| |
Introduction
Theano recent days has been trying to build deep learning development environment and install the NVIDIA CUDA Toolkit. During this period, there are thought to build on Windows, but it is understood from the Internet, or find more appropriate in the Linux environment. In the development environment to build this process, in fact, a lot of the pit, a little twists and turns. The build process is not going to write, because the Internet does have a lot of good installation procedures, as well as the official website also has installation tutorial. But, in view of the pit has been for several days, or record it, long memory :)
During installation Ubuntu 14.04 more than four times (Reason: Some software depends on the problem persists, replace the software source seemingly can not but today after the software source into the main server, it is normal, but my wireless web speed is relatively slow, so. after the primary server, the download speed obviously a lot slower.).
Try to install Fedora operating system once, because also thought on this system build environment, because some of the above always rely on Ubuntu 14.04 can not handle. But because the card driver problem, still can not networking; networking can not, naturally, can not install the compiler, the compiler does not, can not install the NIC driver (because of the need to compile drivers), and finally gave up this platform.
Try to install CentOS operating system 1, in fact, did not successfully installed, because my graphics card driver problem, causing all text graphics installer interface are vague, and even "Exit" button in the text I do not recognize out, and finally gave up.
Desktop main parameters
Memory: Kingston 8G
Processor: Intel Core i5-4590 CPU @ 3.3GHz x 4
Graphics: NVIDIA Geforce GTX 750 (GPU parallel computing)
Hard Drive: Western Digital 1TB
Install Ubuntu 14.04 operating system
Make Ubuntu boot disk (floppy disk through the use of written to the U disk);
When desktop starts, select the quick start, start the Ubuntu live cd system from U disk;
Install Ubuntu 14.04 operating system;
Restart the system after installation;
Start the system when entering Grub interface, press e key to enter the edit page, in the penultimate line, ro quiet splash Add nomodeset behind, so after entering the system will not only drive significant problems caused by a blank screen;
After entering the system, edit the file /boot/grub/grub.cfg files, search ro quiet splash keywords, the same additional nomodeset, before such systems do not repeat the steps above every time you start up.
Theano installation environment
Tools require pre-installed
These tools may be used to later, such as compilers.
sudo apt-get install -y python-dev python-pip python-nose gcc g ++ git gfortran
Installation BLAS, LAPACK, ATLAS
These packages must be installed prior to installing Numpy and Scipy
sudo apt-get install -y libopenblas-dev liblapack-dev libatlas-base-dev
Install NumPy
Using the apt-get in the way of official tutorial installed, but it looks like I did after installing the unit does not pass the test, there are a lot of errors generated. Especially later installed SciPy even more so. And I may not have the tools you need before you install the complete package it has a relationship.
# Install NumPy, the installation process will have to wait because of the need to compile, so you have to spend a little time.
# Actually installed behind SciPy and Theano will take the time to compile, the entire installation process or to quite a long time.
sudo pip install numpy
# carry out testing.
python -c 'import numpy; numpy.test ()'
Must ensure that the above test passes, and then continue to the next step, make sure that every step is correct. Errors will not pass the test. Test results are as follows.
Installation SciPy
# Installation method, patiently waiting for it to compile :)
sudo pip install scipy
# carry out testing
python -c 'import scipy; scipy.test ()'
We must also ensure scipy normal pass the test, so that it can continue to install the following Theano. |
|
|
|