# 1201 meeting note
###### tags: `meeting note`

- tasks
1. define input
- derive from other instruments midi
- [chromagram](https://en.wikipedia.org/wiki/Chroma_feature) of melodic instruments
- https://craffel.github.io/pretty-midi/#pretty_midi.PrettyMIDI.get_chroma
- drum pattern or other instruments’ onsets
- speed, number of instrument, time signature, beat position
- other infos
- style, rhythm type, composer, etc.
- need to find other source for these info
- [kaggle jazz-standard](https://www.kaggle.com/datasets/melihcanyardi/jazz-standards)
- [JSD dataset](https://jazzomat.hfm-weimar.de/dbformat/synopsis/solo1.html)
- input format
- chromagram 12x3
- speed (int)
- number of instrument (int)
- time signature (2 int)
- beat position (int)
- bar position (int)
- 
2. define output
- one input per bar, one output per bar (one to one)
- (time(16th note), midi num(+rest))
- one input per bar, one output per step(16th note) (one to many)
- midi num(+rest)
- one input per step, one output per step (one to one) ☑︎
- midi num(+rest)
3. define loss
- octave distance
- pitch distance
- simple cross-entropy?
4. find and preprocess data
- [thejazzpage](http://www.thejazzpage.de/index1.html)
- [minor9](https://bhs.minor9.com/midi/jazzstandards/)
- [Doug McKenzie Jazz](http://bushgrafts.com/wp/) (some)
- [Jazz ML ready MIDI](https://www.kaggle.com/datasets/saikayala/jazz-ml-ready-midi)
5. build basic model
- RNN LSTM (https://www.tensorflow.org/api_docs/python/tf/keras/layers/LSTM)
- RNN GRU (https://www.tensorflow.org/api_docs/python/tf/keras/layers/GRU)
- [tensorflow RNN tutorial](https://www.tensorflow.org/guide/keras/rnn)
- questions
- solo and accompany mode?
- other data resource
- https://github.com/wayne391/symbolic-music-datasets
- https://fourscoreandmore.org/musoRepo/
- https://groups.google.com/a/tensorflow.org/g/magenta-discuss/c/6ZLbzTjjpHM
- RNN in DL course
- https://nthu-datalab.github.io/ml/#lecture12