aloctavodia / statistical-rethinking-with-python-and-pymc3 Goto Github PK
View Code? Open in Web Editor NEWPython/PyMC3 port of the examples in " Statistical Rethinking A Bayesian Course with Examples in R and Stan" by Richard McElreath
Python/PyMC3 port of the examples in " Statistical Rethinking A Bayesian Course with Examples in R and Stan" by Richard McElreath
What about transferring this repository into an organization. Maybe inside pymc-devs? Together with other books/docs? Thoughts @junpenglao, @AustinRochford @twiecki
It might be helpful for contributors to have a conda environment file so they can quickly create a new environment and get things rolling. I have one I'll paste below, but I haven't checked it against all of the notebooks to make sure everything necessary is included and anything extraneous is not.
Here's mine:
name: pymc
channels:
- conda-forge/label/rc
- conda-forge
- defaults
dependencies:
- cycler=0.10.0=py36_0
- freetype=2.7=1
- joblib=0.11=py36_0
- libgpuarray=0.6.2=np112py36_0
- libpng=1.6.28=0
- mako=1.0.6=py36_0
- matplotlib=2.0.0=np112py36_3
- nose=1.3.7=py36_2
- pandas=0.19.2=np112py36_1
- patsy=0.4.1=py36_0
- pyparsing=2.2.0=py36_0
- pytz=2017.2=py36_0
- theano=0.9.0=py36_0
- tqdm=4.11.2=py36_0
- pymc3=3.1rc3=py36_0
- appnope=0.1.0=py36_0
- bleach=1.5.0=py36_0
- decorator=4.0.11=py36_0
- entrypoints=0.2.2=py36_1
- h5py=2.7.0=np112py36_0
- hdf5=1.8.17=1
- html5lib=0.999=py36_0
- ipykernel=4.6.1=py36_0
- ipython=5.3.0=py36_0
- ipython_genutils=0.2.0=py36_0
- jinja2=2.9.6=py36_0
- jsonschema=2.5.1=py36_0
- jupyter_client=5.0.1=py36_0
- jupyter_core=4.3.0=py36_0
- markupsafe=0.23=py36_2
- mistune=0.7.4=py36_0
- mkl=2017.0.1=0
- mkl-service=1.1.2=py36_3
- nbconvert=5.1.1=py36_0
- nbformat=4.3.0=py36_0
- notebook=5.0.0=py36_0
- numpy=1.12.1=py36_0
- openssl=1.0.2k=1
- pandocfilters=1.4.1=py36_0
- path.py=10.1=py36_0
- pexpect=4.2.1=py36_0
- pickleshare=0.7.4=py36_0
- pip=9.0.1=py36_1
- prompt_toolkit=1.0.14=py36_0
- ptyprocess=0.5.1=py36_0
- pygments=2.2.0=py36_0
- python=3.6.1=0
- python-dateutil=2.6.0=py36_0
- pyzmq=16.0.2=py36_0
- readline=6.2=2
- scipy=0.19.0=np112py36_0
- seaborn=0.7.1=py36_0
- setuptools=27.2.0=py36_0
- simplegeneric=0.8.1=py36_1
- six=1.10.0=py36_0
- sqlite=3.13.0=0
- terminado=0.6=py36_0
- testpath=0.3=py36_0
- tk=8.5.18=0
- tornado=4.4.2=py36_0
- traitlets=4.3.2=py36_0
- wcwidth=0.1.7=py36_0
- wheel=0.29.0=py36_0
- xz=5.2.2=1
- zlib=1.2.8=3
- pip:
- ipython-genutils==0.2.0
- jupyter-client==5.0.1
- jupyter-core==4.3.0
- prompt-toolkit==1.0.14
- pygpu==0.6.2
prefix: /Users/Peter/anaconda/envs/pymc
In 4.66, we set beta to generate two numbers for our quadratic model so both are generated from the same distribution. But the book generates it from two different distributions. Why would we do this? Doesn't this decrease the flexibility of the model?
with pm.Model() as m_4_5:
alpha = pm.Normal('alpha', mu=178, sd=100)
beta = pm.Normal('beta', mu=0, sd=10, shape=2) # shape of 2 generates two betas
sigma = pm.Uniform('sigma', lower=0, upper=50)
mu = pm.Deterministic('mu', alpha + beta[0] * d.weight_std + beta[1] * d.weight_std2)
height = pm.Normal('height', mu=mu, sd=sigma, observed=d.height)
trace_4_5 = pm.sample(1000, tune=1000)
Since almost all the chapter is ported (I will also work on Chapter 14), maybe it is good to
I know there are some functions from the rethinking package that we need to do without. I was wondering, however, if there is an implementation of postcheck
which plots posterior predictions that can be validated against actual data. This is very helpful when reviewing models to see if they fit the data well or not.
