def trainClassifier(m, realFilesDir, genFilesDir):
# Reading real music and generating features for training
numSamples = 100
realFiles = multi_training.loadPieces(realFilesDir)
i,o = zip(*[MT.getPieceSegment(realFiles) for _ in range(numSamples)])
i = numpy.array(i)
realFeat = []
for row in range(0,len(i)):
realFeat.append(i[row,:].flatten())
realFeat = np.array(realFeat)
# Reading generated music and generating features for training
genFiles = multi_training.loadPieces(genFilesDir)
i,o = zip(*[MT.getPieceSegment(genFiles) for _ in range(numSamples)])
i = numpy.array(i)
genFeat = []
for row in range(0,len(i)):
genFeat.append(i[row,:].flatten())
genFeat = np.array(genFeat)
# Fitting classifier to training data
dataTrain = np.concatenate((realFeat, genFeat), axis = 0)
labelsTrain = np.array([1] * numSamples + [0] * numSamples)
classifier = SVC()
classifier.fit(dataTrain, labelsTrain)
return classifier
def gen_adaptive(m,
pcs,
times,
keep_thoughts=False,
name=multi_training.ofile):
xIpt, xOpt = map(lambda x: numpy.array(x, dtype='int8'),
multi_training.getPieceSegment(pcs))
all_outputs = [xOpt[0]]
if keep_thoughts:
all_thoughts = []
m.start_slow_walk(xIpt[0])
cons = 1
for time in range(multi_training.batch_len * times):
resdata = m.slow_walk_fun(cons)
nnotes = numpy.sum(resdata[-1][:, 0])
print(cons)
if nnotes < 3:
if cons > 1:
cons = 1
cons -= 0.1
else:
cons += (1 - cons) * 0.1
all_outputs.append(resdata[-1])
if keep_thoughts:
all_thoughts.append(resdata)
noteStateMatrixToMidi(numpy.array(all_outputs), 'output/' + name)
if keep_thoughts:
pickle.dump(all_thoughts, open('output/' + name + '.p', 'wb'))
def gen_adaptive(m, pcs, times, keep_thoughts=False, name="final", rbm=None):
xIpt, xOpt = map(lambda x: np.array(x, dtype="int8"), multi_training.getPieceSegment(pcs))
all_outputs = [xOpt[0]]
if keep_thoughts:
all_thoughts = []
m.start_slow_walk(xIpt[0])
cons = 1
for time in range(multi_training.batch_len * times):
resdata = m.slow_walk_fun(cons)
nnotes = np.sum(resdata[-1][:, 0])
if nnotes < 2:
if cons > 1:
cons = 1
cons -= 0.02
else:
cons += (1 - cons) * 0.3
all_outputs.append(resdata[-1])
if keep_thoughts:
all_thoughts.append(resdata)
from RBM import rbm_train, rbm_reconstruct
if not rbm:
rbm = rbm_train(
np.array(pcs.values()), window_len=16, learning_rate=0.1, training_epochs=1, batch_size=20, n_hidden=1500
)
rbm_outputs, rbm_outputs_discrete = rbm_reconstruct(rbm, np.array(all_outputs), window_len=16)
# pickle.dump(np.array(all_outputs), open('output/composition.sm', 'wb'))
noteStateMatrixToMidi(np.array(rbm_outputs), "output/" + name + " rbm")
noteStateMatrixToMidi(np.array(rbm_outputs_discrete), "output/" + name + " discrete rbm")
noteStateMatrixToMidi(np.array(all_outputs), "output/" + name)
print ("Saved generated music and reconstruction.")
if keep_thoughts:
pickle.dump(all_thoughts, open("output/" + name + ".p", "wb"))
def gen_adaptive(m,pcs,times,keep_thoughts=False,name="final"):
pcs = multi_training.loadPieces("./music/單一小節資料夾")
xIpt, xOpt = map(lambda x: numpy.array(x, dtype='int8'), multi_training.getPieceSegment(pcs))
all_outputs = [xOpt[0]]
if keep_thoughts:
all_thoughts = []
m.start_slow_walk(xIpt[0])
cons = 1
def test_predict_one_step_model():
pcs = multi_training.loadPieces("Train_data")
data, _ = multi_training.getPieceSegment(pcs)
data = torch.Tensor(data[0])
print(data.shape)
model = m.BiaxialRNNModel([300, 300], [100, 50])
print(model)
out = model(data) # Done in order to print
def test_predict_n_step_model():
pcs = multi_training.loadPieces("Train_data")
data, _ = multi_training.getPieceSegment(pcs)
data = torch.Tensor(data[0])
print(data.shape)
model = m.BiaxialRNNModel([300, 300], [100, 50])
print(model)
out = model(data, 5)
print(out)
print(torch.Tensor(np.array(out)).shape)
def classifyGenerated:
testingDirs = [f for f in os.listdir('/home/ubuntu/test/output/') if re.match(r'gen*', f)]
numSamples = 100
for ind in range(len(testingDirs)):
genPcsTrain = multi_training.loadPieces(testingDirs[ind])
i,o = zip(*[MT.getPieceSegment(genPcsTrain) for _ in range(numSamples)])
i = numpy.array(i)
genFeat = []
for row in range(0,len(i)):
genFeat.append(i[row,:].flatten())
genFeat = np.array(genFeat)
pred_label = classifier.predict(genFeat)
