A few small changes

book/book.tex

@@ -38,7 +38,7 @@
 
 \newcommand{\thetitle}{Think Stats}
 \newcommand{\thesubtitle}{Exploratory Data Analysis in Python}
-\newcommand{\theversion}{2.0.24}
+\newcommand{\theversion}{2.0.25}
 
 % these styles get translated in CSS for the HTML version
 \newstyle{a:link}{color:black;}
@@ -624,6 +624,8 @@ \section*{Contributor List}
 \item Nir Soffer sent several excellent pull requests for both the
   book and the supporting code.
 
+\item Joanne Pratt found a number that was off by a factor of 10.
+
 % ENDCONTRIB
 
 \end{itemize}
@@ -2876,7 +2878,7 @@ \section{Percentiles}
 \begin{verbatim}
 def Percentile2(scores, percentile_rank):
     scores.sort()
-    index = percentile_rank * (len(scores)-1) / 100
+    index = percentile_rank * (len(scores)-1) // 100
     return scores[index]
 \end{verbatim}
 
@@ -2904,16 +2906,16 @@ \section{CDFs}
 than or equal to $x$.
 
 Here's what that looks like as a function that takes a sequence,
-{\tt t}, and a value, {\tt x}:
+{\tt sample}, and a value, {\tt x}:
 %
 \begin{verbatim}
-def EvalCdf(t, x):
+def EvalCdf(sample, x):
     count = 0.0
-    for value in t:
+    for value in sample:
         if value <= x:
             count += 1
 
-    prob = count / len(t)
+    prob = count / len(sample)
     return prob
 \end{verbatim}
 
@@ -3777,21 +3779,15 @@ \section{The lognormal distribution}
 but this representation of the data does not make the difference
 particularly dramatic.  \index{respondent} \index{model}
 
-Figure~\ref{brfss_weight_normal} shows normal probability plots
-for adult weights, $w$, and for their logarithms, $\log_{10} w$.
-Now it is apparent that the data deviate substantially from the
-normal model.  The lognormal model is a good match for the data
-within a few standard deviations of the mean, but it deviates in
-the tails.  I conclude that the lognormal distribution is
-a good model for this data.
+Figure~\ref{brfss_weight_normal} shows normal probability plots for
+adult weights, $w$, and for their logarithms, $\log_{10} w$.  Now it
+is apparent that the data deviate substantially from the normal model.
+On the other hand, the lognormal model is a good match for the data.
 \index{normal distribution} \index{distribution!normal}
 \index{Gaussian distribution} \index{distribution!Gaussian}
 \index{lognormal distribution} \index{distribution!lognormal}
-\index{standard deviation}
-\index{adult weight}
-\index{weight!adult}
-\index{model}
-\index{normal probability plot}
+\index{standard deviation} \index{adult weight} \index{weight!adult}
+\index{model} \index{normal probability plot}
 
 
 \section{The Pareto distribution}
@@ -4356,7 +4352,7 @@ \section{Kernel density estimation}
 \begin{verbatim}
 >>> sample = [random.gauss(mean, std) for i in range(500)]
 >>> sample_pdf = thinkstats2.EstimatedPdf(sample)
->>> thinkplot.Pdf(pdf, label='sample KDE')
+>>> thinkplot.Pdf(sample_pdf, label='sample KDE')
 \end{verbatim}
 
 \verb"sample" is a list of 500 random heights.
@@ -4686,7 +4682,7 @@ \section{Moments}
 the mean, the moment of inertia of the spinning weights is the variance
 of the values.  If you are not familiar with moment of inertia, see
 \url{http://en.wikipedia.org/wiki/Moment_of_inertia}.  \index{moment
-  of inertia}.
+  of inertia}
 
 When you report moment-based statistics, it is important to think
 about the units.  For example, if the values $x_i$ are in cm, the
@@ -4765,7 +4761,7 @@ \section{Skewness}
 
 \begin{verbatim}
 def Median(xs):
-    cdf = thinkstats2.MakeCdfFromList(xs)
+    cdf = thinkstats2.Cdf(xs)
     return cdf.Value(0.5)
 
 def PearsonMedianSkewness(xs):
@@ -5488,8 +5484,8 @@ \section{Spearman's rank correlation}
 
 \begin{verbatim}
 def SpearmanCorr(xs, ys):
-    xranks = pandas.Series(xs)
-    yranks = pandas.Series(ys)
+    xs = pandas.Series(xs)
+    ys = pandas.Series(ys)
     return xs.corr(ys, method='spearman')
 \end{verbatim}
 
@@ -5993,7 +5989,7 @@ \section{Sampling distributions}
         xbar = np.mean(xs)
         means.append(xbar)
 
-    cdf = thinkstats2.MakeCdfFromList(means)
+    cdf = thinkstats2.Cdf(means)
     ci = cdf.Percentile(5), cdf.Percentile(95)
     stderr = RMSE(means, mu)
 \end{verbatim}
@@ -7162,12 +7158,12 @@ \section{Power}
 \end{verbatim}
 
