def Train(self, input, target):
X_train, X_test, Y_train, Y_test = train_test_split(input, target, train_size=0.75)
Y_train = np.asarray(Y_train)
Y_test = np.array(Y_test)
X_train = np.reshape(X_train, [-1, X_train[0].shape[0], X_train[0].shape[1]])
X_test = np.reshape(X_test, [-1, X_train[0].shape[0], X_train[0].shape[1]])
model = Sequential()
model.add(Conv1D(16, 3, padding='same', input_shape=input[0].shape))
model.add(LeakyReLU(alpha=0.2))
model.add(BatchNormalization())
model.add(GRU(16, return_sequences=True))
# model.add(Activation("sigmoid"))
# model.add(LSTM(lstm_out))
model.add(Flatten())
model.add(Dense(8, activity_regularizer=l2(0.001)))
# model.add(GRU(lstm_out, return_sequences=True))
# model.add(LSTM(lstm_out))
# model.add(Dense(20, activity_regularizer=l2(0.001)))
model.add(Activation("relu"))
model.add(Dense(2))
model.compile(loss=mean_absolute_error, optimizer='nadam',
metrics=[RootMeanSquaredError(), MAE])
print(model.summary())
batch_size = 12
epochs = 100
reduce_lr_acc = ReduceLROnPlateau(monitor='val_loss', factor=0.9, patience=epochs / 10, verbose=1, min_delta=1e-4, mode='max')
model.fit(X_train, Y_train,
epochs=epochs,
batch_size=batch_size, validation_data=(X_test, Y_test), callbacks=[reduce_lr_acc])
model.save("PositionEstimation.h5", overwrite=True)
# acc = model.evaluate(X_test,
# Y_test,
# batch_size=batch_size,
# verbose=0)
predicted = model.predict(X_test, batch_size=batch_size)
# predicted = out.ravel()
res = pd.DataFrame({"predicted_x": predicted[:, 0],
"predicted_y": predicted[:, 1],
"original_x": Y_test[:, 0],
"original_y": Y_test[:, 1]})
res.to_excel("res.xlsx")
def free_attn_lstm(dataset_object: LSTM_data):
X_train, X_test, Y_train, Y_test = dataset_object.get_memory()
X_train, X_test = X_train[:, :, :-12], X_test[:, :, :-12]
regressor = Sequential()
# Adding the first LSTM layer and some Dropout regularisation
regressor.add(LSTM(units=NEURONS,
return_sequences=True,
activation=ACTIVATION,
recurrent_activation="sigmoid",
input_shape=(X_train.shape[1], X_train.shape[2]),
bias_regularizer=regularizers.l2(BIAIS_REG),
activity_regularizer=regularizers.l2(L2)
))
regressor.add(Dropout(DROPOUT))
regressor.add(LSTM(units=NEURONS,
activation=ACTIVATION,
recurrent_activation="sigmoid",
return_sequences=True,
bias_regularizer=regularizers.l2(BIAIS_REG),
activity_regularizer=regularizers.l2(L2)
))
regressor.add(Dropout(DROPOUT))
# Adding a second LSTM layer and some Dropout regularisation
regressor.add(LSTM(units=NEURONS,
activation=ACTIVATION,
recurrent_activation="sigmoid",
bias_regularizer=regularizers.l2(BIAIS_REG),
activity_regularizer=regularizers.l2(L2)
))
regressor.add(Dropout(DROPOUT))
# Adding the output layer
regressor.add(Dense(units=1,
activation='relu',
bias_regularizer=regularizers.l2(BIAIS_REG),
activity_regularizer=regularizers.l2(L2)
)
)
optim = Adam()
# Compiling the RNN
regressor.compile(optimizer=optim, loss='mean_squared_error')
# Fitting the RNN to the Training set
history= regressor.fit(X_train,
Y_train,
epochs=EPOCHS,
batch_size=BATCH_SIZE,
validation_data=(X_test, Y_test),
callbacks=[REDUCE_LR, EARLY_STOP]
)
regressor.save("data/weights/free_attn_lstm_no_senti")
plot_train_loss(history)
