【发布时间】：2021-01-06 13:21:20
【问题描述】：
所以每当我运行我的 TensorFlow 模型时，误差幅度 (loss / val_loss) 图都非常倒退和第四，我想知道如何才能阻止它/减少它这是一张图片 Graph
这里是代码，如果有人想运行它应该可以正常工作，只要你有点子
import pandas as pd
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
import numpy as np
import matplotlib.pyplot as plt
import datetime
import tensorboard
from keras.models import Sequential
from keras.layers import Dense
train_df = pd.read_csv('https://www.dropbox.com/s/ednsabkdzs8motw/ROK%20INPUT%20DATA%20-%20Sheet1.csv?dl=1')
eval_df = pd.read_csv('https://www.dropbox.com/s/irnqwc1v67wmbfk/ROK%20EVAL%20DATA%20-%20Sheet1.csv?dl=1')
train_df['Troops'] = train_df['Troops'].astype(float)
train_df['Enemy Troops'] = train_df['Enemy Troops'].astype(float)
train_df['Damage'] = train_df['Damage'].astype(float)
eval_df['Troops'] = eval_df['Troops'].astype(float)
eval_df['Enemy Troops'] = eval_df['Enemy Troops'].astype(float)
eval_df['Damage'] = eval_df['Damage'].astype(float)
damage = train_df.pop('Damage')
dataset = tf.data.Dataset.from_tensor_slices((train_df.values, damage.values))
test_labels = eval_df.pop('Damage')
test_features = eval_df.copy()
model = keras.Sequential(
[
tf.keras.layers.InputLayer(input_shape = (8,)),
tf.keras.layers.Dense(8, activation='relu'),
tf.keras.layers.Dense(64, activation='relu'),
tf.keras.layers.Dense(64, activation='relu'),
tf.keras.layers.Dense(1),
]
)
model.compile(optimizer='adam', loss='mean_squared_error')
model.summary()
history = model.fit(train_df, damage, validation_split=0.2, epochs=5000)
def plot_loss(history):
plt.plot(history.history['loss'], label='loss')
plt.plot(history.history['val_loss'], label='val_loss')
plt.ylim([0, 2000])
plt.xlabel('Epoch')
plt.ylabel('Error [MPG]')
plt.legend()
plt.grid(True)
plot_loss(history)
plt.show()
【问题讨论】：
-
单元比输入多的隐藏层的想法/目的是什么？降维通常用于减少输入并学习其表示。
-
我对机器学习很陌生，所以我基于我看到的许多教程中的值，这些教程使用了比输入更大的单位。如果这是一个不好的做法，我绝对可以改变它
标签： python pandas tensorflow keras