All 8 entries tagged Tensorflow
No other Warwick Blogs use the tag Tensorflow on entries | View entries tagged Tensorflow at Technorati | There are no images tagged Tensorflow on this blog
March 05, 2020
list of submitted projects and information about late submission penalties
Follow-up to Submission date and instructions for your project work from Rudo's blog
Dear all,
thus far I have received submissions from (10/02/30 10:00)
- Jettawat Khantaveramongkol
- Juan Roman Roche
- Hendrik Schlueter
- Alexios Malaquias
- DanielPepin Fowan
- Burak Civitcioglu
- Roman Perrier (one day late)
- Jeremy Cadran (two days late)
Please note that I intend to apply late submission penalties, following roughly similar Warwick penalties which penalize a late submission with a 5% reduction in final points for every day late. A late day starts in the first minute after the dealine and the time of submission follows from the time of the last upload.
Regards,
-Rudo
March 02, 2020
PS: don't worry too much about maximizing epochs
Follow-up to Submission date and instructions for your project work from Rudo's blog
Just to add: do not worry too much about running your code to the maximum of epochs. It is clear to us that most people do not have access to state of the art graphics cards or hundreds of processors. What is more important is to see that you came up with a good idea where machine learning can be used, executed the right commands to get some training and analysis done, and found the right data to train on. In fact, we migth even try to run some of your code for more epochs to see how results can get better!
Submission date and instructions for your project work
Follow-up to teaching schedule, now ready from Rudo's blog
Dear all,
you should now all be close to finishing your project work for the machine learning course. I hope it has all gone well and that you had some fun exploring machine learning. The official date for submission is
Friday, March 6 at 17:00.
For submission, please remember that we want you to submit a running machine learning code using a jupyter notebook with comments that explain step by step what you did and why. Please also supply the data so that Djena and I can run the code ourselves. The easiest way to send us the submission is via
https://files.warwick.ac.uk/rroemer/sendto
where you can upload both jupyter notebook and data (preferably as a single .zip file). If your data files are too big for upload there (I have 4.5GB in total on that upload drive), please send a link at which we can then download the data. Please remember that it is your responsibility to make sure data and notebook work "out of the box", i.e. without us having to change anything before they work, including directories.
We aim to have feedback for you by the end of March and will communicate this by email to each of you (so please also let me have your preferred email address in your jupyter notebook).
Cheers!
December 12, 2019
teaching schedule, now ready
Follow-up to Machine Learning – a hands–on course from Rudo's blog
Dear all,
the schedule of the course has now become clearer: we continue to teach every Monday 0930-1230 in room E4.28 (except for one day in January and one day in February) for 10 weeks overall. For the 1st 5 weeks (with 4 more weeks to come), I will actually "teach", i.e. with prepared lectures for 2 hours 0930-1130, while Djenabou will be available for 1 hr more 1130-1230 to help you with the exercises. Then in the last 5 weeks, Djenabou will be available 0930-1230 to help with your project work (she won't do it for you, but might be able to help with technical problems).
A full schedule should be available soon.
See you next Monday!
-Rudo
December 09, 2019
Thoughts on the 1st lecture session
Follow-up to Installing your own TensorFlow/Kears + FastAI on Ubuntu from Rudo's blog
Dear all,
[1] today in the lecture, we saw that it was not entirely simple to install your own ML machine with Tensorflow/Keras and FastAI. I have now made a blog entry on how to do this. You can find it here.
[2] In addition, when running the FastAI code, we saw that it might take a long time even for just the pets categorization. I just ran the pets notebook (please download a new version) and have made the pets-stage-1 and pets-stage-1-50 trained models available on the download area. You need to move these models into where FastAI keeps them which in my set-up is
/home/phsht/.fastai/data/oxford-iiit-pet/images/models
Just copy the twp .pth files there and you should be able to run the pets code after the
learn.load('pets-stage-1')
fastai commands in the notebook.
[3] When you "play" with the notebook called "FastAI-lesson2-download.ipynb", note that the file data.zip contains the cats/teddys/wolves that I managed to download and use for the exercise.
[4] Some of you argued that your own machines may not be able to do meaningful ML as you do not have a GPU. Have a look at what the FastAI guys have collected in terms of possible free GPU instances, i.e. Using a GPU. I have not yet used any of these services, but would be keen to learn from your experiences!
OK, keep checking this blog for up-to-date information before we meet again next week.
Installing your own TensorFlow/Kears + FastAI on Ubuntu
Follow-up to Things to do before the 1st lectures from Rudo's blog
How to make your own Keras/FastAI ML-capable machine using Ubuntu
Install the Anaconda-version of Python
Navigate to https://docs.anaconda.com/anaconda/install/ to find the links to download your own Anaconda. For Ubuntu (19.10), choose Installing on Linux as shown in the figure below.
