some updates

content/blog/2022-07-06-switching-to-dvc-from-jupyter-vscode.md

@@ -36,7 +36,29 @@ we'll be working with.
 
 ## Start training with the notebook
 
-## Integrate with DVC
+Many times you'll start an ML project with a few cells in a notebook just to
+test out some thoughts you might have. So you might have a simple notebook where
+you set some hyperparameters, load your data, train a model and evaluate its
+metrics, and then save the model. That's what we're doing in the
+`bicycle_experiments.ipynb` file.
+
+So we have all of the cells in place and start running experiments. This is
+usually fine for training models for a while. Although there will eventually be
+a point where you are powering through experiments for the day and you want to
+compare metrics across experiments. You might also end up with a great model,
+but you have no idea what code you used or which data was used to train this
+model.
+
+That makes reproducing the experiment impossible and you're left with a model
+you may not be able to use in production. Once you reach the point where you are
+trying to reproduce models or compare metrics from multiple experiments, it
+might make sense to look at a data versioning and model experiment tracking tool
+like DVC.
+
+## Refactor the notebook to use DVC
+
+You can take your existing Jupyter notebooks and break them out in files and
+stages that DVC tracks for you.
 
 ## Look at metrics from experiments