22-T67/keras_multi-attention_multi_task.ipynb

In [1]:

import os
os.environ['CUDA_DEVICE_ORDER'] = 'PCB_BUS_ID'
os.environ['CUDA_VISIBLE_DEVICES'] = '0'

In [2]:

import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
#新增加的两行
from pylab import mpl
# 设置显示中文字体
mpl.rcParams["font.sans-serif"] = ["SimHei"]

mpl.rcParams["axes.unicode_minus"] = False

In [3]:

data = pd.read_csv('./new_train_data.csv', index_col='date')
# data.drop(columns=['wd'], inplace=True) # 风向还没想好怎么处理
data.head()

Out[3]:

	24_PM2.5	24_PM10	24_SO2	24_NO2	24_O3	24_CO	23_PM2.5	23_PM10	23_SO2	23_NO2	...	NH3_resdient	NH3_agricultural	VOC_industrial	VOC_transportation	VOC_resdient	VOC_power	PM2.5_industrial	PM2.5_transportation	PM2.5_resdient	PM2.5_power
date
2015-01-03 01:00:00	136.0	214.0	317.0	38.0	8.0	3.71	114.0	176.0	305.0	38.0	...	0.033910	0.359273	1.177423	1.084925	0.937173	0.037724	0.926851	0.077715	0.827110	0.436028
2015-01-03 02:00:00	114.0	176.0	305.0	38.0	8.0	3.55	97.0	154.0	306.0	37.0	...	0.033910	0.359273	1.177423	1.134240	0.937173	0.036215	0.926851	0.081248	0.827110	0.418587
2015-01-03 03:00:00	97.0	154.0	306.0	37.0	7.0	3.51	87.0	141.0	316.0	38.0	...	0.033910	0.327791	1.177423	1.232869	0.937173	0.035712	0.926851	0.088313	0.827110	0.412773
2015-01-03 04:00:00	87.0	141.0	316.0	38.0	7.0	3.55	85.0	139.0	292.0	37.0	...	0.033910	0.350014	1.177423	1.273965	0.937173	0.036718	0.926851	0.091256	0.827110	0.424400
2015-01-03 05:00:00	85.0	139.0	292.0	37.0	7.0	3.62	106.0	167.0	316.0	37.0	...	0.071588	0.388904	1.177423	1.290403	1.978475	0.039736	0.926851	0.092434	1.746121	0.459282

5 rows × 187 columns

In [4]:

import seaborn as sns

In [9]:

out_cols = ['PM2.5', 'PM10', 'SO2', 'NO2', 'O3', 'CO']
feature_cols = [x for x in data.columns if x not in out_cols and x != 'date']
len(feature_cols), len(out_cols)

Out[9]:

(37, 6)

In [6]:

# for col in data.columns:
#     data[col] = data[col].astype(float)
#     if data[col].max() > 1:
#         data[col] = np.log1p(data[col])

In [10]:

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
import tensorflow.keras.backend as K

2023-03-29 17:14:45.550871: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcudart.so.11.0

In [11]:

class TransformerBlock(layers.Layer):
    def __init__(self, embed_dim, num_heads, ff_dim, name, rate=0.1):
        super().__init__()
        self.att = layers.MultiHeadAttention(num_heads=num_heads, key_dim=embed_dim, name=name)
        self.ffn = keras.Sequential(
            [layers.Dense(ff_dim, activation="relu"), layers.Dense(embed_dim),]
        )
        self.layernorm1 = layers.LayerNormalization(epsilon=1e-6)
        self.layernorm2 = layers.LayerNormalization(epsilon=1e-6)
        self.dropout1 = layers.Dropout(rate)
        self.dropout2 = layers.Dropout(rate)

    def call(self, inputs, training):
        attn_output = self.att(inputs, inputs)
        attn_output = self.dropout1(attn_output, training=training)
        out1 = self.layernorm1(inputs + attn_output)
        ffn_output = self.ffn(out1)
        ffn_output = self.dropout2(ffn_output, training=training)
        return self.layernorm2(out1 + ffn_output)