Next, click on Anaconda installer for Linux to start the download for the Python 3.7 version. Having downloaded, presumably to the directory ~/Downloads, navigate there with a termial. Then you should perhaps make the file executable, i.e., in the terminal, type
chmod u+x Anaconda3-2019.10-Linux-x86_64.sh
Then in the same terminal, you can execute the installer
./Anaconda3-2019.10-Linux-x86_64.sh
This will install Anaconda into a directory of your choice in your $HOME space. Feel free to change the installation directory. At the end, the installer will ask if you want to allow it to activate anaconda automatically. Say yes to this.
Next, edit your .bashrc file in your home directory and add "conda deactivate" after the lines which the installer added as shown here
Save the change and restart your terminal. You should now have a working Python installation based on Anaconda.
Note that such an installation may overwrite a previous python installation, so please be careful to make sure this is really what you want.t
Activate your new anaconda installation as
conda activate
and you should see
Make conda environments to avoid version clashes
You will rarely need Keras and FastAI to work simultaneously. And the python packages they need may have clashing versions. In order to avoid this, we found it useful to define "environments" as follows
Tensorflow/Kears:
conda create -n TF anaconda
FastAI:
conda create -n FastAI anaconda
When you type this in a terminal, you are being asked to install some Python packages. Simply say yes/accept.
Installing Tensorflow+Keras
Open a terminal, then start the "TF" environment by typing
conda activate TF
You will see that the command prompt is now prepended with the name of the environment, i.e. (TF).
Next, let's install Tensorflow.
In the same terminal with the TF environment, type
conda install -c conda-forge tensorflow
(see https://anaconda.org/conda-forge/tensorflow for other possibilities). This installs Tensorflow. If you are having a GPU in your system, then replacing tensorflow with tensorflow-gpu will install the required GPU compatible codes.
Installing Keras
Similarly, install Keras via
conda install -c conda-forge keras
(see https://anaconda.org/conda-forge/keras for other possibilities). Keras is now installed.
There are other ways to install the combination, but they can lead to other requirements to run the code. If you follow the way we did it here, the dependencies are set in the way that we are using for the jupyter notebooks to run.
Installing FastAI
This installation is actually simpler. As before, remember to activate the correct environment, i.e.
conda activate FastAI
If you stick to the names "TF" and "FastAI" for the course, then you are compatible with what we are doing. Hopefully, this makes it easier to avoid any mistakes. OK, now let's install FastAI!
conda install -c fastai fastai
(see https://anaconda.org/fastai/fastai).
Testing the installation for Keras
Let's see if everything worked as advertised. Start a jupyter notebook in the TF environment,
conda activate TF
jupyter notebook &
Then in the Jupyter notebook, navigate to our course material and load, e.g., Keras-MNIST.ipynb. As usual with Jupyter notebooks, evaluate the first cell (via SHIFT-ENTER if using the keyboard). If everything works well, you will get
and "Using Tensoflow backend" is telling you that Keras has correctly identified how to use Tensorflow.
Testing for a correct FastAI installation
As before, start the jupyter notebook in the correct (FastAI) environment, i.e.
conda activate FastAI
jupyter notebook &
Now use the notebook to open a FastAI code, i.e. FastAI-lesson1-pets.ipynb. if the following lines execute correctly, your FastAI installation works well!
And that's really it! Except it's not. Both Keras and FastAI make use of some further Python packages that you might need to install as well. Here is a, perhaps incomplete, list: mathplotlib, sklearn, ...
December 06, 2019
Things to do before the 1st lectures
Follow-up to Machine Learning – a hands–on course from Rudo's blog
Dear all,
[1] some of you will have registered for the short Machine Learning (ML) course that we are planning to give, starting next week. ML is best "learned" by doing it, so we are planning to use the computer room E4.28 for exercises during all the lectures. But we think you might also find it more useful if you were to bring your laptops along and install+run the ML codes there.
[2] We are expecting to run all the ML on a Ubuntu installation (Francois Chollet, one of the key people in the field, says "it’s possible to use Windows, too, but I don’t recommend it"). Other Linux installations should also work. If you only have Windows, Rudo recommends installing
VirtualBox, https://www.virtualbox.org/
with a Ubuntu 19.10 installation (http://releases.ubuntu.com/19.10/).
[3] During the course, we intend to offer you downloadable ML source codes via Rudo's blog at
https://blogs.warwick.ac.uk/rroemer/entry/machine_learning_a/
where you can also see a copy of this email. The direct download link is in the first entry of the blog and given here as well, i.e.
https://cloud.csc.warwick.ac.uk/index.php/s/JomfpoMj87X3Jbk
[4] We want to share literature on ML with you using Mendeley, i.e. https://www.mendeley.com/. Although the Mendeley Desktop might be convenient to install (we have it on our machines), the online interface could be ok for the course. But in order to "share" PDFs with you, you need to have a (free) Mendeley account. Please create one.
[5] Last, we, Djena and Rudo, are both new to ML, so this will be a course where we will all explore ML together! We hope you'll enjoy it.
- Rudo and Djena
December 05, 2019
Machine Learning – a hands–on course
Dear all,
for this course, you should be able to find all course material for download at
https://cloud.csc.warwick.ac.uk/index.php/s/JomfpoMj87X3Jbk
Please let me know in case this does not work!
-Rudo