You can find the function on Richard's github page. It's used, for example, on page 306 of the book, code 10.27.
Moreso for my own personal tracking.
Need to update links in Pymc3 Main README to point to new repository for ipython notebooks
This is the proposal for code 6.12 in Python
import numpy as np
import scipy.stats as stats
import pymc3 as pm
import theano
def sim_train_test(N=20, k=3, rho=[0.15, -0.4], b_sigma=100):
n_dim = 1 + len(rho)
if n_dim < k:
n_dim = k
Rho = np.diag(np.ones(n_dim))
Rho[0, 1:3:1] = rho
i_lower = np.tril_indices(n_dim, -1)
Rho[i_lower] = Rho.T[i_lower]
x_train = stats.multivariate_normal.rvs(cov=Rho, size=N)
x_test = stats.multivariate_normal.rvs(cov=Rho, size=N)
mm_train = np.ones((N,1))
np.concatenate([mm_train, x_train[:, 1:k]], axis=1)
#Hacer modelo con pymc3
with pm.Model() as m_sim:
vec_V = pm.MvNormal('vec_V', mu=0, tau=b_sigma * np.eye(n_dim),
shape=(1, n_dim), testval=np.random.randn(1, n_dim)*.01)
mu = pm.Deterministic('mu', 0 + theano.tensor.dot(x_train, vec_V.T))
y = pm.Normal('y', mu=mu, sd=1, observed=x_train[:, 0])
trace_m_sim = pm.sample(1000, tune=1000)
vec = pm.summary(trace_m_sim)['mean'][:n_dim]
vec = np.array([i for i in vec]).reshape(n_dim, -1)
dev_train = - 2 * sum(stats.norm.logpdf(x_train, loc = np.matmul(x_train, vec), scale = 1))
mm_test = np.ones((N,1))
mm_test = np.concatenate([mm_test, x_test[:, 1:k +1]], axis=1)
dev_test = - 2 * sum(stats.norm.logpdf(x_test[:,0], loc = np.matmul(mm_test, vec), scale = 1))
# print('Deviation for training', '\t', 'Deviation for test')
# print(np.mean(dev_train), '\t', np.mean(dev_test))
return np.mean(dev_train), np.mean(dev_test)
n = 20
tries = 50
param = 5
r = np.zeros(shape=(param - 1, 4))
train = []
test = []
for j in range(2, param + 1):
print(j)
for i in range(1, tries):
tr, te = sim_train_test(N=n, k=param)
train.append(tr), test.append(te)
r[j -2, :] = np.mean(train), np.std(train, ddof=1), np.mean(test), np.std(test, ddof=1)
Unfortunately, it is slower than R version (it's very, very slow). I know this code doesn't have comments, and to undestand it you have to look this one. I will try to opmitize it (although I don't know how).
If anyone has any ideas, I'd like to hear them.
I've been working through the not yet released 2nd version of McElreath's book and have been using this repository to help guide me along the way. Its been extremely helpful, thanks! As of about ch.6 there is enough difference between the versions that it has become quite difficult to use. Any interest in getting a head start on the 2nd edition code translations so when he releases it, the code with be available in python as well? The 2nd edition can be found on his website w/ a password on one of the first 2019 youtube videos.
We need contributors to work on Chapter 8
We need contributors to work on Chapter 9
I think the environment name pymc
in environment.yml
is a bit confusing and may cause conflicts with existing environments of the user. A better name might be stat-rethink-pymc3
.
This will require changes in the README.md
and environment.yml
.
It seems to me that the Code 5.55, as it presently stands, does not reflect the intended "index variable" approach introduced in the book? Hence, I propose the following code:
with pm.Model() as m5_16_alt:
a = pm.Normal('a',mu = 0.6, sd=10, shape=len(d['clade_id'].unique()))
mu = pm.Deterministic('mu', a[d['clade_id'].values])
sigma = pm.Uniform('sigma', lower= 0 , upper= 10)
kcal_per_g = pm.Normal('kcal_per_g', mu = mu, sd=sigma, observed = d['kcal.per.g'])
trace_5_16_alt = pm.sample(1000, tune=1000)
Proposed code includes shape
parameter for the variable a
and uses index variable the way it was intended by the author of the book. Is my reasoning on this correct?