In [12]:

from tensorflow.keras import Model

In [13]:

from tensorflow.keras.initializers import Constant

In [14]:

# def build_multi_task(num_heads, ff_dim):
#     def build_output(lstm_out, out_name):
#         transformer_block = TransformerBlock(128, num_heads, ff_dim, name='first_attn')
#         out = transformer_block(lstm_out)
#         out = layers.GlobalAveragePooling1D()(out)
#         out = layers.Dropout(0.1)(out)
#         out = layers.Dense(32, activation="relu")(out)
        # out = layers.Dense(1, name=out_name, activation="sigmoid")(out)
#         return out
#     inputs = layers.Input(shape=(1,len(feature_cols)), name='input')
#     x = layers.Conv1D(filters=64, kernel_size=1, activation='relu')(inputs)
#     # x = layers.Dropout(rate=0.1)(x)
#     lstm_out = layers.Bidirectional(layers.LSTM(units=64, return_sequences=True))(x)

#     pm25 = build_output(lstm_out, 'pm25')
#     pm10 = build_output(lstm_out, 'pm10')
#     so2 = build_output(lstm_out, 'so2')
#     no2 = build_output(lstm_out, 'no2')
#     o3 = build_output(lstm_out, 'o3')
#     co = build_output(lstm_out, 'co')

#     model = Model(inputs=[inputs], outputs=[pm25, pm10, so2, no2, o3, co])
#     model.summary()
#     return model

In [12]:

# def build_multi_task(num_heads, ff_dim):
#     def build_output(out, out_name):
#         self_block = TransformerBlock(64, num_heads, ff_dim, name='first_attn')
#         out = self_block(out)
#         out = layers.GlobalAveragePooling1D()(out)
#         out = layers.Dropout(0.1)(out)
#         out = layers.Dense(32, activation="relu")(out)
#         out = layers.Dense(1, name=out_name, activation="sigmoid")(out)
#         return out
#     inputs = layers.Input(shape=(1,len(feature_cols)), name='input')
#     x = layers.Conv1D(filters=64, kernel_size=1, activation='relu')(inputs)
#     # x = layers.Dropout(rate=0.1)(x)
#     lstm_out = layers.Bidirectional(layers.LSTM(units=64, return_sequences=True))(x)
#     lstm_out = layers.Dense(128, activation='relu')(lstm_out)
#     transformer_block = TransformerBlock(128, num_heads, ff_dim)
#     out = transformer_block(lstm_out)
#     out = layers.GlobalAveragePooling1D()(out)
#     out = layers.Dropout(0.1)(out)
#     out = layers.Dense(64, activation='relu')(out)
#     out = K.expand_dims(out, axis=1)

#     pm25 = build_output(out, 'pm25')
#     pm10 = build_output(out, 'pm10')
#     so2 = build_output(out, 'so2')
#     no2 = build_output(out, 'no2')
#     o3 = build_output(out, 'o3')
#     co = build_output(out, 'co')

#     model = Model(inputs=[inputs], outputs=[pm25, pm10, so2, no2, o3, co])
#     model.summary()
#     return model

In [16]:

def build_multi_task(num_heads, ff_dim):
    def build_output(out, out_name):
        self_block = TransformerBlock(64, num_heads, ff_dim, name=f'{out_name}_attn')
        out = self_block(out)
        out = layers.GlobalAveragePooling1D()(out)
        out = layers.Dropout(0.1)(out)
        out = layers.Dense(32, activation="relu")(out)
        # out = layers.Dense(1, name=out_name, activation="sigmoid")(out)
        return out
    inputs = layers.Input(shape=(1,len(feature_cols)), name='input')
    x = layers.Conv1D(filters=64, kernel_size=1, activation='relu')(inputs)
    # x = layers.Dropout(rate=0.1)(x)
    lstm_out = layers.Bidirectional(layers.LSTM(units=64, return_sequences=True))(x)
    lstm_out = layers.Dense(128, activation='relu')(lstm_out)
    transformer_block = TransformerBlock(128, num_heads, ff_dim, name='first_attn')
    out = transformer_block(lstm_out)
    out = layers.GlobalAveragePooling1D()(out)
    out = layers.Dropout(0.1)(out)
    out = layers.Dense(64, activation='relu')(out)
    out = K.expand_dims(out, axis=1)