evaluate(regressor,X_test,Y_test, dataset_object,name="free_attn_lstm", senti="no")
def test_seq_to_seq(self):
#print (self.get_random_states())
train_x = []
# Data size: 10 x (image + 2 actions) x board/action size
train_x = np.random.randint(0, 2, size=(10, 3, 9))
# train_x = [
# [
# [0.1, 1.0],
# [0.1, 1.0],
# [0.1, 1.0],
# [0.1, 1.0],
# [0.1, 1.0],
# ]]
# 1 being the batch size
# 10 being the length
#train_x = np.random.randint(low=0, high=2, size=(1, 10, 9))
train_y = [[0.11, 0.11, 0.11]] * 10
#train_y = [ 0.11 ]
train_y = np.array(train_y)
model = Sequential()
#model.add(layers.Flatten(input_shape=(3, 9))),
#model.add(layers.Embedding(input_shape=(10, 9), ))
model.add(
layers.LSTM(units=100, input_shape=(3, 9), return_sequences=True))
model.add(layers.Dropout(rate=0.25))
model.add(layers.Dense(50, activation='relu'))
model.add(layers.Dense(1, activation=None))
model.compile(optimizer='adam', loss=tf.losses.MSE, metrics=['mae'])
print(model.summary())
model.fit(x=train_x, y=train_y, epochs=100, verbose=0)
loss = model.evaluate(train_x, train_y, verbose=2)
self.assertLess(loss[0], 1e-04)
def dense_net(dataset_object:LSTM_data):
X_train, X_test, Y_train, Y_test = dataset_object.get_memory()
print(X_test.shape, X_train.shape)
X_train = X_train.reshape(X_train.shape[0],X_train.shape[2])
X_test=X_test.reshape(X_test.shape[0], X_test.shape[2])
X_train, X_test = X_train[:, :-12], X_test[:, :-12]
print(X_test.shape, X_train.shape)
regressor = Sequential()
regressor.add(Dense(units=EPOCHS,
activation='relu',
bias_regularizer=regularizers.l2(BIAIS_REG),
activity_regularizer=regularizers.l2(L2)
))
regressor.add(Dense(units=EPOCHS,
activation='relu',
bias_regularizer=regularizers.l2(BIAIS_REG),
activity_regularizer=regularizers.l2(L2)
))
regressor.add(Dense(units=EPOCHS,
activation='relu',
bias_regularizer=regularizers.l2(BIAIS_REG),
activity_regularizer=regularizers.l2(L2)
))
regressor.add(Dropout(DROPOUT))
regressor.add(Dense(units=1,
activation='relu',
bias_regularizer=regularizers.l2(BIAIS_REG),
activity_regularizer=regularizers.l2(L2)
))
optim = Adam()
# Compiling the RNN
regressor.compile(optimizer=optim, loss='mean_squared_error')
# Fitting the RNN to the Training set
history= regressor.fit(X_train,
Y_train,
epochs=EPOCHS,
batch_size=BATCH_SIZE,
validation_data=(X_test, Y_test),
callbacks=[EARLY_STOP, REDUCE_LR])
regressor.save("data/weights/dense_no_senti")
plot_train_loss(history)
evaluate(regressor, X_test,Y_test, dataset_object,name="dense", senti="yes")
Lm1 = Sequential()
#q1. try without input?
Lm1.add(Input(shape=(784, )))
Lm1.add(Dense(num_classes, activation='softmax'))
#q2.try SparseCategoricalCrossentropy without one-hot
loss_object = tf.keras.losses.categorical_crossentropy
optimizer = tf.keras.optimizers.SGD(0.01)
#
train_loss = tf.keras.metrics.Mean(name='train_loss')
#try SparseCategoricalAccuracy
train_accuracy = tf.keras.metrics.CategoricalAccuracy(name='train_accuracy')
test_loss = tf.keras.metrics.Mean(name='test_loss')
test_accuracy = tf.keras.metrics.CategoricalAccuracy(name='test_accuracy')
checkpoint_path = "./checkpoints/"
checkpoint_dir = os.path.dirname(checkpoint_path)
cp_callback = tf.keras.callbacks.ModelCheckpoint(checkpoint_path,
verbose=1,
period=1)
#q3 metrics=xxx without []?