accuracy = sum([pred_label[i] == np.array([1] * len(genFeat)) for i in range(len(genFeat))]) / len(genFeat) * 100
print testingDirs[ind], " ", accuracy
def gen_adaptive(m, pcs, times, name="final"):
xIpt, xOpt = map(lambda x: numpy.array(x, dtype="int8"), multi_training.getPieceSegment(pcs))
all_outputs = [xOpt[0]]
m.start_slow_walk(xIpt[0])
cons = 1
for time in range(multi_training.batch_len * times):
resdata = m.slow_walk_fun(cons)
nnotes = numpy.sum(resdata[-1][:, 0])
if nnotes < 2:
if cons > 1:
cons = 1
cons -= 0.02
else:
cons += (1 - cons) * 0.3
all_outputs.append(resdata[-1])
noteStateMatrixToMidi(numpy.array(all_outputs), "/home/azureuser/hackzurich/output/" + name)
def gen_adaptive(m, pcs, times, keep_thoughts=False, name="final", rbm=None):
xIpt, xOpt = map(lambda x: np.array(x, dtype='int8'),
multi_training.getPieceSegment(pcs))
all_outputs = [xOpt[0]]
if keep_thoughts:
all_thoughts = []
m.start_slow_walk(xIpt[0])
cons = 1
for time in range(multi_training.batch_len * times):
resdata = m.slow_walk_fun(cons)
nnotes = np.sum(resdata[-1][:, 0])
if nnotes < 2:
if cons > 1:
cons = 1
cons -= 0.02
else:
cons += (1 - cons) * 0.3
all_outputs.append(resdata[-1])
if keep_thoughts:
all_thoughts.append(resdata)
from RBM import rbm_train, rbm_reconstruct
if not rbm:
rbm = rbm_train(np.array(pcs.values()),
window_len=16,
learning_rate=0.1,
training_epochs=1,
batch_size=20,
n_hidden=1500)
rbm_outputs, rbm_outputs_discrete = rbm_reconstruct(rbm,
np.array(all_outputs),
window_len=16)
#pickle.dump(np.array(all_outputs), open('output/composition.sm', 'wb'))
noteStateMatrixToMidi(np.array(rbm_outputs), 'output/' + name + ' rbm')
noteStateMatrixToMidi(np.array(rbm_outputs_discrete),
'output/' + name + ' discrete rbm')
noteStateMatrixToMidi(np.array(all_outputs), 'output/' + name)
print('Saved generated music and reconstruction.')
if keep_thoughts:
pickle.dump(all_thoughts, open('output/' + name + '.p', 'wb'))
def gen_adaptive(m,pcs,times,keep_thoughts=False,name="final"):
xIpt, xOpt = map(lambda x: numpy.array(x, dtype='int8'), multi_training.getPieceSegment(pcs))
all_outputs = [xOpt[0]]
if keep_thoughts:
all_thoughts = []
m.start_slow_walk(xIpt[0])
cons = 1
for time in range(multi_training.batch_len*times):
resdata = m.slow_walk_fun( cons )
nnotes = numpy.sum(resdata[-1][:,0])
if nnotes < 2:
if cons > 1:
cons = 1
cons -= 0.02
else:
cons += (1 - cons)*0.3
all_outputs.append(resdata[-1])
if keep_thoughts:
all_thoughts.append(resdata)
noteStateMatrixToMidi(numpy.array(all_outputs),'output/'+name)
if keep_thoughts:
pickle.dump(all_thoughts, open('output/'+name+'.p','wb'))
# Loading model at 4500 epochs to generate music for training and testing
m.learned_config = pickle.load(open("params4500.p", "rb"))
for i in range(1,15):
main.gen_adaptive(m,pcs,10,name="beeth_train_gen_"+str(i))
for i in range(1,15):
main.gen_adaptive(m,pcs,10,name="beeth_gen"+str(i))
# Loading model at 8000 epochs to generate music for testing
m.learned_config = pickle.load(open("params8000.p", "rb"))
for i in range(1,15):
main.gen_adaptive(m,pcs,10,name="beeth_gen"+str(i))
# Reading real music and generating features for training
realPcs = multi_training.loadPieces("../DataSet/liszt")
i,o = zip(*[MT.getPieceSegment(realPcs) for _ in range(100)])
i = numpy.array(i)
realFeat = []
for row in range(0,len(i)):
realFeat.append(i[row,:].flatten())
realFeat = np.array(realFeat)
# Reading generated music and generating features for training
genPcsTrain = multi_training.loadPieces("../test_data/gen_train")
i,o = zip(*[MT.getPieceSegment(genPcsTrain) for _ in range(100)])
i = numpy.array(i)
genFeat = []
for row in range(0,len(i)):
genFeat.append(i[row,:].flatten())
genFeat = np.array(genFeat)
for time in range(三小節):
nnotes = numpy.sum(resdata[-1][:,0])
if nnotes < 2:
if cons > 1:
cons = 1
cons -= 0.02
else:
cons += (1 - cons)*0.3
all_outputs.append(resdata[-1])
if keep_thoughts:
all_thoughts.append(resdata)
// 開始接上最後一小節
pcs = multi_training.loadPieces("./music/過門資料夾")
xIpt, xOpt = map(lambda x: numpy.array(x, dtype='int8'), multi_training.getPieceSegment(pcs))
all_outputs = [xOpt[0]]
if keep_thoughts:
all_thoughts = []
m.start_slow_walk(xIpt[0])
cons = 1
// 利用model產生過門
resdata = m.slow_walk_fun( cons )
nnotes = numpy.sum(resdata[-1][:,0])
if nnotes < 2:
if cons > 1:
cons = 1
cons -= 0.02
else:
cons += (1 - cons)*0.3
