时间序列数据建模流程范例
国内的新冠肺炎疫情从发现至今已经持续3个多月了，这场起源于吃野味的灾难给大家的生活造成了诸多方面的影响。
有的同学是收入上的，有的同学是感情上的，有的同学是心理上的，还有的同学是体重上的。
那么国内的新冠肺炎疫情何时结束呢？什么时候我们才可以重获自由呢？
本篇文章将利用TensorFlow2.0建立时间序列RNN模型，对国内的新冠肺炎疫情结束时间进行预测。
一，准备数据
本文的数据集取自tushare，获取该数据集的方法参考了以下文章。
《https://zhuanlan.zhihu.com/p/109556102》
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.keras import models,layers,losses,metrics,callbacks
%matplotlib inline
%config InlineBackend.figure_format = 'svg'
df = pd.read_csv("./data/covid-19.csv",sep = "\t")
df.plot(x = "date",y = ["confirmed_num","cured_num","dead_num"],figsize=(10,6))
plt.xticks(rotation=60)
(array([-10., 0., 10., 20., 30., 40., 50.]),
<a list of 7 Text xticklabel objects>)
dfdata = df.set_index("date")
dfdiff = dfdata.diff(periods=1).dropna()
dfdiff = dfdiff.reset_index("date")
dfdiff.plot(x = "date",y = ["confirmed_num","cured_num","dead_num"],figsize=(10,6))
plt.xticks(rotation=60)
dfdiff = dfdiff.drop("date",axis = 1).astype("float32")
#用某日前8天窗口数据作为输入预测该日数据
WINDOW_SIZE = 8
def batch_dataset(dataset):
dataset_batched = dataset.batch(WINDOW_SIZE,drop_remainder=True)
return dataset_batched
ds_data = tf.data.Dataset.from_tensor_slices(tf.constant(dfdiff.values,dtype = tf.float32)) \
.window(WINDOW_SIZE,shift=1).flat_map(batch_dataset)
ds_label = tf.data.Dataset.from_tensor_slices(
tf.constant(dfdiff.values[WINDOW_SIZE:],dtype = tf.float32))
#数据较小，可以将全部训练数据放入到一个batch中，提升性能
ds_train = tf.data.Dataset.zip((ds_data,ds_label)).batch(38).cache()
二，定义模型
使用Keras接口有以下3种方式构建模型：使用Sequential按层顺序构建模型，使用函数式API构建任意结构模型，继承Model基类构建自定义模型。
此处选择使用函数式API构建任意结构模型。
#考虑到新增确诊，新增治愈，新增死亡人数数据不可能小于0，设计如下结构
class Block(layers.Layer):
def __init__(self, **kwargs):
super(Block, self).__init__(**kwargs)
def call(self, x_input,x):
x_out = tf.maximum((1+x)*x_input[:,-1,:],0.0)
return x_out
def get_config(self):
config = super(Block, self).get_config()
return config
tf.keras.backend.clear_session()
x_input = layers.Input(shape = (None,3),dtype = tf.float32)
x = layers.LSTM(3,return_sequences = True,input_shape=(None,3))(x_input)
x = layers.LSTM(3,return_sequences = True,input_shape=(None,3))(x)
x = layers.LSTM(3,return_sequences = True,input_shape=(None,3))(x)
x = layers.LSTM(3,input_shape=(None,3))(x)
x = layers.Dense(3)(x)
#考虑到新增确诊，新增治愈，新增死亡人数数据不可能小于0，设计如下结构
#x = tf.maximum((1+x)*x_input[:,-1,:],0.0)
x = Block()(x_input,x)
model = models.Model(inputs = [x_input],outputs = [x])
model.summary()
Model: "model"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, None, 3)] 0
_________________________________________________________________
lstm (LSTM) (None, None, 3) 84
_________________________________________________________________
lstm_1 (LSTM) (None, None, 3) 84
_________________________________________________________________
lstm_2 (LSTM) (None, None, 3) 84
_________________________________________________________________
lstm_3 (LSTM) (None, 3) 84
_________________________________________________________________
dense (Dense) (None, 3) 12
_________________________________________________________________
block (Block) (None, 3) 0
=================================================================
Total params: 348
Trainable params: 348
Non-trainable params: 0
_________________________________________________________________
三，训练模型
训练模型通常有3种方法，内置fit方法，内置train_on_batch方法，以及自定义训练循环。此处我们选择最常用也最简单的内置fit方法。
注：循环神经网络调试较为困难，需要设置多个不同的学习率多次尝试，以取得较好的效果。
#自定义损失函数，考虑平方差和预测目标的比值
class MSPE(losses.Loss):
def call(self,y_true,y_pred):
err_percent = (y_true - y_pred)**2/(tf.maximum(y_true**2,1e-7))
mean_err_percent = tf.reduce_mean(err_percent)
return mean_err_percent
def get_config(self):
config = super(MSPE, self).get_config()
return config
import datetime
import os
optimizer = tf.keras.optimizers.Adam(learning_rate=0.01)
model.compile(optimizer=optimizer,loss=MSPE(name = "MSPE"))
current_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
logdir = os.path.join(".\\data\\keras_model",current_time)
tb_callback = tf.keras.callbacks.TensorBoard(logdir, histogram_freq=1)
#如果loss在100个epoch后没有提升，学习率减半。
lr_callback = tf.keras.callbacks.ReduceLROnPlateau(monitor="loss",factor = 0.5, patience = 100)
#当loss在200个epoch后没有提升，则提前终止训练。
stop_callback = tf.keras.callbacks.EarlyStopping(monitor = "loss", patience= 200)
callbacks_list = [tb_callback,lr_callback,stop_callback]
history = model.fit(ds_train,epochs=500,callbacks = callbacks_list)
Epoch 1/500
1/1 [==============================] - 8s 8s/step - loss: 3.3592
Epoch 2/500
WARNING:tensorflow:Method (on_train_batch_end) is slow compared to the batch update (0.149920). Check your callbacks.
