background
Our project is to develop a dialogue system, is to identify the specific requirements of the user to say, and then give the machine a user's voice feedback to achieve Doran dialogue. Requiring a simple fact, which is much to look at where we need to migrate our entire system to a dialogue on the hardware equipment, and small mobile version of the dialogue system. In the dialogue we have chosen the open-source management dialogue system rasa, the hardware we chose raspberry pie (support deep learning framework of python).
Hardware Configuration
Raspberry Pi 4B: voice module, python2.7 and python3.7, Raspberry Pi system (2019.07)
Pit abbreviated
In the raspberry come installed by far the most important thing is rasa corresponds to the version, release all your other libraries, such as python, tf, etc., are to be required as a benchmark version rasa (rasa will always be the boss). One of the most painful is to install a TensorFlow, rasa 1.4X need tf 1.15.0, rasa 1.3X need tf 1.14.0, rasa 1.2X need tf 1.13.1. Because Windows jacket over rasa1.3.9, so a direct pip to install, and always found a timeout (pit: ctm dregs in network speed, of course, has changed the source) or did not meet the TensorFlow1.14.0 version, simply download is not down (Do It is the source of the problem). I have this clever (in fact, stupid) Well then look for Raspberry Pi supported TensorFlow1.14.0, the results really have a version in line with the (obviously white written -tesorflow1.14.0-cp37-), excitement / happiness Download the first TensorFlow1.14 installed in, and then install rasa lengthy download process (of course, this process is very messed up), rasa download dozens of which depend on what OpenCV libraries like the slow speed and slag (mmp). Some twists and turns finally considered successfully installed (TF actually not being given, pit), happy to try rasa init, ~ I ~ Daily wrong, that my "policies" have a problem, this is not your brother initialization little chestnut ? How can wrong, (I stupidly changed the next policy configuration, but this is a big problem ah) through the identification of the group Friends of the major concluded that rasa environmental problem, well I think it is!
After the above I think TensorFlow and rasa no problem, compared to my Windows (why compare it, ooo, ooo), I found only python is not the same version, Windows above python3.6 and raspberry come in python 3.7, we do from success only to drop raspberry Pi version so close it? Think of all happy ~ drop drop drop, on the way there have been continuous advisory group of friends, by the way took my brother (hey hey hey, this simple to use linux have learned) meal operation, not been downloaded to compile the source code, compile whl file, change the configuration file ..... ultimately successful installed python3.6 can be used, but python3.6 the pip can not be used, and this is a big problem ah, no pip behind the library how to install it on python3.6 (meal operate flexible connection, or so)? ? ? Encountered during the foreign partner in the success of the Raspberry Pi tops Rasa , his configuration is TensorFlow1.14.0 and python3.6, as well as OpenCV installation, so that I firmly believe that victory is at hand. Play with blood continue gogogo, of course ...... or has failed.
Now I think it is not a problem Raspberry Pi board or directly burn a python3.6 into a mirror, so I'm looking for three board, 3B, 3B +, 4B, in fact, wanted to see if there is a default installed python3 above. 6 (I was stupid to start fooling themselves yet), down to the raspberry Pi version too difficult. I found two sub-python version is 3.5.3, but rasa1.2X minimum requirements python3.5.4 (Do not ask me why I do not see rasa1.X, because now rasa1.4.5 the function 1 is not too old?) , heaven ah ah earth, the first cry. So I downloaded all the recent raspberry pie image (I do not believe there is no python3.6), download, initialization, programming, know that this process be more pain now! And found to be all-python3.5.3 python2.7 and python2.7 and python3.7, for that is not py3.6. Then take a look at the problem is not the version of the child, to 3B and 3B + installation TensorFlow1.14.0 and rasa 1.3X, all so calm, finally policies in question (I turn back). Late at night, snow or, in the dormitory of the two of us brothers pale, sad face, depending on the silent, absolutely I really want to do the heavens! Brother said: first stood never know when inspiration will it? Me: Well, (hey hey hey, brother of expulsion)
In the next morning, I silently turned rasa version, see the version with all dependent libraries, suddenly I saw the figure below:
Because they do not know when, I remember rasa need TensorFlow1.15 and 1.14, it did not go rasa 1.2X. Because first of all, raspberry pie official website version TensorFlow1.14.0 meet python3.7, but also installed a foreign partner, as well as on Windows can think, so do not want to install the old, fear not take a lot of features. Ok! 1.2X version is okay, there is no better than that, of course, began to pip .............. installed, running can achieve my simple function, super happy, excited!
Installation
First of all should know what are the domestic sources, specific see my blog , download times out in time for the source.
Raspberry Pi system must choose a new point, at least ensure that there python3.7, but not recommended installed after July 2019, said that does not support voice module (specifically ask your seller X Bao little sister).
pip rasa time, now only rasa 1.2X can correspond TensorFlow1.13.1. Although corresponding rasa tf1.14.0 1.3, but that is a pit, do not go to step up.
Finally, the installation process may be a bit small problem, but the problem is to update some of the libraries, the next Baidu solved.
Better to use the seller's image, which has been installed a sound card driver (if there is a voice module) and so relatively easy, but we must note that I said above problems.
After experiencing not change the source or download download library, we follow the prompts to address direct download whl file, manually install.