The summary produces following output (excerpt):
a:
Mean SD MC Error 89% HPD interval
-------------------------------------------------------------------
0.544 0.044 0.001 [0.482, 0.620]
0.713 0.044 0.001 [0.641, 0.781]
0.788 0.054 0.002 [0.709, 0.882]
0.506 0.059 0.002 [0.407, 0.593]
sigma:
Mean SD MC Error 89% HPD interval
-------------------------------------------------------------------
0.131 0.019 0.001 [0.101, 0.159]
Compare this output with the book (page 159).
If I run this cell:
rt = chimp_ensemble['pulled_left']
pred_mean = np.zeros((1000, 4))
cond = d.condition.unique()
prosoc_l = d.prosoc_left.unique()
for i in range(len(rt)):
tmp = []
for cp in cond:
for pl in prosoc_l:
tmp.append(np.mean(rt[i][(d.prosoc_left==pl) & (d.chose_prosoc==cp)]))
pred_mean[i, :] = tmp
ticks = range(4)
mp = pred_mean.mean(0)
hpd = pm.hpd(pred_mean)
plt.figure(figsize=(13,6))
plt.fill_between(ticks, hpd[:,1], hpd[:,0], alpha=0.25, color='k')
plt.plot(mp, color='k')
plt.xticks(ticks, ("0/0","1/0","0/1","1/1"))
chimps = d.groupby(['actor', 'prosoc_left', 'condition']).agg('mean')['pulled_left'].values.reshape(7, -1)
for i in range(7):
plt.plot(chimps[i], 'C0')
plt.ylim(0, 1.1);
I get this error:
---------------------------------------------------------------------------
IndexError Traceback (most recent call last)
<ipython-input-18-509a6637b6be> in <module>()
11 for cp in cond:
12 for pl in prosoc_l:
---> 13 tmp.append(np.mean(rt[i][(d.prosoc_left==pl) & (d.chose_prosoc==cp)]))
14 pred_mean[i, :] = tmp
15
IndexError: too many indices for array
which it is very strange. Fortunately, after an hour, I found a fix. Change the loop to this one:
for i in range(len(rt)):
tmp = []
if rt[i].shape == ():
continue
for cp in cond:
for pl in prosoc_l:
tmp.append(np.mean(rt[i][(d.prosoc_left==pl) & (d.chose_prosoc==cp)]))
pred_mean[i, :] = tmp
I might be missing something... but is there a chapter 1?
What about the homework solutions? Where are they?
Numba is imported but not used
The pm.compare
function is using a different syntax (I am using version 3.5). Instead of the old syntax which takes two lists for traces and models respectively, the new syntax takes a dictionary with model:trace as key-value pairs.
The old code in 6.24:
compare_df = pm.compare([trace_m6_11, trace_m6_12, trace_m6_13, trace_m6_14],
[m6_11, m6_12, m6_13, m6_14],
method='pseudo-BMA')
should be revised to
compare_df = pm.compare({m6_11: trace_m6_11,
m6_12: trace_m6_12,
m6_13: trace_m6_13,
m6_14: trace_m6_14},
method='pseudo-BMA')
Hello,
Thanks very much for putting this package together.
I tried to install with :
git clone https://github.com/aloctavodia/Statistical-Rethinking-with-Python-and-PyMC3.git
and
conda env create -f environment.yml
But I got a warning:
Warning: you have pip-installed dependencies in your environment file, but you do not list pip itself as one of your conda dependencies. Conda may not use the correct pip to install your packages, and they may end up in the wrong place. Please add an explicit pip dependency. I'm adding one for you, but still nagging you.
Everything progressed seemingly OK until:
Installing pip dependencies: \ Ran pip subprocess with arguments:
['/Users/me/opt/anaconda3/envs/stat-rethink-pymc3/bin/python', '-m', 'pip', 'install', '-U', '-r', '/Users/me/Dropbox/Coding/McElreath/Statistical-Rethinking-with-Python-and-PyMC3/condaenv.35uxzzws.requirements.txt']
Pip subprocess output:
Collecting git+git://github.com/pymc-devs/pymc3.git@master (from -r /Users/me/Dropbox/Coding/McElreath/Statistical-Rethinking-with-Python-and-PyMC3/condaenv.35uxzzws.requirements.txt (line 1))
Cloning git://github.com/pymc-devs/pymc3.git (to revision master) to /private/var/folders/w9/f07_8c6j6lx90vr9hksp9r3h0000gq/T/pip-req-build-t1jvdh8u
Pip subprocess error:
Running command git clone --filter=blob:none --quiet git://github.com/pymc-devs/pymc3.git /private/var/folders/w9/f07_8c6j6lx90vr9hksp9r3h0000gq/T/pip-req-build-t1jvdh8u
fatal: unable to connect to github.com:
github.com[0: 140.82.121.3]: errno=Operation timed out
error: subprocess-exited-with-error
× git clone --filter=blob:none --quiet git://github.com/pymc-devs/pymc3.git /private/var/folders/w9/f07_8c6j6lx90vr9hksp9r3h0000gq/T/pip-req-build-t1jvdh8u did not run successfully.
