Edward Chiao

WEEK 1

• https://docs.google.com/document/d/1Y6f7kvi5Xjl286vufHKggCmFnpPUQbWU6xq-9II-TdQ/edit?usp=sharing
  • Took notes on recurrent neural nets, read paper as assigned by Richard
  • Reviewed/learned some basic Tensorflow/Keras syntax and API
    • https://keras.io <- for future reference
    • https://www.tensorflow.org/tutorials/sequences/text_generation <- for future reference

WEEK 2

• Ran sample code from jazz chord progression to generate personal code
• Results:
  • 181761/181761 [==============================] - 1500s 8ms/step - loss: 3.8105
  • ()
  • ('----- diversity:', 0.8)
  • ----- Generating with seed:
  • G:sus4(b7) G:sus4(b7) C:maj7 C:maj7 G:sus4(b7) G:sus4(b7) C:maj7 C:maj7 G:sus4(b7) G:sus4(b7) C:sus4(b7)/5 C:sus4(b7)/5 D:maj/3 F:min6 E:min7 E:min7 A:sus4(b7) A:sus4(b7) D:min7 E:min7
  • Traceback (most recent call last):
  • File "main_lstm_realbook.py", line 150, in <module>
  • main(character_mode=False)
  • File "main_lstm_realbook.py", line 123, in main
  • next_index = sample(preds, diversity)
  • File "main_lstm_realbook.py", line 20, in sample
  • return np.argmax(np.random.multinomial(1, a, 1))
  • File "mtrand.pyx", line 4629, in mtrand.RandomState.multinomial
  • ValueError: sum(pvals[:-1]) > 1.0
    • this is with main(Character_mode) = False

Epoch 1/1

2019-02-08 01:19:17.525060: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA

1177081/1177081 [==============================] - 7511s 6ms/step - loss: 0.4461

()

('----- diversity:', 0.8)

----- Generating with seed:

C:maj/3 D:7/3 G:min

C:maj/3 D:7/3 G:min Traceback (most recent call last):

File "main_lstm_realbook.py", line 150, in <module>

main(character_mode=True)

File "main_lstm_realbook.py", line 123, in main

next_index = sample(preds, diversity)

File "main_lstm_realbook.py", line 20, in sample

return np.argmax(np.random.multinomial(1, a, 1)

File "mtrand.pyx", line 4629, in mtrand.RandomState.multinomial

ValueError: sum(pvals[:-1]) > 1.0

This was the result for main(Character_mode) = True

push

anyconnectVPN stuff

after running vpn...

`ssh eddy@weinberg-lenovo-1.music.gatech.edu`

enter "password" for password, and boom we're in

`conda activate deep` to create conda environment

Week 3

• Found a way to transfer repository from local to remote machine
• Ran new file that Richard sent on Github for text generation
• Ran 10 samples of the learning algorithm to generate a paragraph of text

Week 4

• Changed the repository code to run on chord progressions
  • Trained algorithm on the .txt file of chord progressions received from Richard
• Created 10 different samples with and without beam search from the algorithm

Week 5

• Manually generated chords on MuseScore based on the RNN-generated progressions
  • What is a better, potentially automated process for making this??

Week 6

• Create code that examines the .txt files
  • Make rules for all chords types: analyze root node and alternations
• Each chord has rules for the intervals of half steps, take a look at midi note values, maybe convert between the two

• I was able to write a script that takes in the chords generated by the RNN and give the specific notes of each chord
• Next steps: making the script perform on very specific chords that are fed in (ex: Fmaj7/2 or G#min(s5, s7), etc.) the code cannot yet take in such specific instructions, but is able to run on basic chords like 7ths, suspendeds, diminished, maj/min, etc.