Errata

In Chapter 1, there is an external link that does not work: "step-by-step guide to creating a boxplot." at location 684. Please update with a valid external URL.

Note from the Author or Editor:
The link:
https://oreil.ly/wTpnE

should be replace with:
https://web.archive.org/web/20190415201405/https://www.oswego.edu/~srp/stats/bp_con.htm

In Chapter 2 there is an external link that does not work "Fooled by Randomness Through Selection Bias" at location 1347. Please update a valid external URL.

Note from the Author or Editor:
This is referring to link https://oreil.ly/v_Q0u

The correct link is now:
https://www.priceactionlab.com/Blog/2012/06/fooled-by-randomness-through-selection-bias/

The behavior of the get_dummies method has changed recently. Instead of creating integer columns containing 0 and 1, the function now creates boolean columns with True and False values. This causes statsmodels model building to fail with an exception.

To revert to the original behavior, add the keyword argument `dtype=int` to the get_dummies method calls.

The link to the pandas documentation ( https://oreil.ly/UGX-4 ) results in a 404 error. The O'Reilly redirect appear to attempt to access https://pandas.pydata.org/pandas-docs/stable/getting_started/basics.html#basics-dtypes .

Note from the Author or Editor:
We need to change to redirect
https://oreil.ly/UGX-4
to redirect to:
https://pandas.pydata.org/docs/user_guide/basics.html#dtypes
Ideally, this can be done without changing the short URL

--redirect all set (O'Reilly errata team)

Text currently states:
...flights by the cause of delay at Dallas/Fort Worth Airport since 2010.

should be:
...flights by the cause of delay at Dallas/Fort Worth Airport in 2010.

The mean of the sample of 20 datasets that was used to generate Figure 2-9 was $55,734.

Replace $62,231 with $55,734.

This was already changed once to $55,836, but the actual value should be $55,734. I remember that I found this confusing too, so I suggest we add a clarification to this.

... for which the mean was $55,734. Note that this is the mean of the subset of 20 records and not the mean of the bootstrap analysis, $55,836.

Changed in repository

The mean of the random values generated using the rexp(n=100, rate=0.2) function in R is ~5, which makes sense given that the mean number of events per time period is 0.2. However, for the Python code given in the book as stats.expon.rvs(0.2, size=100) we have the mean of the random values generated ~1.2, where loc=0.2 is the starting location for the exponential distribution. To get the same range of random values as those obtained with R we need to use stats.expon.rvs(scale=5, size=100) instead.

Note from the Author or Editor:
The errata is correct and requires a change in the book.

Suggested change:

The +scipy+ implementation in +Python+ specifies the exponential distribution using +scale+ instead of rate. With scale being the inverse of rate, the corresponding command in Python is:

.Python
[source,python]
----
stats.expon.rvs(scale=1/0.2, size=100)
stats.expon.rvs(scale=5, size=100)
----

In Google Analytics the average session time does not measure the time spent on a given page (as stated in the book), the correct metric is average time on page. Furthermore, on the last paragraph we have "Also note that Google Analytics, which is how we measure session time, cannot measure session time for the last session a person visits." I think it would be more correct to say: Also note that Google Analytics, which is how we measure average time on page, cannot measure the time spent on the last page within a session. Finally, Google Analytics does indeed set the time spent on the last page in a session to zero, and a single-page session is also set to zero. Having said that, this is true only if there are no user interaction events triggered on that page, such as click events, scroll events, video events, etc.

Note from the Author or Editor:
Thank you for the feedback. We change the text int the book to:

Also note that Google Analytics, which is how we measure average time on page, cannot measure the time spent
on the last page within a session. ((("Google Analytics")))
Google Analytics will set the time spent on the last page in a session to zero, unless the user interacts with the page, e.g. clicks or scrolls. This is also the case for single-page sessions. The data requires additional processing to take this into account.

