benanne

import lasagne as nn
Conv2DLayer = nn.layers.Conv2DDNNLayer

def inception_module(l_in, num_1x1, reduce_3x3, num_3x3, reduce_5x5, num_5x5, gain=1.0, bias=0.1):
    """
    inception module (without the 3x3s1 pooling and projection because that's difficult in Theano right now)
    """
    shape = l_in.get_output_shape()
    out_layers = []

benanne

sedielem@koe:~/git/Theano/theano/sandbox/cuda/tests$ THEANO_FLAGS=warn_float64=ignore nosetests test_dnn.py
Using gpu device 1: Tesla K40c
..........
----------------------------------------------------------------------
Ran 10 tests in 218.306s

OK

benanne

sedielem@koe:~/git/Theano/theano/sandbox/cuda/tests$ nosetests test_dnn.py
Using gpu device 1: Tesla K40c
EEEEEESS.S
======================================================================
ERROR: test_conv (theano.sandbox.cuda.tests.test_dnn.TestDnnInferShapes)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/sedielem/git/Theano/theano/sandbox/cuda/tests/test_dnn.py", line 259, in test_conv
    dnn.GpuDnnConv
  File "/home/sedielem/git/Theano/theano/tests/unittest_tools.py", line 237, in _compile_and_check

benanne

In [1]: import theano
Using gpu device 0: GeForce GT 540M

In [2]: import theano.tensor as T

In [3]: y = T.tensor4('y')

In [4]: f = theano.function([y], y)

In [5]: theano.printing.debugprint(y)

benanne

class GalaxyDivNormLayer(nntools.layers.Layer):
    """
    rectification + divisive normalization
    """
    def __init__(self, input_layer):
        super(GalaxyDivNormLayer, self).__init__(input_layer)

        self.question_slices = [slice(0, 3), slice(3, 5), slice(5, 7), slice(7, 9), slice(9, 13), slice(13, 15),
                                slice(15, 18), slice(18, 25), slice(25, 28), slice(28, 31), slice(31, 37)]

benanne

import multiprocessing as mp

def buffered_gen_mp(source_gen, buffer_size=2):
    """
    Generator that runs a slow source generator in a separate process.
    buffer_size: the maximal number of items to pre-generate (length of the buffer)
    """
    if buffer_size < 2:
        raise RuntimeError("Minimal buffer size is 2!")

benanne

In [1]: import theano
iUsing gpu device 0: GeForce GTX 480
m
In [2]: import theano.tensor as T

In [3]: x, y = T.matrices('x', 'y')

In [4]: prod = T.dot(x, y)

In [5]: f = theano.function([x,y], prod)

benanne

* GeForce GTX 780Ti "Superclocked"
* drivers 340.24
* CUDA 6.5


sander@sander-precision:~/tmp/schluter/Theano$ for x in full valid subsample grads; do cuda-memcheck nosetests theano/sandbox/cuda/tests/test_conv_cuda_ndarray.py:test_gemm_$x; done
========= CUDA-MEMCHECK
Using gpu device 0: GeForce GTX 780 Ti
.
----------------------------------------------------------------------

benanne

* GeForce GTX 780Ti "Superclocked"
* drivers 340.24
* CUDA 6.0


sander@sander-precision:~/tmp/schluter/Theano$ for x in full valid subsample grads; do cuda-memcheck nosetests theano/sandbox/cuda/tests/test_conv_cuda_ndarray.py:test_gemm_$x; done
========= CUDA-MEMCHECK
Using gpu device 0: GeForce GTX 780 Ti
========= Invalid __global__ read of size 4
=========     at 0x000000e0 in sgemm_sm_heavy_nt_ldg

benanne

"""This tutorial introduces the LeNet5 neural network architecture
using Theano.  LeNet5 is a convolutional neural network, good for
classifying images. This tutorial shows how to build the architecture,
and comes with all the hyper-parameters you need to reproduce the
paper's MNIST results.


This implementation simplifies the model in the following ways:

 - LeNetConvPool doesn't implement location-specific gain and bias parameters