google-magenta-midi-music-on-linux-hello-world

Bash install scripts to get Google Magenta working easily on a linux ubuntu computer. Magenta will allow you to convert a folder with several of your midi files to an Artificial Intelligence tensorflow generated set of output midi files.

[![Instructional video at] (http://img.youtube.com/vi/RbF_7LQJ9Go/0.jpg)] (https://youtu.be/RbF_7LQJ9Go?list=PL57Dnr1H_egvXDT3NIne7YQY54Rx8fS6J)

Update Aug 1st, 2016

Some of the bash files have been changed to keep up with the changes on the magenta repository. The steps are:

basically open a terminal (right click-->open Terminal) and git clone this repo

git clone https://github.com/hpssjellis/google-magenta-midi-music-on-linux-hello-world.git

Then go into the repository and run setup.sh (use auto fill so you don't have to type the long name)

cd google-magenta-midi-music-on-linux-hello-world

bash setup.sh

To activate the other bash files you need to migrate to the magenta main folder

 cd ~/mymagenta/magenta

Then activate the other bash files. example

bash a01-rnn_basic.sh

July 17, 2016

Due to changes in Magenta, these scripts have been updated. They are slightly different than the video, but work along the same idea.

June 19th, 2016

Youtube Video at

https://youtu.be/RbF_7LQJ9Go

or at the playlist at https://youtu.be/RbF_7LQJ9Go?list=PL57Dnr1H_egvXDT3NIne7YQY54Rx8fS6J

[![Instructional video at] (http://img.youtube.com/vi/RbF_7LQJ9Go/0.jpg)] (https://youtu.be/RbF_7LQJ9Go?list=PL57Dnr1H_egvXDT3NIne7YQY54Rx8fS6J)

June 17, 2016

Just tried this on a fresh ubuntu version 14.04 server 64 bit installation (I had the CD lying around and an old computer to put it on) and it worked fine except for tensorboard since I did not install LAMP.

Also had to install "git" on a fresh install so I included the bash file a03-extra-installs.sh which installs git and timidity (command line midi player). To run it enter the command

bash a03-extra-installs.sh

June 15, 2016

I think I fixed it, the following line below needed to be changed to

bazel run //magenta/models:basic_rnn_create_dataset --

Don't think it is in the Docs yet.

June 14, 2016 (evening) Darn. Murphy's Law. The day I tweet that this site is up and running https://github.com/tensorflow/magenta changed something that broke my script! For some reason this code does not work. I know lib/encoders.py has been renamed to lib/basic_one_hot_encoder_test.py but that should not be relevant. The script can't seem to find convert_sequences_to_melodies

# TFRecord file containing NoteSequence protocol buffers from convert_midi_dir_to_note_sequences.py.
SEQUENCES_TFRECORD=/tmp/notesequences.tfrecord

# TFRecord file that TensorFlow's SequenceExample protos will be written to. This is the training dataset.
TRAIN_DATA=/tmp/training_melodies.tfrecord

# Optional evaluation dataset. Also, a TFRecord file containing SequenceExample protos.
EVAL_DATA=/tmp/evaluation_melodies.tfrecord

# Fraction of input data that will be written to the eval dataset (if eval_output flag is set).
EVAL_RATIO=0.10

# Name of the encoder to use. See magenta/lib/encoders.py.
ENCODER=basic_one_hot_encoder


bazel run //magenta:convert_sequences_to_melodies -- 
--input=$SEQUENCES_TFRECORD 
--train_output=$TRAIN_DATA 
--eval_output=$EVAL_DATA 
--eval_ratio=$EVAL_RATIO 
--encoder=$ENCODER

June 14, 2016 (morning)

This works, still testing it. No idea if it will work on your computer but it does work on my ubuntu version 15.10 laptop

basically open a terminal (right click-->open Terminal) and git clone this repo

git clone https://github.com/hpssjellis/google-magenta-midi-music-on-linux-hello-world.git

Then go into the repository and run setup.sh (use auto fill so you don't have to type the long name)

cd google-magenta-midi-music-on-linux-hello-world

bash setup.sh

Theoretically that should be it. setup.sh installs into a home folder called mymagenta a folder called Magenta which is the main workspace, bit confusing since it has a folder inside it also called magenta. The script also installs bazel and tensorflow. Sets up some symlinks in the magenta workspace, builds a few extra folders. If all goes well it runs the bash file a01-helloworld.sh which runs a combined set of the example programs with only 200 interactions instead of 20000. Then it finally activates tensorboard

Your final 5 midi examples end up in the /tmp/basic_rnn_generated folder and then copy the midi files to however you want to play them. I installed timidity to play the midi files but have now started using https://onlinesequencer.net/

Then start using your own midi files and your own primer.mid file and change variables and iterations. I would like to make another helloWorld that tidies up where all these temporary files go, presently it is a mess for beginners