python - Keras 网络没有学习

Question

我正在尝试学习如何使用 Keras 创建神经网络，但网络无法学习。

这是我的测试，它应该只是学习输出输入，但这甚至不起作用。我刚刚将 X 数据复制到 Y 数据中，并将其输入到训练中。

即使我让它运行 1000 epoch 或更长时间，损失和准确性也不会改变。

import os
os.environ["TF_CPP_MIN_LOG_LEVEL"]="3"
import tensorflow as tf
from keras.models import Sequential
from keras.layers import Dense, Dropout
import numpy

gpus = tf.config.experimental.list_physical_devices("GPU")
if gpus:
    try:
        # Currently, memory growth needs to be the same across GPUs
        for gpu in gpus:
            tf.config.experimental.set_memory_growth(gpu, True)
        logical_gpus = tf.config.experimental.list_logical_devices("GPU")
        print(len(gpus), "Physical GPUs,", len(logical_gpus), "Logical GPUs")
    except RuntimeError as e:
        # Memory growth must be set before GPUs have been initialized
        print(e)

# Input
X_daten = [
    [-4],
    [-3],
    [-2],
    [-1],
    [ 0],
    [ 1],
    [ 2],
    [ 3],
    [ 4]
]
Y_daten = X_daten.copy()

test_anzahl = 2

X_train = numpy.array(X_daten[:-test_anzahl])
Y_train = numpy.array(Y_daten[:-test_anzahl])
X_test  = numpy.array(X_daten[-test_anzahl:])
Y_test  = numpy.array(Y_daten[-test_anzahl:])

print("1 X_train ", X_train.shape)
print("1 Y_train ", Y_train.shape)
print("1 X_test  ", X_test.shape)
print("1 Y_test  ", Y_test.shape)
print("-"*20)

X_train = numpy.reshape(X_train, (X_train.shape[0], 1, X_train.shape[1]))
Y_train = numpy.reshape(Y_train, (Y_train.shape[0], 1, Y_train.shape[1]))
X_test  = numpy.reshape(X_test , ( X_test.shape[0], 1,  X_test.shape[1]))
Y_test  = numpy.reshape(Y_test , ( Y_test.shape[0], 1,  Y_test.shape[1]))

print("2 X_train ", X_train.shape)
print("2 Y_train ", Y_train.shape)
print("2 X_test  ", X_test.shape)
print("2 Y_test  ", Y_test.shape)
print("-"*20)

# Neural Netzwerk
neuronen      = 100
layer         = 2
batch_size    = 10
epoch         = 1000
input_anzahl  = 1
output_anzahl = 1
#dropout       = 0.3
activation    = "sigmoid"
optimizer     = "Adam"

model = Sequential()

model.add(Dense(neuronen, input_shape=(None, input_anzahl), activation=activation))

for _ in range(layer):
    model.add(Dense(neuronen, activation=activation))
    #model.add(Dropout(dropout))

model.add(Dense(output_anzahl, activation=activation)) # Output

model.compile(loss="mean_squared_error", optimizer=optimizer, metrics=["accuracy"])

# Training
model.fit(X_train, Y_train, batch_size=batch_size, epochs=epoch, verbose=1, shuffle=True,         validation_data=(X_test, Y_test))

# Predict
print("\n" + "-"*100)
for i in range(len(X_daten)):
    daten = numpy.array([[X_daten[i]]])
    daten = numpy.reshape(daten, (daten.shape[0], 1, daten.shape[1]))

    ergebnis = model.predict(daten)

    print(" Ergebnis ", ergebnis, " \tY_daten[" + str(i) + "] ", Y_daten[i], " \tX_daten[" + str(i)     + "] ", X_daten[i])

任何帮助将不胜感激:)

Answer 1

我建议首先将激活设置为“线性”。我相信 Sigmoid 被限制在 (0,+1) 范围内，这阻止了网络生成接近目标的值。这可能会让您朝着正确的方向开始。我希望这会有所帮助。