Keras，存在准确性和纳米损失问题

Question

以下是编码：

import os
os.environ\['KAGGLE_CONFIG_DIR'\] = '/content'
import pandas as pd
from tensorflow import keras
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler, MinMaxScaler
import numpy as np
\# !kaggle competitions download -c spaceship-titanic
\# !unzip \\\*.zip && rm \*.zip
df = pd.read_csv("train.csv")
df1 = pd.read_csv("test.csv")
df.head()
df\[\['Deck', 'Num', 'Side'\]\] = df\['Cabin'\].str.split("/",expand=True)
df1\[\['Deck', 'Num', 'Side'\]\] = df1\['Cabin'\].str.split("/",expand=True)
df.head()
df = df.drop("Cabin",axis=1)
df1 = df1.drop("Cabin",axis=1)
df.head()
df\["Billed"\] = df\[\['RoomService','FoodCourt','ShoppingMall','Spa','VRDeck'\]\].sum(axis=1)
df = df.drop(\['RoomService','FoodCourt','ShoppingMall','Spa','VRDeck',\],axis=1)
df1\["Billed"\] = df1\[\['RoomService','FoodCourt','ShoppingMall','Spa','VRDeck'\]\].sum(axis=1)
df1 = df1.drop(\['RoomService','FoodCourt','ShoppingMall','Spa','VRDeck',\],axis=1)
df.head()
df = df.drop('Name',axis=1)
df1 = df1.drop('Name',axis=1)
df.head()
df\[\['Pass_group','group_num'\]\] = df\['PassengerId'\].str.split("\_",expand=True)
df = df.drop("PassengerId",axis=1)
df1\[\['Pass_group','group_num'\]\] = df1\['PassengerId'\].str.split("\_",expand=True)
df1 = df1.drop("PassengerId",axis=1)
df.head()
vocab_homePlanet = df\["HomePlanet"\].value_counts()
vocab_homePlanet
df\['HomePlanet'\] = df\['HomePlanet'\].dropna()
df\['HomePlanet'\].value_counts()
df\["Destination"\].value_counts()
df\["Deck"\].value_counts()
df.head()
df = df.astype(str)
df1 = df1.astype(str)
df
vectorize_layer = keras.layers.TextVectorization(max_tokens=50000,output_mode='int',output_sequence_length=4)
vectorize_layer.adapt(df\['HomePlanet'\])
df\['HomePlanet'\] = vectorize_layer(df\['HomePlanet'\])
vectorize_layer = keras.layers.TextVectorization(max_tokens=50000,output_mode='int',output_sequence_length=4)
vectorize_layer.adapt(df1\['HomePlanet'\])
df1\['HomePlanet'\] = vectorize_layer(df1\['HomePlanet'\])
df.head()
def vectorized_layer(data,input,output_length) :
vectorize_layer = keras.layers.TextVectorization(max_tokens=5000,output_mode='int',output_sequence_length=output_length)
vectorize_layer.adapt(data\[input\])
return vectorize_layer(data\[input\])
df\['Destination'\].value_counts()
df\['Destination'\] = vectorized_layer(df,'Destination',4)
df1\['Destination'\] = vectorized_layer(df1,'Destination',4)
df1\['Deck'\] = vectorized_layer(df1,'Deck',9)
df\['Side'\] = vectorized_layer(df,'Side',3)
df1\['Side'\] = vectorized_layer(df1,'Side',3)
df\['CryoSleep'\] = df\['CryoSleep'\].map({'True':1,'False':0})
df\['VIP'\] = df\['VIP'\].map({'True':1,'False':0})
df\['Transported'\] = df\['Transported'\].map({'True':1,'False':0})
df1\['CryoSleep'\] = df1\['CryoSleep'\].map({'True':1,'False':0})
df1\['VIP'\] = df1\['VIP'\].map({'True':1,'False':0})
\# df1\['Transported'\] = df1\['Transported'\].map({'True':1,'False':0})
x = df.drop(\['Transported'\],axis=1)
y = df\['Transported'\]
\# x_normalized = df\[\['CryoSleep','VIP'\]\]
x = x.to_numpy(dtype='float32')
y = y.to_numpy(dtype='float32')
\# x_test = df1.drop(\['CryoSleep','VIP'\],axis=1)
x_test = df1
\# x_test_normalized = df\[\['CryoSleep','VIP'\]\]
x_test = x_test.to_numpy(dtype='float32')
x.dtype, x_test.dtype
minMaxScaller = MinMaxScaler(feature_range=(0,1))
StdScaller = StandardScaler()
x_minMax = minMaxScaller.fit_transform(x)
x_minMax
x_test_minMax = minMaxScaller.fit_transform(x_test)
\# x_Std = StdScaller.fit_transform(x)
\# x_Std\[1\]
x_train,x_val,y_train,y_val = train_test_split(x_minMax,y,test_size=0.2,random_state=42)
x_train.shape, x_val.shape, y_train.shape, y_val.shape
model = keras.Sequential()
input_shape = x.shape\[1\]
input_shape
x_train.view()
\# model.add(keras.layers.Embedding(500,16))
model.add(keras.layers.Dense(512,activation='relu',input_shape=(,input_shape)))
model.add(keras.layers.BatchNormalization())
model.add(keras.layers.Dropout(0.1))
model.add(keras.layers.Dense(256,activation='relu'))
model.add(keras.layers.BatchNormalization())
model.add(keras.layers.Dropout(0.1))
model.add(keras.layers.Dense(128,activation='relu'))
model.add(keras.layers.BatchNormalization())
model.add(keras.layers.Dropout(0.1))
\# model.add(keras.layers.Flatten())
model.add(keras.layers.Dense(1,activation='sigmoid'))
optimizer = keras.optimizers.Adam(learning_rate=0.001)
model.compile(optimizer=optimizer,loss='mse',metrics=\['accuracy'\])
model.fit(x_train,y_train,epochs=20)
model.evaluate(x_test_minMax)

我尝试更改 Adam 学习率的参数，添加更密集的层，增加 Epochs，但是精度没有改变，仍然有 nan 损失。

Answer 1

以下是结果

纪元1/20 218/218 [================================] - 4s 8ms/步 - 损失：nan - 准确度：0.4967 纪元 2/20 218/218 [==============================] - 2s 7ms/步 - 损失：nan - 准确度：0.4967 纪元 3/20 218/218 [==============================] - 1s 6ms/步 - 损失：nan - 准确度：0.4967