Do experiments

[mceiring]

• Paramter tuning
• Learning rate, regurlarization parameter, dropout etc
• Depth and width of neural network
• - Model specific parameters for CNNs and RNNS or any other type of neural network you are training
• - For unsupervised techniques cluster size etc
• - compare all the approaches
• - document the limitations of all the approaches you try
• - Choose the best performing method and do more detailed analysis
• - evaluation measures
• - Precision, Recall, F1, ROC curve
• - Statistical significance test (see the examples here Links to an external site.)