KERAS-前馈NN-损耗未减少结构数据

如何解决KERAS-前馈NN-损耗未减少结构数据

我正在尝试为（二进制）分类问题创建前馈NN。

我的数据集的头看起来像这样： dataset

我的数据框的形状是（7214，7）。我的目标是二进制，输入变量是：

• 年龄：人数
• juv_fel_count：数字
• juv_misd_count：数字
• juv_other_count：数量
• 先验计数：num
• c_charge_degree：猫

下面以keras为例（https://keras.io/examples/structured_data/structured_data_classification_from_scratch/），我编写了以下代码：

    val_dataframe = dataframe.sample(frac=0.3,random_state=1337)
train_dataframe = dataframe.drop(val_dataframe.index)

print(
    "Using %d samples for training and %d for validation"
    % (len(train_dataframe),len(val_dataframe))
)
import tensorflow as tf
def dataframe_to_dataset(dataframe):
    dataframe = dataframe.copy()
    labels = dataframe.pop("target")
    ds = tf.data.Dataset.from_tensor_slices((dict(dataframe),labels))
    ds = ds.shuffle(buffer_size=len(dataframe))
    return ds


train_ds = dataframe_to_dataset(train_dataframe)
val_ds = dataframe_to_dataset(val_dataframe)

train_ds = train_ds.batch(50)
val_ds = val_ds.batch(50)

from tensorflow.keras.layers.experimental.preprocessing import CategoryEncoding
from tensorflow.keras.layers.experimental.preprocessing import StringLookup
from tensorflow.keras.layers.experimental.preprocessing import normalization

def encode_numerical_feature(feature,name,dataset):
    # Create a normalization layer for our feature
    normalizer = normalization()

    # Prepare a Dataset that only yields our feature
    feature_ds = dataset.map(lambda x,y: x[name])
    feature_ds = feature_ds.map(lambda x: tf.expand_dims(x,-1))

    # Learn the statistics of the data
    normalizer.adapt(feature_ds)

    # normalize the input feature
    encoded_feature = normalizer(feature)
    return encoded_feature


def encode_string_categorical_feature(feature,dataset):
    # Create a StringLookup layer which will turn strings into integer indices
    index = StringLookup()

    # Prepare a Dataset that only yields our feature
    feature_ds = dataset.map(lambda x,-1))

    # Learn the set of possible string values and assign them a fixed integer index
    index.adapt(feature_ds)

    # Turn the string input into integer indices
    encoded_feature = index(feature)

    # Create a CategoryEncoding for our integer indices
    encoder = CategoryEncoding(output_mode="binary")

    # Prepare a dataset of indices
    feature_ds = feature_ds.map(index)

    # Learn the space of possible indices
    encoder.adapt(feature_ds)

    # Apply one-hot encoding to our indices
    encoded_feature = encoder(encoded_feature)
    return encoded_feature

import keras as keras
juv_fel_count = keras.Input(shape=(1,),name="juv_fel_count")
juv_misd_count = keras.Input(shape=(1,name="juv_misd_count")
juv_other_count = keras.Input(shape=(1,name="juv_other_count")
priors_count = keras.Input(shape=(1,name="priors_count")
age = keras.Input(shape=(1,name="age")
c_charge_degree = keras.Input(shape=(1,name="c_charge_degree",dtype="string")

all_inputs = [
    age,juv_fel_count,juv_misd_count,juv_other_count,priors_count,c_charge_degree,]

c_charge_degree_encoded = encode_string_categorical_feature(c_charge_degree,"c_charge_degree",train_ds)
age_encoded = encode_numerical_feature(age,"age",train_ds)
juv_fel_count_encoded = encode_numerical_feature(juv_fel_count,"juv_fel_count",train_ds)
juv_misd_count_encoded = encode_numerical_feature(juv_misd_count,"juv_misd_count",train_ds)
juv_other_count_encoded = encode_numerical_feature(juv_other_count,"juv_other_count",train_ds)
priors_count_encoded = encode_numerical_feature(priors_count,"priors_count",train_ds)

all_features = layers.concatenate(
    [   juv_fel_count_encoded,juv_misd_count_encoded,juv_other_count_encoded,priors_count_encoded,c_charge_degree_encoded,age_encoded
    ])

x = layers.Dense(50,activation="relu")(all_features)
x = layers.Dropout(0.5)(x)
z = layers.Dense(50,activation="relu")(x)
z = layers.Dropout(0.5)(z)
y = layers.Dense(50,activation="relu")(z)
y = layers.Dropout(0.5)(y)
h = layers.Dense(20,activation="relu")(y)
output = layers.Dense(1,activation="sigmoid")(h)
model = keras.Model(all_inputs,output)

model.compile('adam',"binary_crossentropy",metrics=["accuracy"])

history = model.fit(train_ds,epochs=200,validation_data=val_ds)

损失减少到0.5；然后卡在那儿，不再减少。准确度约为0.75。

结果

pd.DataFrame(history.history)

是以下内容：

training

此外，这些是损失和准确性（训练和验证）的图：

Loss

Accuracy

改变学习率或网络浅层节点的数量似乎无法解决问题。

此外，我真的对Keras和机器学习陌生。因此我无法清楚地了解问题出在哪里。