Issue reported on github repository:

The following code makes a variable call to the chi2 value calculated using the permutation test (chi2observed), vice the chi2 value computed using the scipy stats module (chisq).

chisq, pvalue, df, expected = stats.chi2_contingency(clicks)
print(f'Observed chi2: {chi2observed:.4f}')
print(f'p-value: {pvalue:.4f}')

I believe the first print line should be:
print(f'Observed chi2: {chisq:.4f}') since the purpose is to demonstrate using the chi2 module for statistical tests rather than the previous sections permutation test.

This is correct. Code and book text corrected.

Regarding the paragraph

"Location and house size appear to have a strong interaction.
For a home in the lowest +ZipGroup+,
the slope is the same as the slope for the main effect +SqFtTotLiving+,
which is $118 per square foot (this is because _R_ uses _reference_ coding for factor variables; see 'Factor Variables in Regression').
For a home in the highest +ZipGroup+,
the slope is the sum of the main effect plus +SqFtTotLiving:ZipGroup5+,
or $115 + $227 = $342 per square foot.
In other words, adding a square foot in the most expensive zip code group boosts the predicted sale price by a factor of almost three, compared to the average boost from adding a square foot."

I am thinking about two things:
1.) The coefficient for +SqFtTotLiving+ is 1.148e+02, but it is stated that "the main effect +SqFtTotLiving+ [...] is $118 per square foot". I think it should be adjusted to $115 as mentioned in the subsequent sentece.

2.) Since R uses reference coding (and not deviation coding), I wonder whether the last sentence is correct. Is it really the "average boost from adding a square foot" you compare to with the total effect of the most expensive zip code group? I mean, if you don't include any interaction effect the coefficient of +SqFtTotLiving+ would be the "average boost" (as far as I think about it). But in the setting with an interaction effect and reference coding, I would have interpreted it as "compared to the average boost for the lowest zip code group". Or am I wrong and the average boost is the same as the main affect, which in turn is equal for the first ZipGroup?

Best regards

Note from the Author or Editor:
Thank you for your feedback. This corresponds to page 175 second paragraph in the print edition.

1) $118 should be replaced with $115

2) We are going to change the end of the second paragraph for clarification to:
... to the average boost from adding a square foot in the lowest zip code group.

Gitlab is changed.

I believe that Figure 4-12 on page 191 is in error because the code used to generate it (Chapter 4 - Regression and Prediction.R from the practical-statistics-for-data-scientists-master.zip file) appears to be in error.

The code states:

terms1 <- predict(lm_spline, type='terms')
partial_resid1 <- resid(lm_spline) + terms


but surely partial_resid1 should be:

partial_resid1 <- resid(lm_spline) + terms1

which would give rise to a slightly different plot?

Note from the Author or Editor:
I can confirm the error in the R code. The R code is not printed in the book, but the image created is. As mentioned in the errata, the difference in the plot is only small.

I changed the code to create the correct plot.

New figure file images/psds_0412.png added to book repository. This file will need to be processed (cropping the whitespace) to replace the file psds2_0412.png.

Regarding the paragraph

"An example will make this more explicit.
For the model fit in "Logistic Regression and the GLM" on page 210,
the regression coefficient for +purpose_small_business+ is 1.21526.
This means that a loan to a small business compared to a loan to pay off credit card debt reduces the odds of defaulting versus being paid off by exp(1.21526) ≈ 3.4.
Clearly, loans for the purpose of creating or expanding a small business are considerably riskier than other types of loans."

Suggested change:
This means that a loan to a small business compared to a loan to pay off credit card debt *increases* the odds of defaulting versus being paid off by exp(1.21526) ≈ 3.4.

Best regards

Note from the Author or Editor:
The errata is correct. Gitlab document changed accordingly - PG

Why bother with an odds ratio rather than probabilities? We work with odds because
the coefficient βj in the logistic regression is the log of the odds ratio for Xj .

anyway we can't state that coefficient βj is the log of the odds ration for Xj, since that mean we take log of log

i think the correct statement will be "We work with odds because the coefficient βj in the logistic regression is the *change in* log of the odds ratio for Xj.

