<s>
PyMC	B-Application
(	O
formerly	O
known	O
as	O
PyMC3	B-Application
)	O
is	O
a	O
Python	B-Language
package	O
for	O
Bayesian	O
statistical	O
modeling	O
and	O
probabilistic	O
machine	O
learning	O
which	O
focuses	O
on	O
advanced	O
Markov	B-General_Concept
chain	I-General_Concept
Monte	I-General_Concept
Carlo	I-General_Concept
and	O
variational	O
fitting	O
algorithms	O
.	O
</s>
<s>
It	O
is	O
a	O
rewrite	O
from	O
scratch	O
of	O
the	O
previous	O
version	O
of	O
the	O
PyMC	B-Application
software	O
.	O
</s>
<s>
Unlike	O
PyMC2	O
,	O
which	O
had	O
used	O
Fortran	B-Application
extensions	O
for	O
performing	O
computations	O
,	O
PyMC	B-Application
relies	O
on	O
PyTensor	O
,	O
a	O
Python	B-Language
library	O
that	O
allows	O
defining	O
,	O
optimizing	O
,	O
and	O
efficiently	O
evaluating	O
mathematical	O
expressions	O
involving	O
multi-dimensional	O
arrays	O
.	O
</s>
<s>
From	O
version	O
3.8	O
PyMC	B-Application
relies	O
on	O
ArviZ	B-Application
to	O
handle	O
plotting	O
,	O
diagnostics	O
,	O
and	O
statistical	O
checks	O
.	O
</s>
<s>
PyMC	B-Application
and	O
Stan	B-Application
are	O
the	O
two	O
most	O
popular	O
probabilistic	O
programming	O
tools	O
.	O
</s>
<s>
PyMC	B-Application
is	O
an	O
open	B-Application
source	I-Application
project	O
,	O
developed	O
by	O
the	O
community	O
and	O
fiscally	O
sponsored	O
by	O
NumFOCUS	O
.	O
</s>
<s>
After	O
Theano	B-Algorithm
announced	O
plans	O
to	O
discontinue	O
development	O
in	O
2017	O
,	O
the	O
PyMC	B-Application
team	O
evaluated	O
TensorFlow	B-Language
Probability	I-Language
as	O
a	O
computational	O
backend	O
,	O
but	O
decided	O
in	O
2020	O
to	O
take	O
over	O
the	O
development	O
of	O
Theano	B-Algorithm
.	O
</s>
<s>
Large	O
parts	O
of	O
the	O
Theano	B-Algorithm
codebase	O
have	O
been	O
refactored	O
and	O
compilation	O
through	O
JAX	O
and	O
Numba	B-Language
were	O
added	O
.	O
</s>
<s>
The	O
PyMC	B-Application
team	O
has	O
released	O
the	O
revised	O
computational	O
backend	O
under	O
the	O
name	O
PyTensor	O
and	O
continues	O
the	O
development	O
of	O
PyMC	B-Application
.	O
</s>
<s>
PyMC	B-Application
implements	O
non-gradient-based	O
and	O
gradient-based	O
Markov	B-General_Concept
chain	I-General_Concept
Monte	I-General_Concept
Carlo	I-General_Concept
(	O
MCMC	O
)	O
algorithms	O
for	O
Bayesian	O
inference	O
and	O
stochastic	O
,	O
gradient-based	O
variational	O
Bayesian	O
methods	O
for	O
approximate	O
Bayesian	O
inference	O
.	O
</s>
