def evalModelTesting(model, params): roll = dataUtils.createRepresentation(params["testDir"], limitSongs=50, reductionRatio=params["reductionRatio"]) #remove songs that contain notes out of params["notesMap"] #only with Y, we can take the roll and remove the 1st dimension Y = np.zeros((0,len(params["notesMap"]))) count = 0 for song in roll: songNotes = set() for i in range(song.shape[0]): for j in range(song.shape[1]): if song[i][j] == 1 and j not in songNotes: songNotes.add(j) if songNotes.issubset(set(params["notesMap"])): notesDel = set(range(129)).difference(params["notesMap"]) song = np.delete(song, list(notesDel), 1) Y = np.concatenate((Y,song)) count += 1 print "%d songs" % count return evalModel(model, Y)
def evalModelTesting(model, params):
roll = dataUtils.createRepresentation(
params["testDir"],
limitSongs=50,
reductionRatio=params["reductionRatio"])
#remove songs that contain notes out of params["notesMap"]
#only with Y, we can take the roll and remove the 1st dimension
Y = np.zeros((0, len(params["notesMap"])))
count = 0
for song in roll:
songNotes = set()
for i in range(song.shape[0]):
for j in range(song.shape[1]):
if song[i][j] == 1 and j not in songNotes:
songNotes.add(j)
if songNotes.issubset(set(params["notesMap"])):
notesDel = set(range(129)).difference(params["notesMap"])
song = np.delete(song, list(notesDel), 1)
Y = np.concatenate((Y, song))
count += 1
print "%d songs" % count
return evalModel(model, Y)
from keras.callbacks import ModelCheckpoint
#We need to ensure we're using Python 2.7.x for it to work
if sys.version_info[0] is not 2 and sys.version_info[1] is not 7:
print("Please run this program with Python 2.7")
sys.exit(-1)
if len(sys.argv) < 2:
print("Please specify directory for the model")
print("Loading params...")
path = sys.argv[1] if sys.argv[1][-1] == '/' else sys.argv[1] + '/'
params = dataUtils.loadData(path + "params.nn")
print("Loading data...")
roll = dataUtils.createRepresentation("/Users/Bernat/Dropbox/UoE/Dissertation/midiFiles/", limitSongs=params["limitSongs"], reductionRatio=params["reductionRatio"]) #array of "piano roll" like representations
print("Regenerating inputs")
X, Y = dataUtils.createModelInputs(roll, padding=params["padding"], seqLength=params["seqLength"], inc=params["inc"])
X, Y, notesMap = dataUtils.compressInputs(X, Y)
Y = np.delete(Y, list([53,55]), 1)
notesMap.remove(87)
notesMap.remove(89)
input_dim = len(notesMap)
if notesMap != params["notesMap"]:
print("Notes maps don't match!")
print(notesMap)
print(params["notesMap"])
print("Building model...")
#We need to ensure we're using Python 2.7.x for it to work
if sys.version_info[0] is not 2 and sys.version_info[1] is not 7:
print("Please run this program with Python 2.7")
sys.exit(-1)
if len(sys.argv) < 2:
print("Please specify directory for the model")
print("Loading params...")
path = sys.argv[1] if sys.argv[1][-1] == '/' else sys.argv[1] + '/'
params = dataUtils.loadData(path + "params.nn")
print("Loading data...")
roll = dataUtils.createRepresentation(
"/Users/Bernat/Dropbox/UoE/Dissertation/midiFiles/",
limitSongs=params["limitSongs"],
reductionRatio=params["reductionRatio"]
) #array of "piano roll" like representations
print("Regenerating inputs")
X, Y = dataUtils.createModelInputs(roll,
padding=params["padding"],
seqLength=params["seqLength"],
inc=params["inc"])
X, Y, notesMap = dataUtils.compressInputs(X, Y)
Y = np.delete(Y, list([53, 55]), 1)
notesMap.remove(87)
notesMap.remove(89)
input_dim = len(notesMap)
if notesMap != params["notesMap"]:
print(key + ": " + str(value))
while 1:
user_input = raw_input("Continue? [y]/n: ")
if user_input == 'n':
print("Exiting...")
sys.exit(-1)
elif user_input == 'y' or user_input == '':
break
else:
print("Please press ENTER or 'y' if you want to continue, or 'n' if you wish to abort.")
continue
#Load data
print("Loading data...")
roll = dataUtils.createRepresentation(params["dataDir"], limitSongs=params["limitSongs"], reductionRatio=params["reductionRatio"]) #array of "piano roll" like representations
#Transform
print("Creating output sequences...")
X, Y = dataUtils.createModelInputs(roll, padding=params["padding"], seqLength=params["seqLength"], inc=params["inc"])
X, Y, notesMap = dataUtils.compressInputs(X, Y)
input_dim = len(notesMap)
#Training data shape:
# X -> (nb_samples, timesteps, input_dim) => "sequences of tones"
# Y -> (nb_samples, input_dim) => "next tone for every sequence"
#Build model
print("Building model...")
model = Sequential()
model.add(LSTM(input_dim, input_dim*2, return_sequences=True))