1/1 [==============================] - 0s 245ms/step - loss: 3.2057
Epoch 3/500
1/1 [==============================] - 0s 103ms/step - loss: 3.0463
Epoch 4/500
1/1 [==============================] - 0s 100ms/step - loss: 2.8786
Epoch 5/500
1/1 [==============================] - 0s 95ms/step - loss: 2.7016
Epoch 6/500
1/1 [==============================] - 0s 100ms/step - loss: 2.5145
Epoch 7/500
1/1 [==============================] - 0s 113ms/step - loss: 2.3169
Epoch 8/500
1/1 [==============================] - 0s 98ms/step - loss: 2.1090
Epoch 9/500
1/1 [==============================] - 0s 99ms/step - loss: 1.8919
Epoch 10/500
1/1 [==============================] - 0s 194ms/step - loss: 1.6681
Epoch 11/500
1/1 [==============================] - 0s 94ms/step - loss: 1.4420
Epoch 12/500
1/1 [==============================] - 0s 99ms/step - loss: 1.2206
Epoch 13/500
1/1 [==============================] - 0s 114ms/step - loss: 1.0131
Epoch 14/500
1/1 [==============================] - 0s 185ms/step - loss: 0.8308
Epoch 15/500
1/1 [==============================] - 0s 100ms/step - loss: 0.6851
Epoch 16/500
1/1 [==============================] - 0s 113ms/step - loss: 0.5832
Epoch 17/500
1/1 [==============================] - 0s 123ms/step - loss: 0.5255
Epoch 18/500
1/1 [==============================] - 0s 121ms/step - loss: 0.5047
Epoch 19/500
1/1 [==============================] - 0s 194ms/step - loss: 0.5093
Epoch 20/500
1/1 [==============================] - 0s 98ms/step - loss: 0.5254
Epoch 21/500
1/1 [==============================] - 0s 98ms/step - loss: 0.5403
Epoch 22/500
1/1 [==============================] - 0s 131ms/step - loss: 0.5457
Epoch 23/500
1/1 [==============================] - 0s 115ms/step - loss: 0.5394
Epoch 24/500
1/1 [==============================] - 0s 106ms/step - loss: 0.5227
Epoch 25/500
1/1 [==============================] - 0s 191ms/step - loss: 0.4988
Epoch 26/500
1/1 [==============================] - 0s 122ms/step - loss: 0.4715
Epoch 27/500
1/1 [==============================] - 0s 105ms/step - loss: 0.4446
Epoch 28/500
1/1 [==============================] - 0s 214ms/step - loss: 0.4209
Epoch 29/500
1/1 [==============================] - 0s 123ms/step - loss: 0.4024
Epoch 30/500
1/1 [==============================] - 0s 103ms/step - loss: 0.3897
Epoch 31/500
1/1 [==============================] - 0s 128ms/step - loss: 0.3830
Epoch 32/500
1/1 [==============================] - 0s 135ms/step - loss: 0.3814
Epoch 33/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3835
Epoch 34/500
1/1 [==============================] - 0s 120ms/step - loss: 0.3881
Epoch 35/500
1/1 [==============================] - 0s 107ms/step - loss: 0.3937
Epoch 36/500
1/1 [==============================] - 0s 86ms/step - loss: 0.3993
Epoch 37/500
1/1 [==============================] - 0s 100ms/step - loss: 0.4040
Epoch 38/500
1/1 [==============================] - 0s 247ms/step - loss: 0.4073
Epoch 39/500
1/1 [==============================] - 0s 147ms/step - loss: 0.4089
Epoch 40/500
1/1 [==============================] - 0s 138ms/step - loss: 0.4089
Epoch 41/500
1/1 [==============================] - 0s 120ms/step - loss: 0.4075
Epoch 42/500
1/1 [==============================] - 0s 116ms/step - loss: 0.4048
Epoch 43/500
1/1 [==============================] - 0s 246ms/step - loss: 0.4014
Epoch 44/500
1/1 [==============================] - 0s 112ms/step - loss: 0.3975
Epoch 45/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3936
Epoch 46/500
1/1 [==============================] - 0s 168ms/step - loss: 0.3899
Epoch 47/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3867
Epoch 48/500
1/1 [==============================] - 0s 124ms/step - loss: 0.3842
Epoch 49/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3824
Epoch 50/500
1/1 [==============================] - 0s 103ms/step - loss: 0.3812
Epoch 51/500
1/1 [==============================] - 0s 140ms/step - loss: 0.3807
Epoch 52/500
1/1 [==============================] - 0s 142ms/step - loss: 0.3806
Epoch 53/500
1/1 [==============================] - 0s 135ms/step - loss: 0.3808
Epoch 54/500
1/1 [==============================] - 0s 102ms/step - loss: 0.3812
Epoch 55/500
1/1 [==============================] - 0s 116ms/step - loss: 0.3815
Epoch 56/500
1/1 [==============================] - 0s 114ms/step - loss: 0.3818
Epoch 57/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3818
Epoch 58/500
1/1 [==============================] - 0s 102ms/step - loss: 0.3817
Epoch 59/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3815
Epoch 60/500
1/1 [==============================] - 0s 92ms/step - loss: 0.3811
Epoch 61/500
1/1 [==============================] - 0s 116ms/step - loss: 0.3807
Epoch 62/500
1/1 [==============================] - 0s 116ms/step - loss: 0.3803
Epoch 63/500
1/1 [==============================] - 0s 96ms/step - loss: 0.3800
Epoch 64/500
1/1 [==============================] - 0s 121ms/step - loss: 0.3797
Epoch 65/500
