juri-marcucci / casi_python

Python code for Computer Age Statistical Inference

CASI_Python

Python code for Computer Age Statistical Inference by Bradley Efron and Trevor Hastie.

Required packages

• numpy
• pandas
• statsmodels
• matplotlib
• scipy
• mpmath (for section 3.1)
• sklearn (scikit-learn)
• graphviz (for section 8.4)

This code was developed with Python 3.7.

Data

The data can be obtained from the book site. The notebooks assume that the data has been saved in a data/ subdirectory.

Possible errata or differences with the book

• Section 2.1: I get slightly different values for Table 2.1
• Section 6.2, Table 6.3: I get different values for $\hat{sd}$. I get the same values when I drop $x$ from the exponent in Eq 6.21. See ch06s02.ipynb.
• Section 7.1, eq 7.12: Is $\hat{B}$ missing a $\sigma^2$?
• Section 7.2: I think eq 7.23 should actually be $\displaystyle \hat{p}^{\text JS}_i = \frac{1}{n}\left[(n + 0.75) \sin^2 \left(\frac{\hat{\mu}^{\text JS}_i}{2 \sqrt{n + 0.5}}\right) - 0.375 \right] $
• Section 7.4: Eq 7.46 is missing "1/1000", i.e., it isn't taking the mean
• Section 8.3:
  • The data page says the values in the galaxy dataset are log-redshift, but they are likely log-log-redshift, if 1.22 \leq r \leq 3.32 as equation 8.38 says
  • Equation 8.38 says 17.2 \leq m \leq 21.5, but the y-axis on figure 8.5 shows negative values, approximately -21 \leq m \leq -17
• Section 10.4, pg 171: Equation 10.55 might give the impression that the powers of the bin counts are used in the model (i.e., that the $\hat{\beta}$ values are coefficients from the model). In fact, orthogonal polynomials are used. See ch10s04.ipynb.
• Section 20.1, pg 396: The book says that "X has been standardized so that each of its columns has mean 0 and sum of squares 1", but in fact I think it has been standardized so that standard deviation is 1, not sum of squares; see ch20s01.ipynb
• Section 20.2:
  • pg 402: The book cites equation 12.51 for the Cp estimates, but unlike (12.51), the values in Table 20.2 were not divided by the number of observations; see ch20s02.ipynb.
  • pg 404, 4th line: very minor typo: change "carred" to "carried"
• Section 20.4, pg 415: Equation (20.54), the $\beta$ should probably have a superscript $(b)$

To do

• Section 4.3: Figure 4.2
• Section 5.2: Figures 5.2 and 5.3 ?
• Section 6.3
• Section 8.4: Figure 8.7 doesn't completely match the book
• Section 10.5: Infinitesimal jackknife / influence function s.e. for Table 10.2
• Section 20.3
• Section 20.4

Contributing

Corrections and contributions are welcome. Please check the section below before submitting contributions.

DO NOT contribute

Do not contribute anything that solves a homework exercise or makes an exercise trivial.

For chapters 1-10 and 20, here is a list of things to stay away from. For other chapters, check with the posted homework problems.

• Section 4.1: the calculation of equation 4.10. This is homework exercise 4.1a. This also rules out contributing Figure 4.1
• Section 4.4: Figure 4.3. This would make homework exercise 4.5 trivial.
• Section 7.2: entire section. Calculating the JS estimates is exercise 7.3
• Section 7.3: sd values in Table 7.3. This is exercise 7.7a
• Section 7.4: entire section, probably. The section mainly uses JS estimates, and calculating those estimates is exercise 7.3
• Section 8.1: Figure 8.2, Table 8.1, and Table 8.2. These are covered by exercises 8.1 and 8.3.
• Section 8.3: entire section, probably; covered by exercise 8.6
• Section 9.2: Figure 9.2; would probably make exercise 9.3 trivial
• Section 9.3: Table 9.4? Might make exercise 9.4 too easy.
• Section 9.5: Table 9.8; covered by exercise 9.7
• Section 10.2: Figure 10.2; covered by exercise 10.3