Lm1.compile(optimizer=optimizer, loss=loss_object, metrics=[train_accuracy])
#q4 train_ds?
Lm1.fit(train_ds, epochs=3, callbacks=[cp_callback])
loss, acc = Lm1.evaluate(train_ds)
print("saved model, loss: {:5.2f}, acc: {:5.2f}".format(loss, acc))
class MyAutoEncoder(object):
# archType - 1 => 300|256|300 : archType - 2 => 300|128|300
# archType - 3 => 300|64|300 : archType - 4 => 300|32|300
# archType - 5 => 300|16|300 : archType - 6 => 300|128|64|128|300
# archType - 7 => 300|256|128|128|256|300 : archType - 8 => 300|128|64|32|64|128|300
# archType - 9 => 300||256|128|64|128|256|300 : archType - 10 => 300|128|64|32|16|32|64|128|300
# archType - 11 => 300|256|128|64|32|64|128|256|300 : archType - 12 => 300|256|128|64|32|16|32|64|128|256|300
def __init__(self, logFilePath, inputDim=0, archType=0):
self.logFilePath = logFilePath
if archType == 0: return # We are loading a saved model
# Create auto encoder+decoder
self.autoEncoderModel = Sequential()
self.autoEncoderModel.add(
Dense(inputDim, input_shape=(inputDim, ),
activation='relu')) # Input layer
if archType == 1:
self.autoEncoderModel.add(Dense(256, activation='relu'))
elif archType == 2:
self.autoEncoderModel.add(Dense(128, activation='relu'))
elif archType == 3:
self.autoEncoderModel.add(Dense(64, activation='relu'))
elif archType == 4:
self.autoEncoderModel.add(Dense(32, activation='relu'))
elif archType == 5:
self.autoEncoderModel.add(Dense(16, activation='relu'))
elif archType == 6:
self.autoEncoderModel.add(Dense(128, activation='relu'))
self.autoEncoderModel.add(Dense(64, activation='relu'))
self.autoEncoderModel.add(Dense(128, activation='relu'))
elif archType == 7:
self.autoEncoderModel.add(Dense(256, activation='relu'))
self.autoEncoderModel.add(Dense(128, activation='relu'))
self.autoEncoderModel.add(Dense(256, activation='relu'))
elif archType == 8:
self.autoEncoderModel.add(Dense(128, activation='relu'))
self.autoEncoderModel.add(Dense(64, activation='relu'))
self.autoEncoderModel.add(Dense(32, activation='relu'))
self.autoEncoderModel.add(Dense(64, activation='relu'))
self.autoEncoderModel.add(Dense(128, activation='relu'))
elif archType == 9:
self.autoEncoderModel.add(Dense(256, activation='relu'))
self.autoEncoderModel.add(Dense(128, activation='relu'))
self.autoEncoderModel.add(Dense(64, activation='relu'))
self.autoEncoderModel.add(Dense(128, activation='relu'))
self.autoEncoderModel.add(Dense(256, activation='relu'))
elif archType == 10:
self.autoEncoderModel.add(Dense(128, activation='relu'))
self.autoEncoderModel.add(Dense(64, activation='relu'))
self.autoEncoderModel.add(Dense(32, activation='relu'))
self.autoEncoderModel.add(Dense(16, activation='relu'))
self.autoEncoderModel.add(Dense(32, activation='relu'))
self.autoEncoderModel.add(Dense(64, activation='relu'))
self.autoEncoderModel.add(Dense(128, activation='relu'))
elif archType == 11:
self.autoEncoderModel.add(Dense(256, activation='relu'))
self.autoEncoderModel.add(Dense(128, activation='relu'))
self.autoEncoderModel.add(Dense(64, activation='relu'))
self.autoEncoderModel.add(Dense(32, activation='relu'))
self.autoEncoderModel.add(Dense(64, activation='relu'))
self.autoEncoderModel.add(Dense(128, activation='relu'))
self.autoEncoderModel.add(Dense(256, activation='relu'))
elif archType == 12:
self.autoEncoderModel.add(Dense(256, activation='relu'))
self.autoEncoderModel.add(Dense(128, activation='relu'))
self.autoEncoderModel.add(Dense(64, activation='relu'))
self.autoEncoderModel.add(Dense(32, activation='relu'))
self.autoEncoderModel.add(Dense(16, activation='relu'))
self.autoEncoderModel.add(Dense(32, activation='relu'))
self.autoEncoderModel.add(Dense(64, activation='relu'))
self.autoEncoderModel.add(Dense(128, activation='relu'))
self.autoEncoderModel.add(Dense(256, activation='relu'))
else:
raise ValueError("Incorrect architecture type given.")