1/1 [==============================] - 0s 124ms/step - loss: 0.3796
Epoch 66/500
1/1 [==============================] - 0s 86ms/step - loss: 0.3795
Epoch 67/500
1/1 [==============================] - 0s 100ms/step - loss: 0.3795
Epoch 68/500
1/1 [==============================] - 0s 111ms/step - loss: 0.3796
Epoch 69/500
1/1 [==============================] - 0s 108ms/step - loss: 0.3797
Epoch 70/500
1/1 [==============================] - 0s 117ms/step - loss: 0.3798
Epoch 71/500
1/1 [==============================] - 0s 141ms/step - loss: 0.3798
Epoch 72/500
1/1 [==============================] - 0s 108ms/step - loss: 0.3798
Epoch 73/500
1/1 [==============================] - 0s 130ms/step - loss: 0.3798
Epoch 74/500
1/1 [==============================] - 0s 117ms/step - loss: 0.3797
Epoch 75/500
1/1 [==============================] - 0s 112ms/step - loss: 0.3795
Epoch 76/500
1/1 [==============================] - 0s 119ms/step - loss: 0.3794
Epoch 77/500
1/1 [==============================] - 0s 105ms/step - loss: 0.3792
Epoch 78/500
1/1 [==============================] - 0s 101ms/step - loss: 0.3791
Epoch 79/500
1/1 [==============================] - 0s 116ms/step - loss: 0.3789
Epoch 80/500
1/1 [==============================] - 0s 104ms/step - loss: 0.3789
Epoch 81/500
1/1 [==============================] - 0s 105ms/step - loss: 0.3788
Epoch 82/500
1/1 [==============================] - 0s 109ms/step - loss: 0.3788
Epoch 83/500
1/1 [==============================] - 0s 113ms/step - loss: 0.3788
Epoch 84/500
1/1 [==============================] - 0s 99ms/step - loss: 0.3788
Epoch 85/500
1/1 [==============================] - 0s 101ms/step - loss: 0.3789
Epoch 86/500
1/1 [==============================] - 0s 106ms/step - loss: 0.3789
Epoch 87/500
1/1 [==============================] - 0s 104ms/step - loss: 0.3789
Epoch 88/500
1/1 [==============================] - 0s 100ms/step - loss: 0.3790
Epoch 89/500
1/1 [==============================] - 0s 94ms/step - loss: 0.3790
Epoch 90/500
1/1 [==============================] - 0s 116ms/step - loss: 0.3790
Epoch 91/500
1/1 [==============================] - 0s 95ms/step - loss: 0.3789
Epoch 92/500
1/1 [==============================] - 0s 114ms/step - loss: 0.3789
Epoch 93/500
1/1 [==============================] - 0s 116ms/step - loss: 0.3789
Epoch 94/500
1/1 [==============================] - 0s 101ms/step - loss: 0.3789
Epoch 95/500
1/1 [==============================] - 0s 125ms/step - loss: 0.3788
Epoch 96/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 97/500
1/1 [==============================] - 0s 217ms/step - loss: 0.3788
Epoch 98/500
1/1 [==============================] - 0s 108ms/step - loss: 0.3788
Epoch 99/500
1/1 [==============================] - 0s 119ms/step - loss: 0.3788
Epoch 100/500
1/1 [==============================] - 0s 122ms/step - loss: 0.3788
Epoch 101/500
1/1 [==============================] - 0s 123ms/step - loss: 0.3788
Epoch 102/500
1/1 [==============================] - 0s 128ms/step - loss: 0.3788
Epoch 103/500
1/1 [==============================] - 0s 107ms/step - loss: 0.3788
Epoch 104/500
1/1 [==============================] - 0s 143ms/step - loss: 0.3788
Epoch 105/500
1/1 [==============================] - 0s 123ms/step - loss: 0.3788
Epoch 106/500
1/1 [==============================] - 0s 111ms/step - loss: 0.3788
Epoch 107/500
1/1 [==============================] - 0s 104ms/step - loss: 0.3788
Epoch 108/500
1/1 [==============================] - 0s 138ms/step - loss: 0.3788
Epoch 109/500
1/1 [==============================] - 0s 254ms/step - loss: 0.3788
Epoch 110/500
1/1 [==============================] - 0s 134ms/step - loss: 0.3788
Epoch 111/500
1/1 [==============================] - 0s 119ms/step - loss: 0.3788
Epoch 112/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 113/500
1/1 [==============================] - 0s 114ms/step - loss: 0.3788
Epoch 114/500
1/1 [==============================] - 0s 87ms/step - loss: 0.3788
Epoch 115/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3788
Epoch 116/500
1/1 [==============================] - 0s 83ms/step - loss: 0.3788
Epoch 117/500
1/1 [==============================] - 0s 112ms/step - loss: 0.3788
Epoch 118/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 119/500
1/1 [==============================] - 0s 114ms/step - loss: 0.3788
Epoch 120/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3788
Epoch 121/500
1/1 [==============================] - 0s 119ms/step - loss: 0.3788
Epoch 122/500
1/1 [==============================] - 0s 223ms/step - loss: 0.3788
Epoch 123/500
1/1 [==============================] - 0s 99ms/step - loss: 0.3788
Epoch 124/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 125/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3788
Epoch 126/500
1/1 [==============================] - 0s 114ms/step - loss: 0.3788
Epoch 127/500
1/1 [==============================] - 0s 129ms/step - loss: 0.3788
Epoch 128/500
1/1 [==============================] - 0s 124ms/step - loss: 0.3788