│ exit code: 128
╰─> See above for output.
note: This error originates from a subprocess, and is likely not a problem with pip.
error: subprocess-exited-with-error
× git clone --filter=blob:none --quiet git://github.com/pymc-devs/pymc3.git /private/var/folders/w9/f07_8c6j6lx90vr9hksp9r3h0000gq/T/pip-req-build-t1jvdh8u did not run successfully.
│ exit code: 128
╰─> See above for output.
note: This error originates from a subprocess, and is likely not a problem with pip.
failed
CondaEnvException: Pip failed
Would appreciate any help possible!
Cant understand this piece of code in 4.54 and 4.58:
weigth_seq = np.arange(25, 71)
# Given that we have a lot of samples we can use less of them for plotting (or we can use all!)
chain_N_thinned = chain_N[::10]
mu_pred = np.zeros((len(weigth_seq), len(chain_N_thinned)*chain_N.nchains))
for i, w in enumerate(weigth_seq):
mu_pred[i] = chain_N_thinned['alpha'] + chain_N_thinned['beta'] * w
What is chain_N[::10]? and chain_N.nchains doing? What is mu_pred supposed to be?
Why the posterior distribution of the sum of bl and br is not in the same scale (centered at 2.05) as shown in the book (top page 144)?
thanks,
The commented line for model m4_3 is missing a comma between 'mu' and alpha.
#mu = pm.Deterministic('mu' alpha + beta * d2.weight) # try uncomenting this line and comenting the above line
We need contributors to work on Chapter 12
I get the message saying
Sorry, something went wrong. Reload?
. I believe this happens for other Chapters as well.
We need contributors to work on Chapter 13
Just wondering if anybody dig into the code to see why WAIC return from the book and rethinking is sometimes larger than we computed using pm.compare
. It appears in some models but not the others. I dont think it is too much of a problem, as the orders return from both is the same.
We need contributors to work on Chapter 11
We need contributors to work on Chapter 10
We need contributors to work on Chapter 7
This is the proposal for code 6.14 in Python
num_param = np.arange(2, param + 1)
plt.figure()
plt.scatter(num_param, r[:, 0], color='C0')
plt.xticks(num_param)
for j in range(param - 1):
plt.vlines(num_param[j], r[j,0] - r[j, 1], r[j,0] + r[j,1], color='C0',
zorder=-1, alpha=0.50)
plt.scatter(num_param + 0.1, r[:, 2], facecolors='none', edgecolors='k')
for j in range(param - 1):
plt.vlines(num_param[j] + 0.1, r[j,2] - r[j, 3], r[j,2] + r[j,3], color='k',
zorder=-2, alpha=0.50)
dist = 0.20
plt.text(num_param[1] - dist, r[1, 0] - dist, 'in', color='C0')
plt.text(num_param[1] + dist, r[1, 2] - dist, 'out', color='k')
plt.text(num_param[1] + dist, r[1, 2] + r[1,3] - dist, '+1 SD', color='k', fontsize=9)
plt.text(num_param[1] + dist, r[1, 2] - r[1,3] - dist, '+1 SD', color='k', fontsize=9)
plt.xlabel('Number of parameters')
plt.ylabel('Deviance')
plt.title('N = {}'.format(n))
plt.show()
Remember from this that you have to provide: n
, the number of cases; tries
, the number of simulations; and param
, the number of parameters.
n = 20
tries = 50
param = 5
r = np.zeros(shape=(param - 1, 4))
train = []
test = []
for j in range(2, param + 1):
print(j)
for i in range(1, tries):
tr, te = sim_train_test(N=n, k=param)
train.append(tr), test.append(te)
r[j -2, :] = np.mean(train), np.std(train, ddof=1), np.mean(test), np.std(test, ddof=1)
Is there a way you'd like to track progress on this? I'd love to contribute, but don't want to duplicate a chapter someone is already working on.
I think it will be great to add the launch in azure notebook badge, like the one here: https://github.com/wesm/pydata-book
<a href="https://notebooks.azure.com/import/gh/aloctavodia/Statistical-Rethinking-with-Python-and-PyMC3"><img src="https://notebooks.azure.com/launch.png" /></a>
We need contributors to work on Chapter 14
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.