    pm25 = build_output(out, 'pm25')
    pm10 = build_output(out, 'pm10')
    so2 = build_output(out, 'so2')
    no2 = build_output(out, 'no2')
    o3 = build_output(out, 'o3')
    co = build_output(out, 'co')

    merge = layers.Concatenate(axis=1)([pm25, pm10, so2, no2, o3, co])
    merge = K.expand_dims(merge, axis=1)
    merge_attn = TransformerBlock(32*6, 3, 16, name='last_attn')

    out = merge_attn(merge)
    out = layers.GlobalAveragePooling1D()(out)
    out = layers.Dropout(0.1)(out)

    pm25 = layers.Dense(32, activation='relu')(out)
    pm10 = layers.Dense(32, activation='relu')(out)
    so2 = layers.Dense(32, activation='relu')(out)
    no2 = layers.Dense(32, activation='relu')(out)
    o3 = layers.Dense(32, activation='relu')(out)
    co = layers.Dense(32, activation='relu')(out)

    pm25 = layers.Dense(1, activation='sigmoid', name='pm25')(pm25)
    pm10 = layers.Dense(1, activation='sigmoid', name='pm10')(pm10)
    so2 = layers.Dense(1, activation='sigmoid', name='so2')(so2)
    no2 = layers.Dense(1, activation='sigmoid', name='no2')(no2)
    o3 = layers.Dense(1, activation='sigmoid', name='o3')(o3)
    co = layers.Dense(1, activation='sigmoid', name='co')(co)

    model = Model(inputs=[inputs], outputs=[pm25, pm10, so2, no2, o3, co])
    model.summary()
    return model

In [17]:

from tensorflow.python.keras.utils.vis_utils import plot_model
from tensorflow.keras import optimizers

In [19]:

use_cols = feature_cols + out_cols
use_data = data[use_cols].dropna()

In [20]:

for col in use_cols:
    use_data[col] = use_data[col].astype('float32')

In [21]:

maxs = use_data.max()
mins = use_data.min()
use_cols = use_data.columns

In [22]:

for col in use_cols:
    # use_data[col] = use_data[col].apply(lambda x: 0 if x < 0 else x)
    # use_data[col] = np.log1p(use_data[col])
    use_data[col] = (use_data[col] - mins[col]) / (maxs[col] - mins[col])

In [23]:

train_data, valid = train_test_split(use_data[use_cols], test_size=0.1, random_state=42, shuffle=True)
valid_data, test_data = train_test_split(valid, test_size=0.5, random_state=42, shuffle=True)

In [24]:

opt = optimizers.Adam(learning_rate=5e-5)

In [23]:

def compile_multitask_model(model, gamma):
      w = 1 / gamma
      model.compile(optimizer=opt,
                  loss={'pm25': 'mae', 
                        'pm10': 'mae', 
                        'so2': 'mae', 
                        'no2': 'mae', 
                        'o3': 'mae', 
                        'co': 'mae'},
                  metrics=['mse'])
      return model

In [24]:

model = compile_multitask_model(model, gamma=6)

In [25]:

train_cols = [x for x in out_cols if x not in out_cols]

In [26]:

train_data

Out[26]:

