theano

Version:: 1.0.5
Category:: ai
Cluster:: Loki

Author / Distributor

Description

Theano is a deep learning library and optimizing compiler for numerical computation in Python. It allows users to define, optimize, and evaluate mathematical expressions involving multi-dimensional arrays efficiently.

Version 1.0.5 is the final official release of Theano. It includes:

GPU acceleration via CUDA 10.2
Just-In-Time (JIT) compilation with GCC 8
NumPy integration for efficient scientific computing
Symbolic differentiation, loop optimization, and memory reuse
Compatibility with legacy deep learning frameworks (e.g., Lasagne, Blocks)

Documentation

theano.__version__        → "1.0.5"

Configuration:
----------------
$ THEANO_FLAGS=device=gpu,floatX=float32

Import and compile a function:
------------------------------
>>> import theano
>>> import theano.tensor as T
>>> x = T.dscalar('x')
>>> y = T.dscalar('y')
>>> z = x + y
>>> f = theano.function([x, y], z)
>>> f(2, 3)  # Returns 5.0

Config options:
  - device=[cpu|gpu]
  - floatX=[float32|float64]
  - optimizer=[fast_run|fast_compile|None]
  - mode=[FAST_RUN|FAST_COMPILE|DEBUG]

Help:
  >>> theano.config
  >>> theano.printing.pydotprint()

Examples/Usage

Load the module:

$ module load theano-py37-cuda10.2-gcc/1.0.5

Example Python session using GPU:

import theano
from theano import tensor as T

x = T.fscalar('x')
y = T.fscalar('y')
z = x * y
f = theano.function([x, y], z)

print(f(3.0, 4.0))  # Output: 12.0

Set GPU environment:

$ export THEANO_FLAGS=device=gpu,floatX=float32

Unload the module:

$ module unload theano-py37-cuda10.2-gcc/1.0.5

Installation

Source code is obtained from Theano