Epoch 129/500
1/1 [==============================] - 0s 147ms/step - loss: 0.3788
Epoch 130/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 131/500
1/1 [==============================] - 0s 131ms/step - loss: 0.3788
Epoch 132/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 133/500
1/1 [==============================] - 0s 165ms/step - loss: 0.3788
Epoch 134/500
1/1 [==============================] - 0s 147ms/step - loss: 0.3788
Epoch 135/500
1/1 [==============================] - 0s 139ms/step - loss: 0.3788
Epoch 136/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 137/500
1/1 [==============================] - 0s 96ms/step - loss: 0.3788
Epoch 138/500
1/1 [==============================] - 0s 117ms/step - loss: 0.3788
Epoch 139/500
1/1 [==============================] - 0s 110ms/step - loss: 0.3788
Epoch 140/500
1/1 [==============================] - 0s 123ms/step - loss: 0.3788
Epoch 141/500
1/1 [==============================] - 0s 129ms/step - loss: 0.3788
Epoch 142/500
1/1 [==============================] - 0s 116ms/step - loss: 0.3788
Epoch 143/500
1/1 [==============================] - 0s 147ms/step - loss: 0.3788
Epoch 144/500
1/1 [==============================] - 0s 110ms/step - loss: 0.3788
Epoch 145/500
1/1 [==============================] - 0s 86ms/step - loss: 0.3788
Epoch 146/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 147/500
1/1 [==============================] - 0s 143ms/step - loss: 0.3788
Epoch 148/500
1/1 [==============================] - 0s 121ms/step - loss: 0.3788
Epoch 149/500
1/1 [==============================] - 0s 90ms/step - loss: 0.3788
Epoch 150/500
1/1 [==============================] - 0s 94ms/step - loss: 0.3788
Epoch 151/500
1/1 [==============================] - 0s 126ms/step - loss: 0.3788
Epoch 152/500
1/1 [==============================] - 0s 110ms/step - loss: 0.3788
Epoch 153/500
1/1 [==============================] - 0s 191ms/step - loss: 0.3788
Epoch 154/500
1/1 [==============================] - 0s 103ms/step - loss: 0.3788
Epoch 155/500
1/1 [==============================] - 0s 108ms/step - loss: 0.3788
Epoch 156/500
1/1 [==============================] - 0s 101ms/step - loss: 0.3788
Epoch 157/500
1/1 [==============================] - 0s 112ms/step - loss: 0.3788
Epoch 158/500
1/1 [==============================] - 0s 119ms/step - loss: 0.3788
Epoch 159/500
1/1 [==============================] - 0s 108ms/step - loss: 0.3788
Epoch 160/500
1/1 [==============================] - 0s 123ms/step - loss: 0.3788
Epoch 161/500
1/1 [==============================] - 0s 107ms/step - loss: 0.3788
Epoch 162/500
1/1 [==============================] - 0s 126ms/step - loss: 0.3788
Epoch 163/500
1/1 [==============================] - 0s 94ms/step - loss: 0.3788
Epoch 164/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 165/500
1/1 [==============================] - 0s 194ms/step - loss: 0.3788
Epoch 166/500
1/1 [==============================] - 0s 140ms/step - loss: 0.3788
Epoch 167/500
1/1 [==============================] - 0s 189ms/step - loss: 0.3788
Epoch 168/500
1/1 [==============================] - 0s 109ms/step - loss: 0.3788
Epoch 169/500
1/1 [==============================] - 0s 87ms/step - loss: 0.3788
Epoch 170/500
1/1 [==============================] - 0s 101ms/step - loss: 0.3788
Epoch 171/500
1/1 [==============================] - 0s 162ms/step - loss: 0.3788
Epoch 172/500
1/1 [==============================] - 0s 129ms/step - loss: 0.3788
Epoch 173/500
1/1 [==============================] - 0s 116ms/step - loss: 0.3788
Epoch 174/500
1/1 [==============================] - 0s 111ms/step - loss: 0.3788
Epoch 175/500
1/1 [==============================] - 0s 117ms/step - loss: 0.3788
Epoch 176/500
1/1 [==============================] - 0s 111ms/step - loss: 0.3788
Epoch 177/500
1/1 [==============================] - 0s 100ms/step - loss: 0.3788
Epoch 178/500
1/1 [==============================] - 0s 95ms/step - loss: 0.3788
Epoch 179/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 180/500
1/1 [==============================] - 0s 112ms/step - loss: 0.3788
Epoch 181/500
1/1 [==============================] - 0s 101ms/step - loss: 0.3788
Epoch 182/500
1/1 [==============================] - 0s 96ms/step - loss: 0.3788
Epoch 183/500
1/1 [==============================] - 0s 129ms/step - loss: 0.3788
Epoch 184/500
1/1 [==============================] - 0s 153ms/step - loss: 0.3788
Epoch 185/500
1/1 [==============================] - 0s 132ms/step - loss: 0.3788
Epoch 186/500
1/1 [==============================] - 0s 114ms/step - loss: 0.3788
Epoch 187/500
1/1 [==============================] - 0s 112ms/step - loss: 0.3788
Epoch 188/500
1/1 [==============================] - 0s 117ms/step - loss: 0.3788
Epoch 189/500
1/1 [==============================] - 0s 146ms/step - loss: 0.3788
Epoch 190/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 191/500
1/1 [==============================] - 0s 119ms/step - loss: 0.3788
Epoch 192/500
