- Is it the way HCDC will use the model in practice? If so, make clear in your text
- Refit the model using data from 2013-the current week
- Make prediction for the next 4 weeks
- Convert to R_t
- Should be clear about the assumption of each model, strengths and weaknesses, when it can be used and when it should not be used (eg. if vaccination is widely introduced then I guess they should move away from this model, or just don't use data from as far as 2013 to fit it)
- Regarding this statement:
We can say with HCDC, after reviewing forecasting models, there are various approaches, including mechanistic, statistical, and machine learning models. However, for policymakers, they need estimates of uncertainty in forecasting accuracy, so there are three models that can provide this: ARIMA (SARIMA), Prophet, and Bayesian additive regression tree (BART).
Mechanistic models can provide uncertainty too, no? I'm unsure whether it is a valid reason. I'm also unsure about whether policymakers need uncertainty, or it is just us thinking so.
4.
For BART model with 2 week prediction ahead had 3 predictors: admission week, lag 1 week, lag 2 week.
This sentence is very confusing. So you are standing at week t and trying to predict week t+2. Your sentence sounds like you are using t and t+1 to predict t+2, which is incorrect because you are standing at t (hope this is not what you're doing lol). Can you make it clear (by showing the head of the training set, or other ways) to clarify that what you did is using t and t-1 to predict t+1 and t+2?
5. I think PCC and PICP is easier to interpret (since we compare them to 1), but you didn't show them
6. The text in figures are too small to read for my age
7. 
I think you meant the vertical line (when there is no bar) seperate training-prediction, not horizontal line? And what are the dots, are they the actual data? This figure is a bit confusing.
8. Now that you have the case predictions, maybe you can use the case prediction to produce a prediction of R_t, taking the prediction intervals into account? I think this is useful for HCDC to interpret the direction of the outbreak given the forecasting
Mechanistic models can provide uncertainty too, no? I'm unsure whether it is a valid reason. I'm also unsure about whether policymakers need uncertainty, or it is just us thinking so.
4.
This sentence is very confusing. So you are standing at week t and trying to predict week t+2. Your sentence sounds like you are using t and t+1 to predict t+2, which is incorrect because you are standing at t (hope this is not what you're doing lol). Can you make it clear (by showing the head of the training set, or other ways) to clarify that what you did is using t and t-1 to predict t+1 and t+2?
5. I think PCC and PICP is easier to interpret (since we compare them to 1), but you didn't show them
6. The text in figures are too small to read for my age
7.
I think you meant the vertical line (when there is no bar) seperate training-prediction, not horizontal line? And what are the dots, are they the actual data? This figure is a bit confusing.
8. Now that you have the case predictions, maybe you can use the case prediction to produce a prediction of R_t, taking the prediction intervals into account? I think this is useful for HCDC to interpret the direction of the outbreak given the forecasting