Congratulations! You made it to the end of this fourth course, which is also the end of this specialization. Let's recap all that we've covered so far about machine learning. You learned that AI or artificial intelligence is the field of study that includes machine learning and a sub-discipline of ML called deep learning. Recall that ML, for supervised learning at least, is a thing labeler. Then you leave the ML model with good labels examples and not with hard-coded instructions. That's what makes deep learning so amazing, is that even with no instructions, the models themselves can classify complex tasks like image recognition, like you saw on your Google Photos demo. You then learn a few of the different business areas that ML can transform, like operations, sales, marketing, and more. We then describe the key terminology that data scientists use. We call it instances or observations, or records or rows of data, labels or the correct answer column or columns in your dataset, and that the remaining columns that you use as inputs to ML model are called feature columns. Getting those feature columns right or creating new columns is called feature engineering. It's often the hardest part of any ML project. We then went over my favorite three secrets for ML, which again are; number 1, you don't have to start out saying that you're going to do an ML or an AI project at the outset. You're going to first explore and clean to describe your data to see if ML is even an option. Remember, if you can't do analytics, you can't do machine learning. We then learned in secret number 2, it's not just about training the actual models to perfection. You first have to create a great set of examples to train from, and then productionalize the model afterward. Next up, we compare the spectrum of ML tool options on the Google Cloud Platform, and from the left-hand side, the more programming intensive options like TensorFlow versus on the right-hand side, which are minimal effort, things like pre-trained ML APIs and AutoML. You got to then try your hand at using pre-trained ML APIs like image classification and sensitive analysis, where we have the advantage of not having to bring your own dataset to do ML. We then discussed the aspects of what makes a good feature column, and making sure to remain critical and curious about your dataset. You remember the Eiffel Tower image? Then you got to practice exploring and creating data pipelines with advanced Cloud Dataproc features like using UNNEST to work with that advanced data e-commerce schema. Then we discussed how you can get around knowing the unknowable future data without having a time machine. Remember those three keywords? Split your data. Specifically split it into sample then a repeatable fashion, into training, validation, and testing. After you've processed and cleaned and got your dataset ML ready, we had a very quick training and evaluation demo inside of BigQuery machine learning. You got to create forecasting and classification models using your course e-commerce dataset and your labs. Then we discussed how to productionalize your models by using a tool like Cloud Machine Learning Engine. Lastly, if you find machine-learning itself fascinating and you want to dive deeper into TensorFlow, check out the Machine Learning on Google Cloud Platform specialization. Here, I've broken down the decision tree based on what roles you have most interest in. Honestly, it's been a really exciting ride, and I hope you enjoyed the journey from data to insights. Thank you and we'll see you again soon.