1/1 [==============================] - 0s 248ms/step - loss: 0.3788
Epoch 193/500
1/1 [==============================] - 0s 134ms/step - loss: 0.3788
Epoch 194/500
1/1 [==============================] - 0s 156ms/step - loss: 0.3788
Epoch 195/500
1/1 [==============================] - 0s 108ms/step - loss: 0.3788
Epoch 196/500
1/1 [==============================] - 0s 93ms/step - loss: 0.3788
Epoch 197/500
1/1 [==============================] - 0s 95ms/step - loss: 0.3788
Epoch 198/500
1/1 [==============================] - 0s 100ms/step - loss: 0.3788
Epoch 199/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3788
Epoch 200/500
1/1 [==============================] - 0s 150ms/step - loss: 0.3788
Epoch 201/500
1/1 [==============================] - 0s 144ms/step - loss: 0.3788
Epoch 202/500
1/1 [==============================] - 0s 125ms/step - loss: 0.3788
Epoch 203/500
1/1 [==============================] - 0s 104ms/step - loss: 0.3788
Epoch 204/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 205/500
1/1 [==============================] - 0s 113ms/step - loss: 0.3788
Epoch 206/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 207/500
1/1 [==============================] - 0s 114ms/step - loss: 0.3788
Epoch 208/500
1/1 [==============================] - 0s 82ms/step - loss: 0.3788
Epoch 209/500
1/1 [==============================] - 0s 131ms/step - loss: 0.3788
Epoch 210/500
1/1 [==============================] - 0s 247ms/step - loss: 0.3788
Epoch 211/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3788
Epoch 212/500
1/1 [==============================] - 0s 107ms/step - loss: 0.3788
Epoch 213/500
1/1 [==============================] - 0s 105ms/step - loss: 0.3788
Epoch 214/500
1/1 [==============================] - 0s 97ms/step - loss: 0.3788
Epoch 215/500
1/1 [==============================] - 0s 97ms/step - loss: 0.3788
Epoch 216/500
1/1 [==============================] - 0s 105ms/step - loss: 0.3788
Epoch 217/500
1/1 [==============================] - 0s 99ms/step - loss: 0.3788
Epoch 218/500
1/1 [==============================] - 0s 87ms/step - loss: 0.3788
Epoch 219/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3788
Epoch 220/500
1/1 [==============================] - 0s 99ms/step - loss: 0.3788
Epoch 221/500
1/1 [==============================] - 0s 97ms/step - loss: 0.3788
Epoch 222/500
1/1 [==============================] - 0s 120ms/step - loss: 0.3788
Epoch 223/500
1/1 [==============================] - 0s 106ms/step - loss: 0.3788
Epoch 224/500
1/1 [==============================] - 0s 88ms/step - loss: 0.3788
Epoch 225/500
1/1 [==============================] - 0s 99ms/step - loss: 0.3788
Epoch 226/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3788
Epoch 227/500
1/1 [==============================] - 0s 100ms/step - loss: 0.3788
Epoch 228/500
1/1 [==============================] - 0s 95ms/step - loss: 0.3788
Epoch 229/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 230/500
1/1 [==============================] - 0s 200ms/step - loss: 0.3788
Epoch 231/500
1/1 [==============================] - 0s 178ms/step - loss: 0.3788
Epoch 232/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3788
Epoch 233/500
1/1 [==============================] - 0s 163ms/step - loss: 0.3788
Epoch 234/500
1/1 [==============================] - 0s 222ms/step - loss: 0.3788
Epoch 235/500
1/1 [==============================] - 0s 122ms/step - loss: 0.3788
Epoch 236/500
1/1 [==============================] - 0s 96ms/step - loss: 0.3788
Epoch 237/500
1/1 [==============================] - 0s 88ms/step - loss: 0.3788
Epoch 238/500
1/1 [==============================] - 0s 97ms/step - loss: 0.3788
Epoch 239/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3788
Epoch 240/500
1/1 [==============================] - 0s 113ms/step - loss: 0.3788
Epoch 241/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3788
Epoch 242/500
1/1 [==============================] - 0s 109ms/step - loss: 0.3788
Epoch 243/500
1/1 [==============================] - 0s 111ms/step - loss: 0.3788
Epoch 244/500
1/1 [==============================] - 0s 117ms/step - loss: 0.3788
Epoch 245/500
1/1 [==============================] - 0s 111ms/step - loss: 0.3788
Epoch 246/500
1/1 [==============================] - 0s 116ms/step - loss: 0.3788
Epoch 247/500
1/1 [==============================] - 0s 132ms/step - loss: 0.3788
Epoch 248/500
1/1 [==============================] - 0s 126ms/step - loss: 0.3788
Epoch 249/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 250/500
1/1 [==============================] - 0s 103ms/step - loss: 0.3788
Epoch 251/500
1/1 [==============================] - 0s 107ms/step - loss: 0.3788
Epoch 252/500
1/1 [==============================] - 0s 112ms/step - loss: 0.3788
Epoch 253/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 254/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 255/500
1/1 [==============================] - 0s 91ms/step - loss: 0.3788
Epoch 256/500
1/1 [==============================] - 0s 103ms/step - loss: 0.3788
