In this video, we're going to take the machine learning process life-cycle which you were just introduced to and talk about how farmer Betty might apply it to her machine learning project. Let's take a look at how this might be used in Betty's agricultural business. First of all, Betty has to think about her business, it's contexts, her business goals, and what question would help her reach those goals, also what would be appropriate for Machine Learning. In our example, she wants to increase crop yield, and she wants to use fertilizers as efficiently as possible both to decrease her cost and to decrease the larger environmental impact of the fertilizer use. We'll say that Betty settles on the question that we discussed earlier. What yield would I get if I used X amount of fertilizer Y on this plant right now? This is a specific question with business value and it's appropriate for machine learning. Betty also identifies some sources of data. Recall, she's got lots of data about whether and soil conditions, records of previous fertilization, and the output of a different QuAM that uses drone footage to determine what growth stage the plants are at. She also thinks about the stake holders, whether this QuAM is going to be distributed widely and used as part of her agricultural consultation business, or if it's just for her own fields. So Betty is moving on to phase two, data acquisition and understanding. Now Betty takes a close look at that data she has, makes sure it all lines up, it's all in the same units, and so on. In other words, she does the cleaning. She also makes sure that there's enough data on fertilizer amounts and types and crop yields for the QuAM to be able to answer the question. When she's sure that the data is in the right state for machine learning, she moves on to phase three. First, she chooses what algorithm she's going to start with. In this case, her question is a regression problem. So she needs to choose a regression algorithm instead of say a classification algorithm. She decides to start with something simple like linear regression, then she trains her QuAM on part of the data and tests it on the rest. Based on the results of the test, she'll go back and modify. We're not going into too much detail here, because we'll be discussing this process in much more depth in the next two courses of the specialization. We do want to emphasize though, it's a safe assumption for any given machine learning project you're going to need to go back and forth between training and testing multiple times. Once Betty is happy with the answers that are being output by her QuAM, she can move on to the fourth phase. If Betty is the only stakeholder there isn't a handoff, but she's still needs to have documentation of what she did with what data when, so that if there are problems she needs to answer about what she did and how she did it, she's not relying just on her own memory to be able to answer them. This is also the time to take a step back and make sure the model satisfies all the criteria she decided on backend phase one. Of course, we're simplifying somewhat here. In this description, Betty went from phase one to phase four in one cycle, and that's really rare. What's much more likely, is that, Betty would have gone back and forth between her question and data a few times, then back-and-forth between her machine learning model and her data. She might even have had to go all the way back to her question and start again from the beginning. But because Betty is an intrepid scientist and businesswoman, she'll get through the process and be ready to deploy her machine learning model.