	24_PM2.5	24_PM10	24_SO2	24_NO2	24_O3	24_CO	23_PM2.5	23_PM10	23_SO2	23_NO2	...	PM2.5_industrial	PM2.5_transportation	PM2.5_resdient	PM2.5_power	PM2.5	PM10	SO2	NO2	O3	CO
date
2017-04-21 07:00:00	0.130977	0.163872	0.142349	0.331325	0.015873	0.175287	0.151767	0.176067	0.172598	0.355422	...	0.259570	0.821789	0.130550	0.558376	0.051975	0.069360	0.030645	0.295181	0.073016	0.103448
2020-12-18 13:00:00	0.162162	0.104421	0.103203	0.349398	0.073016	0.204023	0.178794	0.114329	0.101423	0.361446	...	0.581076	0.619134	0.280542	0.390375	0.056133	0.038872	0.032258	0.162651	0.155556	0.125000
2015-07-28 07:00:00	0.166320	0.106707	0.039146	0.150602	0.107937	0.181034	0.170478	0.109756	0.039146	0.186747	...	0.338909	0.661377	0.164996	0.722157	0.153846	0.099085	0.037097	0.156627	0.085714	0.145115
2016-09-27 21:00:00	0.141372	0.096799	0.033808	0.451807	0.028571	0.222701	0.143451	0.101372	0.028470	0.457831	...	0.422103	0.447063	0.098317	0.545690	0.041580	0.042683	0.027419	0.216867	0.142857	0.099138
2019-09-08 23:00:00	0.081081	0.088415	0.014235	0.656627	0.088889	0.090517	0.081081	0.096037	0.016014	0.728916	...	0.331071	0.135968	0.005011	0.112853	0.051975	0.051829	0.012903	0.403614	0.219048	0.070402
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
2016-05-05 17:00:00	0.097713	0.087652	0.090747	0.114458	0.453968	0.107759	0.099792	0.090701	0.092527	0.144578	...	0.712507	0.655019	0.146814	0.788907	0.064449	0.086128	0.014516	0.054217	0.266667	0.057471
2020-07-01 10:00:00	0.033264	0.022104	0.019573	0.084337	0.371429	0.077586	0.027027	0.019055	0.014235	0.048193	...	0.554177	0.626956	0.018196	0.388151	0.085239	0.064024	0.024194	0.246988	0.260317	0.142241
2019-09-27 09:00:00	0.064449	0.056402	0.037367	0.337349	0.088889	0.106322	0.060291	0.052591	0.055160	0.283133	...	0.658467	0.681554	0.040906	0.254681	0.064449	0.051067	0.022581	0.253012	0.098413	0.089080
2015-02-08 22:00:00	0.054054	0.035823	0.113879	0.174699	0.098413	0.218391	0.054054	0.034299	0.113879	0.120482	...	0.475407	0.329360	0.219197	0.546827	0.141372	0.079268	0.298387	0.367470	0.028571	0.350575
2016-12-03 03:00:00	0.284823	0.160061	0.343416	0.373494	0.012698	0.372126	0.228690	0.131860	0.316726	0.373494	...	0.025917	0.038973	0.195944	0.309794	0.611227	0.292683	0.566129	0.560241	0.015873	0.636494

44112 rows × 187 columns

In [28]:

model_history = model.fit(
    {'input': np.expand_dims(train_data[feature_cols].values, axis=1)},
    {
        'pm25': train_data['PM2.5'], 
        'pm10':train_data['PM10'],
        'so2': train_data['SO2'], 
        'no2': train_data['NO2'], 
        'o3': train_data['O3'], 
        'co': train_data['CO'],
    },
    validation_data=(np.expand_dims(valid_data[feature_cols].values, axis=1), valid_data[out_cols].values),
    epochs=50, batch_size=64, verbose=1)