Epoch 257/500
1/1 [==============================] - 0s 91ms/step - loss: 0.3788
Epoch 258/500
1/1 [==============================] - 0s 94ms/step - loss: 0.3788
Epoch 259/500
1/1 [==============================] - 0s 100ms/step - loss: 0.3788
Epoch 260/500
1/1 [==============================] - 0s 110ms/step - loss: 0.3788
Epoch 261/500
1/1 [==============================] - 0s 101ms/step - loss: 0.3788
Epoch 262/500
1/1 [==============================] - 0s 127ms/step - loss: 0.3788
Epoch 263/500
1/1 [==============================] - 0s 109ms/step - loss: 0.3788
Epoch 264/500
1/1 [==============================] - 0s 102ms/step - loss: 0.3788
Epoch 265/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 266/500
1/1 [==============================] - 0s 111ms/step - loss: 0.3788
Epoch 267/500
1/1 [==============================] - 0s 100ms/step - loss: 0.3788
Epoch 268/500
1/1 [==============================] - 0s 102ms/step - loss: 0.3788
Epoch 269/500
1/1 [==============================] - 0s 217ms/step - loss: 0.3788
Epoch 270/500
1/1 [==============================] - 0s 112ms/step - loss: 0.3788
Epoch 271/500
1/1 [==============================] - 0s 116ms/step - loss: 0.3788
Epoch 272/500
1/1 [==============================] - 0s 108ms/step - loss: 0.3788
Epoch 273/500
1/1 [==============================] - 0s 88ms/step - loss: 0.3788
Epoch 274/500
1/1 [==============================] - 0s 133ms/step - loss: 0.3788
Epoch 275/500
1/1 [==============================] - 0s 164ms/step - loss: 0.3788
Epoch 276/500
1/1 [==============================] - 0s 259ms/step - loss: 0.3788
Epoch 277/500
1/1 [==============================] - 0s 181ms/step - loss: 0.3788
Epoch 278/500
1/1 [==============================] - 0s 103ms/step - loss: 0.3788
Epoch 279/500
1/1 [==============================] - 0s 122ms/step - loss: 0.3788
Epoch 280/500
1/1 [==============================] - 0s 90ms/step - loss: 0.3788
Epoch 281/500
1/1 [==============================] - 0s 103ms/step - loss: 0.3788
Epoch 282/500
1/1 [==============================] - 0s 106ms/step - loss: 0.3788
Epoch 283/500
1/1 [==============================] - 0s 104ms/step - loss: 0.3788
Epoch 284/500
1/1 [==============================] - 0s 99ms/step - loss: 0.3788
Epoch 285/500
1/1 [==============================] - 0s 109ms/step - loss: 0.3788
Epoch 286/500
1/1 [==============================] - 0s 99ms/step - loss: 0.3788
Epoch 287/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 288/500
1/1 [==============================] - 0s 102ms/step - loss: 0.3788
Epoch 289/500
1/1 [==============================] - 0s 93ms/step - loss: 0.3788
Epoch 290/500
1/1 [==============================] - 0s 94ms/step - loss: 0.3788
Epoch 291/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 292/500
1/1 [==============================] - 0s 85ms/step - loss: 0.3788
Epoch 293/500
1/1 [==============================] - 0s 109ms/step - loss: 0.3788
Epoch 294/500
1/1 [==============================] - 0s 95ms/step - loss: 0.3788
Epoch 295/500
1/1 [==============================] - 0s 116ms/step - loss: 0.3788
Epoch 296/500
1/1 [==============================] - 0s 91ms/step - loss: 0.3788
Epoch 297/500
1/1 [==============================] - 0s 108ms/step - loss: 0.3788
Epoch 298/500
1/1 [==============================] - 0s 103ms/step - loss: 0.3788
Epoch 299/500
1/1 [==============================] - 0s 99ms/step - loss: 0.3788
Epoch 300/500
1/1 [==============================] - 0s 124ms/step - loss: 0.3788
Epoch 301/500
1/1 [==============================] - 0s 110ms/step - loss: 0.3788
Epoch 302/500
1/1 [==============================] - 0s 248ms/step - loss: 0.3788
Epoch 303/500
1/1 [==============================] - 0s 108ms/step - loss: 0.3788
Epoch 304/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 305/500
1/1 [==============================] - 0s 110ms/step - loss: 0.3788
Epoch 306/500
1/1 [==============================] - 0s 121ms/step - loss: 0.3788
Epoch 307/500
1/1 [==============================] - 0s 103ms/step - loss: 0.3788
Epoch 308/500
1/1 [==============================] - 0s 109ms/step - loss: 0.3788
Epoch 309/500
1/1 [==============================] - 0s 110ms/step - loss: 0.3788
Epoch 310/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 311/500
1/1 [==============================] - 0s 206ms/step - loss: 0.3788
Epoch 312/500
1/1 [==============================] - 0s 113ms/step - loss: 0.3788
Epoch 313/500
1/1 [==============================] - 0s 130ms/step - loss: 0.3788
Epoch 314/500
1/1 [==============================] - 0s 125ms/step - loss: 0.3788
Epoch 315/500
1/1 [==============================] - 0s 180ms/step - loss: 0.3788
Epoch 316/500
1/1 [==============================] - 0s 128ms/step - loss: 0.3788
Epoch 317/500
1/1 [==============================] - 0s 124ms/step - loss: 0.3788
Epoch 318/500
1/1 [==============================] - 0s 144ms/step - loss: 0.3788
Epoch 319/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 320/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 321/500