Note from the Author or Editor:
In the cited location..
EXISTING
Why bother with an odds ratio rather than probabilities? We work with odds because the coefficient βj in the logistic regression is the log of the odds ratio for Xj .
CHANGE TO (end of second sentence)
Why bother with an odds ratio rather than probabilities? We work with odds because the coefficient βj in the logistic regression is CHANGE IN the log (Odds(Y=1)) associated with a change in Xj .

On page 217 of the printed book (2nd edition), the R code at the end of the page reads:

terms <- predict(logistic_gam, type='terms')
partial_resid <- resid(logistic_model) + terms
df <- data.frame(payment_inc_ratio = loan_data[, 'payment_inc_ratio'],
terms = terms[, 's(payment_inc_ratio)'],
partial_resid = partial_resid[, 's(payment_inc_ratio)'])

I believe that partial_resid here should be:

partial_resid <- resid(logistic_gam) + terms

I'm not sure if the graph produced on page 218 (Figure 5-4) using this data needs correction or not, as the difference using logistic_model and logistic_gam is quite minor, and it is hard to tell comparing a screenshot and the printed page.

Note from the Author or Editor:
The line needs to be changed in the asciidoc code. It is already corrected in the book's Github repository, however I overlooked changing the book text. That is now corrected too.

Since the R code yields TRUE for the prediction knn_pred == 'paid off', the sentence

"The KNN prediction is for the loan to default."

seems to be wrong and "default" should be replaced with "be paid off".

Note from the Author or Editor:
This is correct. The sentence should read:

The KNN prediction is for the loan to be paid off.

scikit-learn implements Tree-complexity pruning like in R

In version 0.22, scikit-learn implemented tree complexity pruning for decision trees.

https://scikit-learn.org/stable/auto_examples/tree/plot_cost_complexity_pruning.html#sphx-glr-auto-examples-tree-plot-cost-complexity-pruning-py
https://scikit-learn.org/stable/modules/tree.html#minimal-cost-complexity-pruning

Replace with:

===== Controlling tree complexity in _Python_
In the +scikit-learn+’s decision tree implementation, the complexity parameter is called +cpp_alpha+. The default value is 0, which means that the tree is not pruned; increasing the value leads to smaller trees. You can use GridSearchCV to find an optimal value.

There are a number of other model parameters, that allow controlling the tree size. For example, we can vary +max_depth+ in the range 5 to 30 and +min_samples_split+ between 20 and 100. The +GridSearchCV+ method in +scikit-learn+ is a convenient way to combine the exhaustive search through all combinations with cross-validation. An optimal parameter set is then selected using the cross-validated model performance.

"Figure 7-7 shows the cumulative percent of variance explained for the default data for the number of clusters ranging from 2 to 15."

Just a few minor things here:

- "2 to 15" should be replaced by "2 to 14"

- "default data" should be replaced by "stock data"

- Due to harmonization, the python code on the following page might be adjusted, so that range(2, 15) is used instead of range(2, 14).

Note from the Author or Editor:
All suggestions confirmed.

Book text changed.

The output of the first line of x is incorrect and should be major_purchase instead of car.

> x
dti payment_inc_ratio home_ purpose_
1 1.00 2.39320 RENT major_purchase
2 5.55 4.57170 OWN small_business
3 18.08 9.71600 RENT other
4 10.08 12.21520 RENT debt_consolidation
5 7.06 3.90888 RENT other


gitlab code corrected

In the F-Statistics section :

"For the residuals, degrees of freedom is 20 (all observations can vary), and SS is the sum of squared difference between the individual observations and the treatment means. Mean squares (MS) is the sum of squares divided by the degrees of freedom."

Bruce, Peter; Bruce, Andrew; Gedeck, Peter. Practical Statistics for Data Scientists (Emplacements du Kindle 2760-2762). O'Reilly Media. Édition du Kindle.

When you run the ANOVA on R or Python, you have 16 for df in Residuals, not 20 !!

Note from the Author or Editor:
The text should read:

For the residuals, degrees of freedom is 16 (20 observations, 16 of which can vary after the the grand mean and the treatment means are set), and SS is the sum of squared difference between the individual observations and the treatment means. Mean squares (MS) is the sum of squares divided by the degrees of freedom.