Epoch 1/50
690/690 [==============================] - 13s 19ms/step - loss: 0.2332 - pm25_loss: 0.0390 - pm10_loss: 0.0279 - so2_loss: 0.0278 - no2_loss: 0.0548 - o3_loss: 0.0448 - co_loss: 0.0388 - pm25_mse: 0.0031 - pm10_mse: 0.0016 - so2_mse: 0.0023 - no2_mse: 0.0054 - o3_mse: 0.0041 - co_mse: 0.0028 - val_loss: 0.6660 - val_pm25_loss: 0.0833 - val_pm10_loss: 0.0885 - val_so2_loss: 0.1071 - val_no2_loss: 0.1400 - val_o3_loss: 0.1600 - val_co_loss: 0.0871 - val_pm25_mse: 0.0168 - val_pm10_mse: 0.0188 - val_so2_mse: 0.0248 - val_no2_mse: 0.0340 - val_o3_mse: 0.0483 - val_co_mse: 0.0170
Epoch 2/50
690/690 [==============================] - 12s 17ms/step - loss: 0.2003 - pm25_loss: 0.0326 - pm10_loss: 0.0244 - so2_loss: 0.0248 - no2_loss: 0.0470 - o3_loss: 0.0376 - co_loss: 0.0339 - pm25_mse: 0.0022 - pm10_mse: 0.0013 - so2_mse: 0.0018 - no2_mse: 0.0040 - o3_mse: 0.0028 - co_mse: 0.0021 - val_loss: 0.6676 - val_pm25_loss: 0.0836 - val_pm10_loss: 0.0868 - val_so2_loss: 0.1035 - val_no2_loss: 0.1513 - val_o3_loss: 0.1520 - val_co_loss: 0.0905 - val_pm25_mse: 0.0169 - val_pm10_mse: 0.0182 - val_so2_mse: 0.0238 - val_no2_mse: 0.0398 - val_o3_mse: 0.0442 - val_co_mse: 0.0179
Epoch 3/50
690/690 [==============================] - 11s 17ms/step - loss: 0.1805 - pm25_loss: 0.0290 - pm10_loss: 0.0223 - so2_loss: 0.0225 - no2_loss: 0.0425 - o3_loss: 0.0334 - co_loss: 0.0308 - pm25_mse: 0.0018 - pm10_mse: 0.0011 - so2_mse: 0.0015 - no2_mse: 0.0033 - o3_mse: 0.0022 - co_mse: 0.0018 - val_loss: 0.6681 - val_pm25_loss: 0.0839 - val_pm10_loss: 0.0857 - val_so2_loss: 0.1027 - val_no2_loss: 0.1514 - val_o3_loss: 0.1545 - val_co_loss: 0.0899 - val_pm25_mse: 0.0169 - val_pm10_mse: 0.0180 - val_so2_mse: 0.0237 - val_no2_mse: 0.0400 - val_o3_mse: 0.0461 - val_co_mse: 0.0175
Epoch 4/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1671 - pm25_loss: 0.0267 - pm10_loss: 0.0206 - so2_loss: 0.0213 - no2_loss: 0.0398 - o3_loss: 0.0303 - co_loss: 0.0284 - pm25_mse: 0.0015 - pm10_mse: 9.8021e-04 - so2_mse: 0.0013 - no2_mse: 0.0029 - o3_mse: 0.0018 - co_mse: 0.0015 - val_loss: 0.6596 - val_pm25_loss: 0.0856 - val_pm10_loss: 0.0833 - val_so2_loss: 0.1020 - val_no2_loss: 0.1457 - val_o3_loss: 0.1520 - val_co_loss: 0.0911 - val_pm25_mse: 0.0171 - val_pm10_mse: 0.0171 - val_so2_mse: 0.0234 - val_no2_mse: 0.0373 - val_o3_mse: 0.0455 - val_co_mse: 0.0176
Epoch 5/50