1/1 [==============================] - 0s 124ms/step - loss: 0.3788
Epoch 322/500
1/1 [==============================] - 0s 102ms/step - loss: 0.3788
Epoch 323/500
1/1 [==============================] - 0s 108ms/step - loss: 0.3788
Epoch 324/500
1/1 [==============================] - 0s 119ms/step - loss: 0.3788
Epoch 325/500
1/1 [==============================] - 0s 119ms/step - loss: 0.3788
Epoch 326/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 327/500
1/1 [==============================] - 0s 124ms/step - loss: 0.3788
Epoch 328/500
1/1 [==============================] - 0s 124ms/step - loss: 0.3788
Epoch 329/500
1/1 [==============================] - 0s 117ms/step - loss: 0.3788
Epoch 330/500
1/1 [==============================] - 0s 178ms/step - loss: 0.3788
Epoch 331/500
1/1 [==============================] - 0s 118ms/step - loss: 0.3788
Epoch 332/500
1/1 [==============================] - 0s 114ms/step - loss: 0.3788
Epoch 333/500
1/1 [==============================] - 0s 135ms/step - loss: 0.3788
Epoch 334/500
1/1 [==============================] - 0s 121ms/step - loss: 0.3788
Epoch 335/500
1/1 [==============================] - 0s 121ms/step - loss: 0.3788
Epoch 336/500
1/1 [==============================] - 0s 95ms/step - loss: 0.3788
Epoch 337/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 338/500
1/1 [==============================] - 0s 95ms/step - loss: 0.3788
Epoch 339/500
1/1 [==============================] - 0s 99ms/step - loss: 0.3788
Epoch 340/500
1/1 [==============================] - 0s 94ms/step - loss: 0.3788
Epoch 341/500
1/1 [==============================] - 0s 101ms/step - loss: 0.3788
Epoch 342/500
1/1 [==============================] - 0s 96ms/step - loss: 0.3788
Epoch 343/500
1/1 [==============================] - 0s 99ms/step - loss: 0.3788
Epoch 344/500
1/1 [==============================] - 0s 122ms/step - loss: 0.3788
Epoch 345/500
1/1 [==============================] - 0s 107ms/step - loss: 0.3788
Epoch 346/500
1/1 [==============================] - 0s 88ms/step - loss: 0.3788
Epoch 347/500
1/1 [==============================] - 0s 91ms/step - loss: 0.3788
Epoch 348/500
1/1 [==============================] - 0s 95ms/step - loss: 0.3788
Epoch 349/500
1/1 [==============================] - 0s 101ms/step - loss: 0.3788
Epoch 350/500
1/1 [==============================] - 0s 86ms/step - loss: 0.3788
Epoch 351/500
1/1 [==============================] - 0s 92ms/step - loss: 0.3788
Epoch 352/500
1/1 [==============================] - 0s 88ms/step - loss: 0.3788
Epoch 353/500
1/1 [==============================] - 0s 101ms/step - loss: 0.3788
Epoch 354/500
1/1 [==============================] - 0s 97ms/step - loss: 0.3788
Epoch 355/500
1/1 [==============================] - 0s 147ms/step - loss: 0.3788
Epoch 356/500
1/1 [==============================] - 0s 212ms/step - loss: 0.3788
Epoch 357/500
1/1 [==============================] - 0s 100ms/step - loss: 0.3788
Epoch 358/500
1/1 [==============================] - 0s 112ms/step - loss: 0.3788
Epoch 359/500
1/1 [==============================] - 0s 117ms/step - loss: 0.3788
Epoch 360/500
1/1 [==============================] - 0s 163ms/step - loss: 0.3788
Epoch 361/500
1/1 [==============================] - 0s 128ms/step - loss: 0.3788
Epoch 362/500
1/1 [==============================] - 0s 307ms/step - loss: 0.3788
Epoch 363/500
1/1 [==============================] - 0s 107ms/step - loss: 0.3788
Epoch 364/500
1/1 [==============================] - 0s 114ms/step - loss: 0.3788
Epoch 365/500
1/1 [==============================] - 0s 106ms/step - loss: 0.3788
Epoch 366/500
1/1 [==============================] - 0s 97ms/step - loss: 0.3788
Epoch 367/500
1/1 [==============================] - 0s 123ms/step - loss: 0.3788
Epoch 368/500
1/1 [==============================] - 0s 106ms/step - loss: 0.3788
Epoch 369/500
1/1 [==============================] - 0s 97ms/step - loss: 0.3788
Epoch 370/500
1/1 [==============================] - 0s 96ms/step - loss: 0.3788
Epoch 371/500
1/1 [==============================] - 0s 106ms/step - loss: 0.3788
Epoch 372/500
1/1 [==============================] - 0s 115ms/step - loss: 0.3788
Epoch 373/500
1/1 [==============================] - 0s 104ms/step - loss: 0.3788
Epoch 374/500
1/1 [==============================] - 0s 100ms/step - loss: 0.3788
Epoch 375/500
1/1 [==============================] - 0s 107ms/step - loss: 0.3788
Epoch 376/500
1/1 [==============================] - 0s 95ms/step - loss: 0.3788
Epoch 377/500
1/1 [==============================] - 0s 111ms/step - loss: 0.3788
Epoch 378/500
1/1 [==============================] - 0s 113ms/step - loss: 0.3788
Epoch 379/500
1/1 [==============================] - 0s 93ms/step - loss: 0.3788
Epoch 380/500
1/1 [==============================] - 0s 97ms/step - loss: 0.3788
Epoch 381/500
1/1 [==============================] - 0s 84ms/step - loss: 0.3788
Epoch 382/500
