tensorflow2.0,基于LSTM的新冠预测

import tensorflow as tf
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from tensorflow import keras
from tensorflow.keras import layers

df = pd.read_csv('./COVID-19_China.csv')

df.head()

	countryEnglishName	confirmedCount	nowconfirmedCount	curedCount	deadCount	deadRate	curedRate
0	China	544	499	28	17	3.125000	5.147059
1	China	639	592	30	17	2.660407	4.694836
2	China	901	839	36	26	2.885683	3.995560
3	China	1377	1297	39	41	2.977487	2.832244
4	China	2076	1971	49	56	2.697495	2.360308

用前三天的新冠数据,预测第四天,构造新的训练数据和标签

value = df['confirmedCount'].values[20:140]
print(len(value))
x = []
y = []
seq = 3
for i in range(len(value)-seq-1):
    x.append(value[i:i+seq])
    y.append(value[i+seq])

118

train_x = (np.array(x[0:90],dtype=np.float32)/100000.).reshape(-1, seq, 1)

train_y = (np.array(y[0:90],dtype=np.float32)/100000.).reshape(-1, 1)
test_x = (np.array(x[90:110],dtype=np.float32)/100000.).reshape(-1, seq, 1)
test_y = (np.array(y[90:110],dtype=np.float32)/100000.).reshape(-1, 1)

print(train_x.shape)

(90, 3, 1)

model = keras.Sequential()
model.add(layers.LSTM(32, input_shape=(train_x.shape[1:]), return_sequences=True))
model.add(layers.LSTM(32, return_sequences=True))
model.add(layers.LSTM(32))
model.add(layers.Dense(1))

model.compile(optimizer=keras.optimizers.Adam(), loss='mae',metrics=['accuracy'])
learning_rate_reduction = keras.callbacks.ReduceLROnPlateau(monitor='val_loss', patience=3, factor=0.5, min_lr=0.00001)

model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
lstm (LSTM)                  (None, 3, 32)             4352      
_________________________________________________________________
lstm_1 (LSTM)                (None, 3, 32)             8320      
_________________________________________________________________
lstm_2 (LSTM)                (None, 32)                8320      
_________________________________________________________________
dense (Dense)                (None, 1)                 33        
=================================================================
Total params: 21,025
Trainable params: 21,025
Non-trainable params: 0
_________________________________________________________________

tf.keras.utils.plot_model(model,show_shapes = True)

history = model.fit(train_x, train_y,
                    batch_size = 128,
                    epochs=1000,
                    validation_data=(test_x, test_y),
                    callbacks=[learning_rate_reduction])

Train on 90 samples, validate on 20 samples
Epoch 1/1000
90/90 [==============================] - 9s 105ms/sample - loss: 0.8074 - accuracy: 0.0000e+00 - val_loss: 0.8302 - val_accuracy: 0.0000e+00


Epoch 1000/1000
90/90 [==============================] - 0s 188us/sample - loss: 0.0089 - accuracy: 0.0000e+00 - val_loss: 0.0085 - val_accuracy: 0.0000e+00

model(train_x).numpy()

array([[0.6455748 ],
       [0.6756083 ],
       [0.725161  ],
       [0.7416195 ],
       [0.7532444 ],
       [0.7631259 ],
       [0.7728118 ],
       [0.7811027 ],
       [0.78719884],
       [0.7901641 ],
       [0.7943473 ],
       [0.7978379 ],
       [0.8003325 ],
       [0.8019932 ],
       [0.8043494 ],
       [0.8063849 ],
       [0.8083892 ],
       [0.81033134],
       [0.8125819 ],
       [0.81462765],
       [0.8154769 ],
       [0.81612027],
       [0.8167974 ],
       [0.8175316 ],
       [0.8181267 ],
       [0.81846243],
       [0.8186874 ],
       [0.81885636],
       [0.81899816],
       [0.81913096],
       [0.8192062 ],
       [0.8193179 ],
       [0.8194875 ],
       [0.8196601 ],
       [0.8198598 ],
       [0.8201023 ],
       [0.8204589 ],
       [0.8208657 ],
       [0.8213866 ],
       [0.8218331 ],
       [0.8224054 ],
       [0.82299423],
       [0.8235421 ],
       [0.82408875],
       [0.8247542 ],
       [0.8253892 ],
       [0.8259906 ],
       [0.8265722 ],
       [0.827014  ],
       [0.8274759 ],
       [0.82790923],
       [0.8282994 ],
       [0.8286484 ],
       [0.82891166],
       [0.82924974],
       [0.82959986],
       [0.82994974],
       [0.8303709 ],
       [0.83072305],
       [0.8310832 ],
       [0.831541  ],
       [0.83207834],
       [0.8324945 ],
       [0.8328698 ],
       [0.8332928 ],
       [0.8340022 ],
       [0.83510035],
       [0.835257  ],
       [0.8354161 ],
       [0.83552617],
       [0.83566976],
       [0.83573765],
       [0.8357913 ],
       [0.83584535],
       [0.83590865],
       [0.8359509 ],
       [0.83600116],
       [0.8360815 ],
       [0.8361014 ],
       [0.8361371 ],
       [0.8361847 ],
       [0.83620495],
       [0.83622694],
       [0.836259  ],
       [0.8362705 ],
       [0.8362912 ],
       [0.83631444],
       [0.8363304 ],
       [0.8363674 ],
       [0.83644485]], dtype=float32)

prediction = list((model(train_x).numpy().reshape(-1))*100000) + list((model(test_x).numpy().reshape(-1))*100000)

plt.plot(value[3:], label='True Value')
plt.plot(prediction[:91], label='LSTM fit')
plt.plot(np.arange(90, 110, 1), prediction[90:], label='LSTM pred')
print(len(value[3:]))
print(len(prediction[90:]))
plt.legend(loc='best')
plt.title('Cumulative infections prediction(China)')
plt.xlabel('Days')
plt.ylabel('Cumulative Cases')
plt.show()

115
20