Script to test my GpuJoin.c_code for Theano. See https://github.com/shackenberg/Theano/compare/adding_GpuJoin.c_code

GpuJoin.c_code.test.py

#!/usr/bin/env python
"""
call this to run the code:
export  CUDA_LAUNCH_BLOCKING=1
THEANO_FLAGS=mode=FAST_RUN,device=gpu,floatX=float32,exception_verbosity=high python jointest.py
"""
 
import theano
import theano.tensor as T
from numpy import *


axis = 0
np_array1 = array(random.random([2]), dtype='float32')
np_array2 = array(random.random([5]), dtype='float32')
 
T_vector1 = T.vector('T_vector1')
T_vector2 = T.vector('T_vector2')
T_result = T.concatenate([T_vector1, T_vector2], axis)
f1 = theano.function([T_vector1, T_vector2], T_result , profile=True)  
numpy_result = concatenate([np_array1, np_array2], axis)
theano_result = f1(np_array1, np_array2)
if not allclose(theano_result,numpy_result):
    print numpy_result
    print theano_result
   

axis = 1
np_array1 = array(random.random([2, 3]), dtype='float32')
np_array2 = array(random.random([2, 11]), dtype='float32')
np_array3 = array(random.random([2, 3]), dtype='float32')
 
T_matrix1 = T.matrix('T_matrix1')
T_matrix2 = T.matrix('T_matrix2')
T_matrix3 = T.matrix('T_matrix3')
T_result = T.concatenate([T_matrix1, T_matrix2, T_matrix3], axis)
f1 = theano.function([T_matrix1, T_matrix2, T_matrix3], T_result , profile=True)  
numpy_result = concatenate([np_array1, np_array2, np_array3], axis)
theano_result = f1(np_array1, np_array2, np_array3)
if not allclose(theano_result,numpy_result):
    print numpy_result
    print theano_result


   

axis = 2
np_array1 = array(random.random([2, 3, 2, 4]), dtype='float32')
np_array2 = array(random.random([2, 3, 11, 4]), dtype='float32')
np_array3 = array(random.random([2, 3, 44, 4]), dtype='float32')
np_array4 = array(random.random([2, 3, 1, 4]), dtype='float32')
 
T_tensor_1 = T.tensor4('T_tensor3_1')
T_tensor_2 = T.tensor4('T_tensor3_2')
T_tensor_3 = T.tensor4('T_tensor3_3')
T_tensor_4 = T.tensor4('T_tensor3_4')

T_result = T.concatenate([T_tensor_1, T_tensor_2, T_tensor_3, T_tensor_4], axis)

f1 = theano.function([T_tensor_1, T_tensor_2, T_tensor_3, T_tensor_4], T_result , profile=True)  
numpy_result = concatenate([np_array1, np_array2, np_array3, np_array4], axis)
theano_result = f1(np_array1, np_array2, np_array3, np_array4)
if not allclose(theano_result,numpy_result):
    print numpy_result
    print theano_result