Skip to content

Commit c0dbe06

Browse files
ferrineferrine
committed
restructure docs
1 parent fdf0011 commit c0dbe06

File tree

1 file changed

+81
-75
lines changed

1 file changed

+81
-75
lines changed

‎pymc_experimental/distributions/multivariate.py

Lines changed: 81 additions & 75 deletions
Original file line numberDiff line numberDiff line change
@@ -127,92 +127,98 @@ def R2D2M2CP(
127127
--------
128128
Here are arguments explained in a synthetic example
129129
130-
>>> import pymc_experimental as pmx
131-
>>> import pymc as pm
132-
>>> import numpy as np
133-
>>> X = np.random.randn(10, 3)
134-
>>> b = np.random.randn(3)
135-
>>> y = X @ b + np.random.randn(10) * 0.04 + 5
136-
>>> with pm.Model(coords=dict(variables=["a", "b", "c"])) as model:
137-
... eps, beta = pmx.distributions.R2D2M2CP(
138-
... "beta",
139-
... y.std(),
140-
... X.std(0),
141-
... dims="variables",
142-
... # NOTE: global shrinkage
143-
... r2=0.8,
144-
... # NOTE: if you are unsure about r2
145-
... r2_std=0.2,
146-
... # NOTE: if you know where a variable should go
147-
... # if you do not know, leave as 0.5
148-
... positive_probs=[0.8, 0.5, 0.1],
149-
... # NOTE: if you have different opinions about
150-
... # where a variable should go.
151-
... # NOTE: if you put 0.5 previously,
152-
... # just put 0.1 there, but other
153-
... # sigmas should work fine too
154-
... positive_probs_std=[0.3, 0.1, 0.2],
155-
... # NOTE: variable importances are relative to each other,
156-
... # but larget numbers put "more" weight in the relation
157-
... # use
158-
... # * 1-10 for small confidence
159-
... # * 10-30 for moderate confidence
160-
... # * 30+ for high confidence
161-
... # EXAMPLE:
162-
... # "a" - is likely to be useful
163-
... # "b" - no idea if it is useful
164-
... # "c" - a must have in the relation
165-
... variables_importance=[10, 1, 34],
166-
... # NOTE: try both
167-
... centered=True
168-
... )
169-
... intercept = y.mean()
170-
... obs = pm.Normal("obs", intercept + X @ beta, eps, observed=y)
130+
.. code-block:: python
131+
132+
import pymc_experimental as pmx
133+
import pymc as pm
134+
import numpy as np
135+
X = np.random.randn(10, 3)
136+
b = np.random.randn(3)
137+
y = X @ b + np.random.randn(10) * 0.04 + 5
138+
with pm.Model(coords=dict(variables=["a", "b", "c"])) as model:
139+
eps, beta = pmx.distributions.R2D2M2CP(
140+
"beta",
141+
y.std(),
142+
X.std(0),
143+
dims="variables",
144+
# NOTE: global shrinkage
145+
r2=0.8,
146+
# NOTE: if you are unsure about r2
147+
r2_std=0.2,
148+
# NOTE: if you know where a variable should go
149+
# if you do not know, leave as 0.5
150+
positive_probs=[0.8, 0.5, 0.1],
151+
# NOTE: if you have different opinions about
152+
# where a variable should go.
153+
# NOTE: if you put 0.5 previously,
154+
# just put 0.1 there, but other
155+
# sigmas should work fine too
156+
positive_probs_std=[0.3, 0.1, 0.2],
157+
# NOTE: variable importances are relative to each other,
158+
# but larget numbers put "more" weight in the relation
159+
# use
160+
# * 1-10 for small confidence
161+
# * 10-30 for moderate confidence
162+
# * 30+ for high confidence
163+
# EXAMPLE:
164+
# "a" - is likely to be useful
165+
# "b" - no idea if it is useful
166+
# "c" - a must have in the relation
167+
variables_importance=[10, 1, 34],
168+
# NOTE: try both
169+
centered=True
170+
)
171+
intercept = y.mean()
172+
obs = pm.Normal("obs", intercept + X @ beta, eps, observed=y)
171173
172174
There can be special cases by choosing specific set of arguments
173175
174176
Here the prior distribution of beta is ``Normal(0, y.std() * r2 ** .5)``
175177
176-
>>> with pm.Model(coords=dict(variables=["a", "b", "c"])) as model:
177-
... eps, beta = pmx.distributions.R2D2M2CP(
178-
... "beta",
179-
... y.std(),
180-
... X.std(0),
181-
... dims="variables",
182-
... # NOTE: global shrinkage
183-
... r2=0.8,
184-
... # NOTE: if you are unsure about r2
185-
... r2_std=0.2,
186-
... # NOTE: if you know where a variable should go
187-
... # if you do not know, leave as 0.5
188-
... centered=False
189-
... )
190-
... intercept = y.mean()
191-
... obs = pm.Normal("obs", intercept + X @ beta, eps, observed=y)
178+
.. code-block:: python
179+
180+
with pm.Model(coords=dict(variables=["a", "b", "c"])) as model:
181+
eps, beta = pmx.distributions.R2D2M2CP(
182+
"beta",
183+
y.std(),
184+
X.std(0),
185+
dims="variables",
186+
# NOTE: global shrinkage
187+
r2=0.8,
188+
# NOTE: if you are unsure about r2
189+
r2_std=0.2,
190+
# NOTE: if you know where a variable should go
191+
# if you do not know, leave as 0.5
192+
centered=False
193+
)
194+
intercept = y.mean()
195+
obs = pm.Normal("obs", intercept + X @ beta, eps, observed=y)
192196
193197
194198
It is fine to leave some of the ``_std`` arguments unspecified.
195199
You can also specify only ``positive_probs``, and all
196200
the variables are assumed to explain same amount of variance (same importance)
197201
198-
>>> with pm.Model(coords=dict(variables=["a", "b", "c"])) as model:
199-
... eps, beta = pmx.distributions.R2D2M2CP(
200-
... "beta",
201-
... y.std(),
202-
... X.std(0),
203-
... dims="variables",
204-
... # NOTE: global shrinkage
205-
... r2=0.8,
206-
... # NOTE: if you are unsure about r2
207-
... r2_std=0.2,
208-
... # NOTE: if you know where a variable should go
209-
... # if you do not know, leave as 0.5
210-
... positive_probs=[0.8, 0.5, 0.1],
211-
... # NOTE: try both
212-
... centered=True
213-
... )
214-
... intercept = y.mean()
215-
... obs = pm.Normal("obs", intercept + X @ beta, eps, observed=y)
202+
.. code-block:: python
203+
204+
with pm.Model(coords=dict(variables=["a", "b", "c"])) as model:
205+
eps, beta = pmx.distributions.R2D2M2CP(
206+
"beta",
207+
y.std(),
208+
X.std(0),
209+
dims="variables",
210+
# NOTE: global shrinkage
211+
r2=0.8,
212+
# NOTE: if you are unsure about r2
213+
r2_std=0.2,
214+
# NOTE: if you know where a variable should go
215+
# if you do not know, leave as 0.5
216+
positive_probs=[0.8, 0.5, 0.1],
217+
# NOTE: try both
218+
centered=True
219+
)
220+
intercept = y.mean()
221+
obs = pm.Normal("obs", intercept + X @ beta, eps, observed=y)
216222
"""
217223
if not isinstance(dims, (list, tuple)):
218224
dims = (dims,)

0 commit comments

Comments
 (0)