Lincoln Index - Three parameter model

[zjachc]

I am receiving the following error when I attempt to compute the likelihood in the three parameter model:
for N, p0, p1 in joint3_pmf.index:

ValueError: not enough values to unpack (expected 3, got 2)

I think the issue is that when I attempt to use make_joint using p1 and the joint2_pmf, I do not get the triplet, but rather only the pair, mean p1 is not attaching and thus the multi-index only has 2 items instead of three.

Do you have any suggestions about how best to fix?

Thank you

[zjachc]

Here is the entire script leading up to the problem:

Priors

params N, p0, p1, reuse prior_N

qs = np.linspace(0, 1, num=51)
prior_p0 = make_uniform(qs, name='p0')
prior_p1 = make_uniform(qs, name='p1')

Assemble into joint prior w/ 3-dim

put first two into DataFrame

joint2 = make_joint(prior_p0, prior_N)
joint2.shape

Stack and put result into a Pmf

joint2_pmf = Pmf(joint2.stack())
joint2_pmf.head(3)

Use make_joint again to add in 3rd param

joint3 = make_joint(prior_p1, joint2_pmf)
joint3.shape

joint3.head(3)

result is a DataFrame w/ values of N & p0 in a multi-index that goes down the rows

and values of p1 in an index that goes across the columns

apply stack again

joint3_pmf = Pmf(joint3.stack())
joint3_pmf.head(3)

joint3_pmf.shape

Loop that computes the likelihoods

likelihood = joint3_pmf.copy()
observed = data.sum()
x = data.copy()

for N, p0, p1 in joint3_pmf.index:
x[0] = N - observed
y = compute_probs(p0, p1)
likelihood[N, p0, p1] = multinomial.pmf(x, N, y)

[AllenDowney]

Hi. To make sure I understand the question, are you reporting an error in my code or asking for help debugging?

[zjachc]

Hi - I used to code as it was written in the book and received an error; however, I'm not sure if there is a code error or a different issue and wanted to see if you had a suggestion as to why the code won't work as written.

[AllenDowney]

Here's the notebook where you can run the code (which works for me):

https://colab.research.google.com/github/AllenDowney/ThinkBayes2/blob/master/notebooks/chap15.ipynb

In the notebook you can print some of the intermediate values, or check the shape of the arrays/DataFrames. That should help with debugging.

[zjachc]

Shapes of the arrays match what was expected in the book. As do the heads up until joint3_pmf. At that point, the triplet doesn't show up, only a pair. It seems there is some issue with make_joint, Pmf, and stack, which has worked fine until now.

[AllenDowney]

When you run the code in the notebook I linked, does it work?

[zjachc]

Yes, it does. I understand the problem is on my end.

Do you have any thoughts as to why this issue emerged in this instance of the three-parameter, but never before?

[AllenDowney]

If you display joint3_pmf.index, does it have three columns?

[zjachc]

Yes - here's the first row:

MultiIndex([((0.0, 0.0), 32)

[zjachc]

MultiIndex([((0.0, 0.0), 32),
((0.0, 0.0), 37),
((0.0, 0.0), 42),
((0.0, 0.0), 47),
((0.0, 0.0), 52),
((0.0, 0.0), 57),
((0.0, 0.0), 62),
((0.0, 0.0), 67),
((0.0, 0.0), 72),
((0.0, 0.0), 77),
...
((1.0, 1.0), 302),
((1.0, 1.0), 307),
((1.0, 1.0), 312),
((1.0, 1.0), 317),
((1.0, 1.0), 322),
((1.0, 1.0), 327),
((1.0, 1.0), 332),
((1.0, 1.0), 337),
((1.0, 1.0), 342),
((1.0, 1.0), 347)],
length=166464)

[AllenDowney]

Yup, that's a sequence of pairs, where the first of each pair is a pair, so that's not the same as a sequence of triples. So either there's a difference between your code and mine or a difference between your environment and mine. I think it's unlikely to be an environment issue, so I'd look for a difference in the code.

[zjachc]

Thank you Allen - I will go through the code again. After running into the issue, I re-wrote the code and still had the problem and then copied it directly, but still no joy.

[AllenDowney]

Maybe put it in a Colab notebook. That would eliminate the environmental explanation. In the process, you might find the problem. If not, you could share the notebook with me.

[zjachc]

I will do that.

[zjachc]

Here's the script:

Three parameter model

Priors

params N, p0, p1, reuse prior_N

prior_N

qs = np.arange(32, 350, step=5)
prior_N = make_uniform(qs, name='N')
prior_N.head(3)

prior_p0, prior_p1

qs = np.linspace(0, 1, num=51)
prior_p0 = make_uniform(qs, name='p0')
prior_p1 = make_uniform(qs, name='p1')

prior_p0.head(3)
prior_p1.head(3)

Assemble into joint prior w/ 3-dim

put first two into DataFrame

joint2 = make_joint(prior_p0, prior_N)
joint2.shape
joint2.head(10)

Stack and put result into a Pmf

joint2_pmf = Pmf(joint2.stack())
joint2_pmf.head(10)

Use make_joint again to add in 3rd param

joint3 = make_joint(prior_p1, joint2_pmf)
joint3.shape

joint3.head(10)

result is a DataFrame w/ values of N & p0 in a multi-index that goes down the rows

and values of p1 in an index that goes across the columns

apply stack again

joint3_pmf = Pmf(joint3.stack())
joint3_pmf.head(10)
joint3_pmf.index

joint3_pmf.shape

Loop that computes the likelihoods

likelihood = joint3_pmf.copy()
observed = data.sum()
x = data.copy()

from scipy.stats import multinomial

for N, p0, p1 in joint3_pmf.index:
x[0] = N - observed
y = compute_probs(p0, p1)
likelihood[N, p0, p1] = multinomial.pmf(x, N, y)

[zjachc]

It's an environment issue, works in Jupyter but not Spyder unfortunately.

[zjachc]

What's interesting is that this does not work in Jupyter but does work in Spyder.

The following was copied directly - no modification.

Result is that the contour plot won't render. However, Spyder did not have this issue.

observed = k01 + k10 + k11

for N, p in joint_pmf.index:
k00 = N - observed
x = [k00, k01, k10, k11]
q = 1-p
y = [qq, qp, pq, pp]
likelihood[N, p] = multinomial.pmf(x, N, y)

---

TypeError Traceback (most recent call last)
Input In [24], in <cell line: 3>()
6 q = 1-p
7 y = [qq, qp, pq, pp]
----> 8 likelihood[N, p] = multinomial.pmf(x, N, y)

TypeError: 'numpy.float64' object does not support item assignment