self.autoEncoderModel.add(Dense(inputDim,
activation='relu')) # Output layer
self.autoEncoderModel.compile(optimizer='adam', loss=losses.MSE)
self.autoEncoderModel.summary()
# Create encoder
inputSample = Input(shape=(inputDim, ))
inputLayer = self.autoEncoderModel.layers[0]
if 0 < archType < 6:
layerTwo = self.autoEncoderModel.layers[1]
self.encoderModel = Model(inputSample,
layerTwo(inputLayer(inputSample)))
elif archType < 8:
layerTwo = self.autoEncoderModel.layers[1]
layerThree = self.autoEncoderModel.layers[2]
self.encoderModel = Model(
inputSample, layerThree(layerTwo(inputLayer(inputSample))))
elif archType < 10:
layerTwo = self.autoEncoderModel.layers[1]
layerThree = self.autoEncoderModel.layers[2]
layerFour = self.autoEncoderModel.layers[3]
self.encoderModel = Model(
inputSample,
layerFour(layerThree(layerTwo(inputLayer(inputSample)))))
elif archType < 12:
layerTwo = self.autoEncoderModel.layers[1]
layerThree = self.autoEncoderModel.layers[2]
layerFour = self.autoEncoderModel.layers[3]
layerFive = self.autoEncoderModel.layers[4]
self.encoderModel = Model(
inputSample,
layerFive(
layerFour(layerThree(layerTwo(inputLayer(inputSample))))))
elif archType == 12:
layerTwo = self.autoEncoderModel.layers[1]
layerThree = self.autoEncoderModel.layers[2]
layerFour = self.autoEncoderModel.layers[3]
layerFive = self.autoEncoderModel.layers[4]
layerSix = self.autoEncoderModel.layers[5]
self.encoderModel = Model(
inputSample,
layerSix(
layerFive(
layerFour(layerThree(layerTwo(
inputLayer(inputSample)))))))
self.encoderModel.summary()
def train(self, trainX, batchSize, epochs, isDenoising=False):
tic = time.perf_counter()
inputLayer = trainX
if isDenoising: # add some noise to the input layer
inputLayer = trainX + np.random.normal(0, 1, trainX.shape) / 2
self.autoEncoderModel.fit(inputLayer,
trainX,
epochs=epochs,
batch_size=batchSize,
shuffle=True,
validation_split=0.2)
toc = time.perf_counter()
with open(self.logFilePath, "a") as resultsWriter:
resultsWriter.write(
f"AutoEncoder training time: {toc - tic:0.4f} seconds \r")
return toc - tic
def encode(self, dataX, isTrainData):
tic = time.perf_counter()
encodedDataX = self.encoderModel.predict(dataX)
toc = time.perf_counter()
if isTrainData:
with open(self.logFilePath, "a") as resultsWriter:
resultsWriter.write(
f"AutoEncoder training encoding time: {toc - tic:0.4f} seconds \r"
)
else:
with open(self.logFilePath, "a") as resultsWriter:
resultsWriter.write(
f"AutoEncoder testing encoding time: {toc - tic:0.4f} seconds \r\r"
)
return encodedDataX, toc - tic