class Generator:
def __init__(self):
self.music = Music()
def write_arffs(self):
for next_file in os.listdir("midi"):
if os.path.isfile("midi/" + next_file):
print "adding midi/" + next_file
self.music.load_song("midi/" + next_file)
print "generating arff files"
self.music.write_arff("test")
def train(self):
print "loading arff files"
instrument_data, note_data, velocity_data, duration_data, time_delta_data = self.music.read_arff(
"test")
self.transform_instrument = preprocessing.OneHotEncoder(categorical_features=[0,1,2,5,6,7,10,11,12,15,16,17,20,21,22])
self.transform_instrument.fit(instrument_data[:, :-1])
self.transform_note = preprocessing.OneHotEncoder(categorical_features=[0,1,2,5,6,7,10,11,12,15,16,17,20,21,22])
self.transform_note.fit(note_data[:, :-1])
self.transform_duration = preprocessing.OneHotEncoder(categorical_features=[0,1,2,5,6,7,10,11,12,15,16,17,20,21,22])
self.transform_duration.fit(duration_data[:, :-1])
self.transform_time_delta = preprocessing.OneHotEncoder(categorical_features=[0,1,2,5,6,7,10,11,12,15,16,17,20,21,22])
self.transform_time_delta.fit(time_delta_data[:, :-1])
print "training instruments"
self.instrument_clf = RandomForestClassifier(n_estimators=10)
self.instrument_clf = self.instrument_clf.fit(
self.transform_instrument.transform(instrument_data[:, :-1]), instrument_data[:, -1:])
print "training notes"
self.note_clf = RandomForestRegressor(n_estimators=10)
print note_data[:, :-1]
self.note_clf = self.note_clf.fit(self.transform_note.transform(note_data[:, :-1]), note_data[:, -1:])
print "training velocity"
self.velocity_clf = RandomForestClassifier(n_estimators=10)
self.velocity_clf = self.velocity_clf.fit(
velocity_data[:, :-1], velocity_data[:, -1:])
print "training duration"
self.duration_clf = RandomForestClassifier(n_estimators=10)
self.duration_clf = self.duration_clf.fit(
self.transform_duration.transform(duration_data[:, :-1]), duration_data[:, -1:])
print "training time_delta"
self.time_delta_clf = RandomForestClassifier(n_estimators=10)
self.time_delta_clf = self.time_delta_clf.fit(
self.transform_time_delta.transform(time_delta_data[:, :-1]), time_delta_data[:, -1:])
def generate_song(self, filename):
self.song = Song()
self.song.add_note(self.get_seed_note())
self.song.add_note(self.get_seed_note())
print "generating new song"
for _ in range(5000):
notes = self.song.get_last_notes()
notes = np.array(notes).reshape(1, -1)
instrument = self.instrument_clf.predict(self.transform_instrument.transform(notes))[0]
note = self.note_clf.predict(self.transform_note.transform(notes))[0]
velocity = self.velocity_clf.predict(notes)[0]
note_duration = self.duration_clf.predict(self.transform_duration.transform(notes))[0] * 2
time_delta = self.time_delta_clf.predict(self.transform_time_delta.transform(notes))[0] * 0.75
self.song.add_note(Note(int(instrument), int(note), int(
velocity), int(time_delta), duration=int(note_duration)))
print "writing song to " + filename
self.song.write_file(filename)
def get_seed_note(self):
return Note(0, random.randrange(0, 126), 100, 0, duration=random.randrange(1, 10) * 10)
class Generator:
def __init__(self):
self.music = Music()
def write_arffs(self):
for next_file in os.listdir("midi"):
if os.path.isfile("midi/" + next_file):
print "adding midi/" + next_file
self.music.load_song("midi/" + next_file)
print "generating arff files"
self.music.write_arff("test")
def train(self):
print "loading arff files"
instrument_data, note_data, velocity_data, duration_data, time_delta_data = self.music.read_arff(
"test")
self.transform_instrument = preprocessing.OneHotEncoder(
categorical_features=[
0, 1, 2, 5, 6, 7, 10, 11, 12, 15, 16, 17, 20, 21, 22
])
self.transform_instrument.fit(instrument_data[:, :-1])
self.transform_note = preprocessing.OneHotEncoder(
categorical_features=[
0, 1, 2, 5, 6, 7, 10, 11, 12, 15, 16, 17, 20, 21, 22
])
self.transform_note.fit(note_data[:, :-1])
self.transform_duration = preprocessing.OneHotEncoder(
categorical_features=[
0, 1, 2, 5, 6, 7, 10, 11, 12, 15, 16, 17, 20, 21, 22
])
self.transform_duration.fit(duration_data[:, :-1])
self.transform_time_delta = preprocessing.OneHotEncoder(
categorical_features=[
0, 1, 2, 5, 6, 7, 10, 11, 12, 15, 16, 17, 20, 21, 22
])
self.transform_time_delta.fit(time_delta_data[:, :-1])
print "training instruments"
self.instrument_clf = RandomForestClassifier(n_estimators=10)
self.instrument_clf = self.instrument_clf.fit(
self.transform_instrument.transform(instrument_data[:, :-1]),
instrument_data[:, -1:])
print "training notes"
self.note_clf = RandomForestRegressor(n_estimators=10)
print note_data[:, :-1]
self.note_clf = self.note_clf.fit(
self.transform_note.transform(note_data[:, :-1]), note_data[:,
-1:])
print "training velocity"
self.velocity_clf = RandomForestClassifier(n_estimators=10)
self.velocity_clf = self.velocity_clf.fit(velocity_data[:, :-1],
velocity_data[:, -1:])
print "training duration"
self.duration_clf = RandomForestClassifier(n_estimators=10)
self.duration_clf = self.duration_clf.fit(
self.transform_duration.transform(duration_data[:, :-1]),
duration_data[:, -1:])
print "training time_delta"
self.time_delta_clf = RandomForestClassifier(n_estimators=10)
self.time_delta_clf = self.time_delta_clf.fit(
self.transform_time_delta.transform(time_delta_data[:, :-1]),
time_delta_data[:, -1:])
def generate_song(self, filename):
self.song = Song()
self.song.add_note(self.get_seed_note())
self.song.add_note(self.get_seed_note())
print "generating new song"
for _ in range(5000):
notes = self.song.get_last_notes()
notes = np.array(notes).reshape(1, -1)
instrument = self.instrument_clf.predict(
self.transform_instrument.transform(notes))[0]
note = self.note_clf.predict(
self.transform_note.transform(notes))[0]
velocity = self.velocity_clf.predict(notes)[0]
note_duration = self.duration_clf.predict(
self.transform_duration.transform(notes))[0] * 2
time_delta = self.time_delta_clf.predict(
self.transform_time_delta.transform(notes))[0] * 0.75
self.song.add_note(
Note(int(instrument),
int(note),
int(velocity),
int(time_delta),
duration=int(note_duration)))
print "writing song to " + filename
self.song.write_file(filename)
def get_seed_note(self):
return Note(0,
random.randrange(0, 126),
100,
0,
duration=random.randrange(1, 10) * 10)
