def getTrainAndValidationData(self, validation_split=7):
(X_train, y_train) = readH5FileTrain(self.__h5file__)
#so it fits into memory without paging
reduce_to = int(X_train.shape[0] * NormalizedSpectrograms.__trainSizeReduction)
X_train = X_train[:reduce_to]
y_train = y_train[:reduce_to]
self.__getMeanAndSd(X_train, y_train)
#normalize spectrograms
X_train = np.subtract(X_train, self.__mean)
X_train = np.divide(X_train, self.__sd)
#normalize bpms
print(self.__y_mean,self.__y_sd)
#shuffle everything
learnLib.shuffle_in_unison(X_train, y_train)
split_at = X_train.shape[0] // validation_split
X_validate = X_train[:split_at]
Y_validate = np.array(list(map(self.normalize_bpm, y_train[:split_at])))
Y_train = np.array(list(map(self.normalize_bpm, y_train[split_at:])))
X_train = (X_train[split_at:])
return (X_train, Y_train), (X_validate, Y_validate)
def getTrainData(self, validation_split=7):
(X_train, y_train) = readH5FileTrain(self.__h5file__)
#so it fits into memory without paging
reduce_to = int(X_train.shape[0] * SubjectSplitSpectrograms.__trainSizeReduction)
X_train = X_train[:reduce_to]
y_train = y_train[:reduce_to]
if not self.__subjectIdependant:
(X, y) = readH5FileValidate(self.__h5file__)
y_train = np.concatenate([y_train, y],0)
X_train = np.concatenate([X_train, X],0)
X_train = np.array(X_train)
y_train = np.array(y_train)
print(X_train.shape)
rnd = np.random.rand(y_train.shape[0])
X_train = X_train[rnd > 0.8]
y_train = y_train[rnd > 0.8]
X_train = X_train[y_train < 140]
y_train = y_train[y_train < 140]
print(X_train.shape)
#shuffle everything
learnLib.shuffle_in_unison(X_train, y_train)
if not self.__subjectIdependant:
split_at = X_train.shape[0] // 4
self.X_validate = X_train[:split_at]
self.Y_validate = np.array(y_train[:split_at])
y_train = y_train[split_at:]
X_train = (X_train[split_at:])
Y_train = np.array(y_train)
return X_train, Y_train
def getTrainData(self, validation_split=7):
(X_train, y_train) = readH5FileTrain(self.__h5file__)
#so it fits into memory without paging
reduce_to = int(X_train.shape[0] *
SubjectSplitSpectrograms.__trainSizeReduction)
X_train = X_train[:reduce_to]
y_train = y_train[:reduce_to]
if not self.__subjectIdependant:
(X, y) = readH5FileValidate(self.__h5file__)
y_train = np.concatenate([y_train, y], 0)
X_train = np.concatenate([X_train, X], 0)
X_train = np.array(X_train)
y_train = np.array(y_train)
print(X_train.shape)
rnd = np.random.rand(y_train.shape[0])
X_train = X_train[rnd > 0.8]
y_train = y_train[rnd > 0.8]
X_train = X_train[y_train < 140]
y_train = y_train[y_train < 140]
print(X_train.shape)
#shuffle everything
learnLib.shuffle_in_unison(X_train, y_train)
if not self.__subjectIdependant:
split_at = X_train.shape[0] // 4
self.X_validate = X_train[:split_at]
self.Y_validate = np.array(y_train[:split_at])
y_train = y_train[split_at:]
X_train = (X_train[split_at:])
Y_train = np.array(y_train)
return X_train, Y_train