The Data Science Workshop: A New, Interactive Approach to Learning Data Science

Working towards making a transition into the field of Data Science this book has been a great resource for both reinforcing skills as well as learning new ones. The layout is a bit more like taking a course with examples and walkthroughs to help build on what was discussed in the chapter. Each subject has a good mix of instruction, visualizations, and easily readable sample code. Materials are well organized online for easy access for whatever chapter or subject you want to dive into.I did find the structure of the book to be a bit tricky to follow with some of the metrics for assessing model performance being discussed after the chapter with the models themselves. That being said the explanations are clear and that may just be a personal quirk on how I want to review the material.Overall I’d recommend the book especially for beginners or those looking to learn new tools such as Altair API for data visualization.

This book is a great supplementary book for an introduction to data science course. The book could be used alone as well, but would be most beneficial for practice in a workshop-style class. I'm considering using this book for assignments for my students. The only side note might be that it is a beginner and low intermediate-level book and would likely be too easy for more experienced students.

_The Data Science Workshop_ delivers on the promise made in its subtitle: “a new, interactive approach to learning data science”. The computer language of the book is Python, which is quickly becoming the lingua franca of data science everywhere, and the authors helpfully take a moment, before beginning in earnest, to justify the use of that language (p. 2). The book contains lots of hands-on coding exercises, starting even in the first chapter, and it makes use of lots of relatively new and valuable Python packages, such as altair (p. 87 ff.), featuretools (Chap. 17), lime (p. 432 ff.), mlxtend (p. 419 ff.) and smote-variants (p. 605 ff). Each chapter also includes a list of SWBATs (“students will be able to”) at the beginning and a summary of key ideas at the end, all of which are helpful.But there are also some structural aspects that are not as good. The early chapters jump into modeling right away, including such notions as model overfitting (p. 28), hyperparameters (p. 28), and training data (p. 29), each of which could use a section all on its own. Skipping over these details is nice for the student who’s eager to jump in and start coding, but it could also make these chapters rather intimidating for the beginner who prioritizes understanding over doing.And the unfortunate truth is that scattered throughout the book there are less than satisfactory explanations of various key concepts and tools, including:statistical notions like hypothesis testing, confidence intervals, and Pearson correlation (Chap. 2);modeling algorithms like RandomForest (Chap. 4), LASSO/Ridge (Chap. 7), and SVMs (Chap. 8);modeling metrics like accuracy, recall, and logarithmic loss (Chap. 6); andsklearn tools like StandardScaler (Chap. 3), PolynomialFeatures (Chap. 6) and why OneHotEncoder is superior to pandas.get_dummies() (Chap. 16).Again, the emphasis seems to be on how to use these tools in a Python environment. Some students will indeed be looking for nothing more than this, but others will want to hear more about LASSO than the fact that “a penalty is introduced in the loss function” (p. 327).There is, moreover, a problem throughout with loose language that sometimes rears its head. To name a few examples:a regression model is described as trying to “find a solution” for a linear equation, even though regression modeling is an exercise in optimization rather than in exact solution (p. 27);in the context of model evaluation, the claim is made that “what we want is to get a model that makes extremely accurate predictions, so we need to assess its performance using some kind of metric”, when accuracy is only one of several metrics we may use (p. 140);Euler’s number is said to be “the natural logarithm” (p. 120);the square root of 2 is said to be “equal to 1.45” (p. 164);sklearn’s OneHotEncoder class is several times called a function (p. 715 ff.).Some of these are hardly terrible mistakes. But the point is just that there is a certain casualness over some important details that is fine for ordinary discourse but less than fine for a technical discussion.In spite of these shortcomings, the book is also to be praised for making clever use of some Python tools, using:Normalizer (pp. 114-15) and RobustScaler (pp. 589-90) from sklearn.preprocessing;sklearn.externals.joblib to save and load a model (pp. 278-9)plot_partial_dependence() from sklearn.inspection (pp. 428 ff.)inverse_transform() from PCA to illustrate the orientation of principal components (pp. 646-7);underutilized pandas DataFrame methods like select_dtypes() (p. 466) and duplicated() (p. 501 ff.); and pop() to separate predictors and target (p. 138);the usecols (p. 178), na_values (p. 531), and error_bad_lines (p. 620) parameters in read_csv() from pandas; andRandomizedSearchCV (p. 376 ff.) and cv_results_ from GridSearchCV (p. 367) to analyze the results of hyperparameter tuning.There are in addition some helpful aids for illustrating concepts that often go unexplained, such as a table to illustrate patsy syntax (p. 63), a paragraph devoted to the ‘k-means++’ value of the init parameter in KMeans (p. 201), and an inventive comparison of overfit models to a student memorizing examples in anticipation of a test (p. 141).In summary, _The Data Science Workshop_ is an excellent resource for what contemporary data science work in Python looks like, and it is full of examples and exercises that would be useful for many who are new to the industry. But the student of data science who is looking for deep explanations of the concepts and algorithms underlying popular Python tools will need to supplement their path through this book with other resources. Still, I recommend _The Data Science_ Workshop highly for anyone who is looking for engaging practice with some of data science’s most popular tools.

DO NOT BUY AMAZON BOOKS. THEY ARE SUCH POOR QUALITY! Mine came and the letters and layout are all distorted, it was cut lopsided and overall looks and feels like a year old made it. For a textbook that im supposed to use to study, it makes it impossible to do so. BUY FROM YOUR LOCAL STORE.