   • The model’s accuracy is printed for each dataset.

Filling in the Blanks:

1. LRModel Class:
   • Fill in the initialization, training, and evaluation methods for the Logistic Regression model.
2. LinearSVMModel Class:
   • Implement the initialization method for the Linear SVM model.
3. Test Data Generation:
   • Complete the generation of test data by complementing the train data. Ensure correct reading positions for test data.

Running the Script:

• Execute the script main.py from the command line.
• You can provide arguments to customize the model type, kernel type, regularization parameter, and data loading method.

Example Command:

python main.py --model_type svm --kernel rbf --C 1.0

Experimental Requirements:

1. Complete Implementation of Protein Classification (Data Loading) - 4 points
   • Ensure the code effectively reads and preprocesses protein structure data and sequences.
2. Comparison of Linear SVM with Other Machine Learning Methods (e.g., LR) - 2 points
   • Implement Linear SVM model and LR model.
   • Analyze and compare the performance of SVM with other methods in terms of accuracy and computational efficiency.
3. Analysis and Discussion on the Impact of SVM Kernel Functions and Regularization Coefficients - 2 points
   • Investigate the effects of different SVM kernel functions on classification performance.
   • Analyze how varying the regularization coefficient (C) affects the model’s performance and generalization.
4. Feature Engineering: - 2 points
   • Provide insights on extracting useful features from protein structure data or utilizing feature selection methods to reduce dimensionality.
   • Modify the feature extraction function for the final experimental analysis.

Total Score: 10 points

Submission:

Submit the code along with an experimental report (up to 4 pages, No Involution) to the e-learning platform.

Note:

• Ensure you now are in ‘pj1’ path.
• Adjust the arguments as per your requirements for experimentation.
• Ensure the experimental report includes detailed explanations of the implemented code, experimental setup, results, analysis, and discussions based on the specified requirements.