 The result is about 70\%, which means that if the actual difference in
-mean pregnancy length is 0.78 weeks, we expect an experiment with this
+mean pregnancy length is 0.078 weeks, we expect an experiment with this
 sample size to yield a negative test 70\% of the time.
 \index{pregnancy length}
 
 This result is often presented the other way around: if the actual
-difference is 0.78 weeks, we should expect a positive test only 30\%
+difference is 0.078 weeks, we should expect a positive test only 30\%
 of the time.  This ``correct positive rate'' is called the {\bf power}
 of the test, or sometimes ``sensitivity''.  It reflects the ability of
 the test to detect an effect of a given size.
@@ -7176,7 +7172,7 @@ \section{Power}
 \index{correct positive}
 
 In this example, the test had only a 30\% chance of yielding a
-positive result (again, assuming that the difference is 0.78 weeks).
+positive result (again, assuming that the difference is 0.078 weeks).
 As a rule of thumb, a power of 80\% is considered acceptable, so
 we would say that this test was ``underpowered.''
 \index{underpowered}
@@ -8020,21 +8016,22 @@ \section{Weighted resampling}
 \index{resampling}
 
 \begin{verbatim}
-def ResampleRowsWeighted(df):
-    weights = df.finalwgt
-    pmf = thinkstats2.Pmf(weights.iteritems())
-    cdf = pmf.MakeCdf()
+def ResampleRowsWeighted(df, column='finalwgt'):
+    weights = df[column]
+    cdf = Cdf(dict(weights))
     indices = cdf.Sample(len(weights))
     sample = df.loc[indices]
     return sample
 \end{verbatim}
 
-{\tt pmf} maps from each row index to its normalized weight.  Converting
-to a Cdf makes the sampling process faster.  {\tt indices} is a sequence
-of row indices; {\tt sample} is a DataFrame that contains the selected
-rows.  Since we sample with replacement, the same row might appear
-more than once.
-\index{Cdf}
+{\tt weights} is a Series; converting it to a dictionary makes
+a map from the indices to the weights.  In {\tt cdf} the values
+are indices and the probabilities are proportional to the
+weights.
+
+{\tt indices} is a sequence of row indices; {\tt sample} is a
+DataFrame that contains the selected rows.  Since we sample with
+replacement, the same row might appear more than once.  \index{Cdf}
 \index{replacement}
 
 Now we can compare the effect of resampling with and without
@@ -8655,7 +8652,7 @@ \section{Data mining}
 
 I check that each explanatory variable has some variability; otherwise
 the results of the regression are unreliable.  I also check the number
-of observations for each model.  Variables that contains a large number
+of observations for each model.  Variables that contain a large number
 of {\tt nan}s are not good candidates for prediction.
 \index{explanatory variable}
 \index{NaN}

code/chap06soln.py

@@ -44,7 +44,7 @@
 cdf[mean] 0.856563066521
 
 With a higher upper bound, the moment-based skewness increases, as
-expected.  Surprisingly, the Person skewness goes down!  The reason
+expected.  Surprisingly, the Pearson skewness goes down!  The reason
 seems to be that increasing the upper bound has a modest effect on the
 mean, and a stronger effect on standard deviation.  Since std is in
 the denominator with exponent 3, it has a stronger effect on the

code/chap08soln.py

@@ -166,7 +166,7 @@ def VertLine(x, y=1):
     stderr = RMSE(estimates, lam)
     print('standard error', stderr)
 
-    cdf = thinkstats2.MakeCdfFromList(estimates)
+    cdf = thinkstats2.Cdf(estimates)
     ci = cdf.Percentile(5), cdf.Percentile(95)
     print('confidence interval', ci)
     VertLine(ci[0])

code/cumulative.py

@@ -57,27 +57,27 @@ def Percentile2(scores, percentile_rank):
           Percentile2(scores, percentile_rank))
 
 
-def EvalCdf(t, x):
+def EvalCdf(sample, x):
     """Computes CDF(x) in a sample.
 
-    t: sequence
+    sample: sequence
     x: value
 
     returns: cumulative probability
     """
     count = 0.0
-    for value in t:
+    for value in sample:
         if value <= x:
             count += 1.0
 
-    prob = count / len(t)
+    prob = count / len(sample)
     return prob
 
-t = [1, 2, 2, 3, 5]
+sample = [1, 2, 2, 3, 5]
 
 print('x', 'CDF(x)')
 for x in range(0, 7):
-    print(x, EvalCdf(t, x))
+    print(x, EvalCdf(sample, x))
 
 
 

code/estimation.py

@@ -130,7 +130,7 @@ def VertLine(x, y=1):
     stderr = RMSE(means, mu)
     print('standard error', stderr)
 
-    cdf = thinkstats2.MakeCdfFromList(means)
+    cdf = thinkstats2.Cdf(means)
     ci = cdf.Percentile(5), cdf.Percentile(95)
     print('confidence interval', ci)
     VertLine(ci[0])