1/1 [==============================] - 0s 100ms/step - loss: 0.3788
Epoch 383/500
1/1 [==============================] - 0s 80ms/step - loss: 0.3788
Epoch 384/500
1/1 [==============================] - 0s 99ms/step - loss: 0.3788
Epoch 385/500
1/1 [==============================] - 0s 97ms/step - loss: 0.3788
Epoch 386/500
1/1 [==============================] - 0s 105ms/step - loss: 0.3788
Epoch 387/500
1/1 [==============================] - 0s 98ms/step - loss: 0.3788
Epoch 388/500
1/1 [==============================] - 0s 92ms/step - loss: 0.3788
Epoch 389/500
1/1 [==============================] - 0s 97ms/step - loss: 0.3788
Epoch 390/500
1/1 [==============================] - 0s 107ms/step - loss: 0.3788
四，评估模型
评估模型一般要设置验证集或者测试集，由于此例数据较少，我们仅仅可视化损失函数在训练集上的迭代情况。
%matplotlib inline
%config InlineBackend.figure_format = 'svg'
import matplotlib.pyplot as plt
def plot_metric(history, metric):
train_metrics = history.history[metric]
epochs = range(1, len(train_metrics) + 1)
plt.plot(epochs, train_metrics, 'bo--')
plt.title('Training '+ metric)
plt.xlabel("Epochs")
plt.ylabel(metric)
plt.legend(["train_"+metric])
plt.show()
plot_metric(history,"loss")
五，使用模型
此处我们使用模型预测疫情结束时间，即 新增确诊病例为0 的时间。
#使用dfresult记录现有数据以及此后预测的疫情数据
dfresult = dfdiff[["confirmed_num","cured_num","dead_num"]].copy()
dfresult.tail()
| confirmed_num | cured_num | dead_num | |
|---|---|---|---|
| 41 | 143.0 | 1681.0 | 30.0 |
| 42 | 99.0 | 1678.0 | 28.0 |
| 43 | 44.0 | 1661.0 | 27.0 |
| 44 | 40.0 | 1535.0 | 22.0 |
| 45 | 19.0 | 1297.0 | 17.0 |
#预测此后100天的新增走势,将其结果添加到dfresult中
for i in range(100):
arr_predict = model.predict(tf.constant(tf.expand_dims(dfresult.values[-38:,:],axis = 0)))
dfpredict = pd.DataFrame(tf.cast(tf.floor(arr_predict),tf.float32).numpy(),
columns = dfresult.columns)
dfresult = dfresult.append(dfpredict,ignore_index=True)
dfresult.query("confirmed_num==0").head()
# 第55天开始新增确诊降为0，第45天对应3月10日，也就是10天后，即预计3月20日新增确诊降为0
# 注：该预测偏乐观
| confirmed_num | cured_num | dead_num | |
|---|---|---|---|
| 49 | 0.0 | 1199.0 | 0.0 |
| 50 | 0.0 | 1178.0 | 0.0 |
| 51 | 0.0 | 1158.0 | 0.0 |
| 52 | 0.0 | 1139.0 | 0.0 |
| 53 | 0.0 | 1121.0 | 0.0 |
dfresult.query("cured_num==0").head()
# 第164天开始新增治愈降为0，第45天对应3月10日，也就是大概4个月后，即7月10日左右全部治愈。
# 注: 该预测偏悲观，并且存在问题，如果将每天新增治愈人数加起来，将超过累计确诊人数。
dfresult.query("dead_num==0").head()
六，保存模型
推荐使用TensorFlow原生方式保存模型。
model.save('./data/tf_model_savedmodel', save_format="tf")
print('export saved model.')
WARNING:tensorflow:From D:\anaconda3\lib\site-packages\tensorflow_core\python\ops\resource_variable_ops.py:1786: calling BaseResourceVariable.__init__ (from tensorflow.python.ops.resource_variable_ops) with constraint is deprecated and will be removed in a future version.
Instructions for updating:
If using Keras pass *_constraint arguments to layers.
INFO:tensorflow:Assets written to: ./data/tf_model_savedmodel\assets
export saved model.
model_loaded = tf.keras.models.load_model('./data/tf_model_savedmodel',compile=False)
optimizer = tf.keras.optimizers.Adam(learning_rate=0.001)
model_loaded.compile(optimizer=optimizer,loss=MSPE(name = "MSPE"))
model_loaded.predict(ds_train)
array([[1.10418066e+03, 8.77097855e+01, 4.04874420e+00],
[1.36194556e+03, 6.87724457e+01, 8.23244667e+00],
[1.48609351e+03, 1.46515198e+02, 7.69261456e+00],
[1.70072229e+03, 1.56482224e+02, 8.63732147e+00],
[2.04423340e+03, 2.59142548e+02, 8.77227974e+00],
[1.94323169e+03, 2.60139252e+02, 9.85194492e+00],
[1.65337769e+03, 3.85723694e+02, 9.85194492e+00],
[1.78068201e+03, 5.08318054e+02, 1.16064005e+01],
[1.39508679e+03, 5.97024536e+02, 1.20112753e+01],
[1.56394910e+03, 6.29915710e+02, 1.30909405e+01],
[1.29776733e+03, 7.12641968e+02, 1.45754795e+01],
[1.05999243e+03, 7.41546326e+02, 1.30909405e+01],
[7.97019580e+03, 1.16713818e+03, 3.42793694e+01],
[2.12892749e+03, 8.09322083e+02, 1.75445592e+00],
[1.38930017e+03, 1.36847192e+03, 1.92990150e+01],
[1.05631006e+03, 1.31863684e+03, 1.91640568e+01],
[1.07735205e+03, 1.42030042e+03, 1.41706057e+01],
[9.93183899e+02, 1.70236719e+03, 1.32258987e+01],
[9.20062866e+02, 1.81798462e+03, 1.83543091e+01],
[2.05685867e+02, 1.77313293e+03, 1.53852291e+01],
[4.67659180e+02, 2.10204468e+03, 1.59250612e+01],
[4.32939819e+02, 2.38710156e+03, 1.47104378e+01],
[3.40880920e+02, 2.22164893e+03, 1.30909405e+01],
[1.12574875e+02, 1.83991199e+03, 2.02437229e+01],
[2.67233795e+02, 2.58046167e+03, 9.58202839e+00],
[2.13576630e+02, 2.41401245e+03, 7.01782370e+00],
[2.27779999e+02, 2.74093066e+03, 3.91378641e+00],
[1.72018616e+02, 3.61005493e+03, 5.93815851e+00],
[2.24623688e+02, 2.87548560e+03, 6.34303284e+00],
[3.01427124e+02, 2.61434961e+03, 4.72353506e+00],
[1.06262260e+02, 2.82764380e+03, 5.66824245e+00],
[6.57563477e+01, 2.73295703e+03, 4.18370247e+00],
[6.26000443e+01, 2.64325391e+03, 5.12840939e+00],
[7.31210632e+01, 2.18178076e+03, 4.18370247e+00],
[7.52252655e+01, 1.67545618e+03, 4.04874420e+00],
[5.20790291e+01, 1.67246606e+03, 3.77882814e+00],
[2.31462364e+01, 1.65552209e+03, 3.64386988e+00],
[2.10420322e+01, 1.52993774e+03, 2.96907926e+00]], dtype=float32)
原创文章 58
获赞 7
访问量 6212