[SOLVED] NCTU-CS Assignment #4 -Support Vector Machine & ANN

Competition

This homework is held on Kaggle as a competition so that you could see how it works.

• Click the link to participate.
• The competition provides you a training and a testing set.
  • training set – train.json
  • testing set – test.json
• Since it’s a competition, you won’t know the answer to the testing set, which is for you to predict and submit.
• The standard procedure of a competition:

• 1. Understand the data
  2. Split the provided training set into training subset and validation set for validation methods.
  3. Preprocessing, model construction, tuning
  4. Retrain the best model with as much data as possible, and predict testing set and make a submission.
  5. Win the competition
• If you have any questions, post them in the Discussion section or on Discord so everyone can see and understand.

Objective

1. Data Input – 5%
   • Download the training set and testing set from Kaggle.
2. Data Preprocessing – 15%
   • Transform data format and shape so your model can process them.
   • Shuffle the data.
   • Any data augmentation that can boost your final results. – 10%
3. Model Construction – 50%
   • Support Vector Machine – 20%
     • for SVM model, you may want to try out different types of kernels and compare the result.
   • Artificial Neural Networks – 30%
     • for ANN model, you could use any Neural Network based model you want and implement it by yourself.
     • Every framework (such as TensorFlow or PyTorch) is allowed.
     • explain the reasoning of your model choice, data augmentation, and training process.
   • Validation method
     • Holdout validation with the ratio 7:37:3
4. Results – 10%
   • Obtain the performances of all experiment settings in tables by the following metrics:

1. • 1. Confusion matrix
     2. Accuracy
     3. Sensitivity(Recall)
     4. Precision
2. Comparison & Conclusion – 10%
   • Also some feedback, anything you want to tell me.
3. Kaggle Submission – 10% (+30%)
   • After the validation, now you have working SVM and ANN models.
   • Retrain one of your best models with the whole train.json, predict test.json, and submit your y_test.csv to Kaggle.

1. • • You can check sample_submission.csv for the submission format.
   • Take a screenshot of the Leaderboard, highlight your name, and put it in the report.
   • Top 10 in the final Private Leaderboard can get 30 bonus scores.

Note that you still need to submit your report and code to the newE3 system.

Data – Recipe Ingredients Dataset

• The objective of the competition is to predict the category of a dish’s cuisine given a list of its ingredients.
• In the dataset, we include the recipe id, the type of cuisine, and the list of ingredients of each recipe (of variable length). The data is stored in JSON format.
• An example of a recipe node in train.json:

• {          “id”: 24717,·          “cuisine”: “indian”,·          “ingredients”: [·              “tumeric”,·              “vegetable stock”,·              “tomatoes”,·              “garam masala”,·              “naan”,·              “red lentils”,·              “red chili peppers”,·              “onions”,·              “spinach”,·              “sweet potatoes”·          ]·          },

• In the test file test.json, the format of a recipe is the same as train.json, only the cuisine type is removed, as it is the target variable you are going to predict.