Here are some lessons learned designing models, to help us assist you more efficiently.
The scientific method serves very well as a guideline, here is how we apply the scientific method:
- Identify the problem that you want to model
- Research the problem, the more detail you can collect, the more accurate your model will be.
- Try to decompose the problem into independent (things that you can change), and dependent (what you can measure) variables
- Some dependent variables may be observed only, others are measurable — both are equally valid
- If you can break the model into discrete independent variables, then you can change one at a time, this can help avoid having too many dependent variables change at once and creating difficulty in determining which independent variable is responsible
- Develop your hypothesis:
- It is helpful to be prepared for unexpected outcomes, or disproving your hypothesis, as they are often the most interesting outcomes
- Attempt to determine a control — a single independent variable (ie a plant grown with water only), so that you can consider other independent variables (ie a plant with water and fertilizer)
- Select the data sources required for your model
- Separate your data sources into the corpus (things you want to compare) from the learning set (things you want to compare too)
- Determine how you want to visualize your model.
With this in mind, we can often guide you on the next steps.
The most important thing to remember, this is a process. Your first model does not have to complete or perfect — simple and working is often preferable. The model can always be improved, but it has to exist first.
Also remember that you are often comparing results to nothing, exists now. This means that even if you model is 10% accurate – that is better than nothing. Error and be quantified and considered along with the model results.