690/690 [==============================] - 11s 17ms/step - loss: 0.1573 - pm25_loss: 0.0252 - pm10_loss: 0.0192 - so2_loss: 0.0201 - no2_loss: 0.0378 - o3_loss: 0.0286 - co_loss: 0.0264 - pm25_mse: 0.0014 - pm10_mse: 8.6174e-04 - so2_mse: 0.0012 - no2_mse: 0.0026 - o3_mse: 0.0016 - co_mse: 0.0013 - val_loss: 0.6741 - val_pm25_loss: 0.0834 - val_pm10_loss: 0.0846 - val_so2_loss: 0.1018 - val_no2_loss: 0.1565 - val_o3_loss: 0.1564 - val_co_loss: 0.0914 - val_pm25_mse: 0.0171 - val_pm10_mse: 0.0177 - val_so2_mse: 0.0233 - val_no2_mse: 0.0424 - val_o3_mse: 0.0469 - val_co_mse: 0.0181
Epoch 6/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1500 - pm25_loss: 0.0238 - pm10_loss: 0.0182 - so2_loss: 0.0192 - no2_loss: 0.0363 - o3_loss: 0.0274 - co_loss: 0.0251 - pm25_mse: 0.0012 - pm10_mse: 7.8217e-04 - so2_mse: 0.0011 - no2_mse: 0.0024 - o3_mse: 0.0015 - co_mse: 0.0012 - val_loss: 0.6692 - val_pm25_loss: 0.0851 - val_pm10_loss: 0.0845 - val_so2_loss: 0.1007 - val_no2_loss: 0.1507 - val_o3_loss: 0.1572 - val_co_loss: 0.0910 - val_pm25_mse: 0.0176 - val_pm10_mse: 0.0176 - val_so2_mse: 0.0231 - val_no2_mse: 0.0404 - val_o3_mse: 0.0482 - val_co_mse: 0.0181
Epoch 7/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1438 - pm25_loss: 0.0229 - pm10_loss: 0.0174 - so2_loss: 0.0186 - no2_loss: 0.0350 - o3_loss: 0.0260 - co_loss: 0.0240 - pm25_mse: 0.0011 - pm10_mse: 7.0840e-04 - so2_mse: 0.0010 - no2_mse: 0.0022 - o3_mse: 0.0013 - co_mse: 0.0011 - val_loss: 0.6538 - val_pm25_loss: 0.0840 - val_pm10_loss: 0.0852 - val_so2_loss: 0.1031 - val_no2_loss: 0.1435 - val_o3_loss: 0.1489 - val_co_loss: 0.0892 - val_pm25_mse: 0.0175 - val_pm10_mse: 0.0179 - val_so2_mse: 0.0238 - val_no2_mse: 0.0376 - val_o3_mse: 0.0431 - val_co_mse: 0.0179
Epoch 8/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1385 - pm25_loss: 0.0221 - pm10_loss: 0.0166 - so2_loss: 0.0178 - no2_loss: 0.0337 - o3_loss: 0.0251 - co_loss: 0.0232 - pm25_mse: 0.0011 - pm10_mse: 6.5187e-04 - so2_mse: 9.3330e-04 - no2_mse: 0.0021 - o3_mse: 0.0012 - co_mse: 0.0010 - val_loss: 0.6569 - val_pm25_loss: 0.0827 - val_pm10_loss: 0.0854 - val_so2_loss: 0.1016 - val_no2_loss: 0.1436 - val_o3_loss: 0.1552 - val_co_loss: 0.0884 - val_pm25_mse: 0.0171 - val_pm10_mse: 0.0180 - val_so2_mse: 0.0235 - val_no2_mse: 0.0376 - val_o3_mse: 0.0472 - val_co_mse: 0.0176
Epoch 9/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1349 - pm25_loss: 0.0212 - pm10_loss: 0.0161 - so2_loss: 0.0175 - no2_loss: 0.0330 - o3_loss: 0.0246 - co_loss: 0.0225 - pm25_mse: 9.9345e-04 - pm10_mse: 6.0136e-04 - so2_mse: 8.9766e-04 - no2_mse: 0.0020 - o3_mse: 0.0012 - co_mse: 9.6550e-04 - val_loss: 0.6593 - val_pm25_loss: 0.0825 - val_pm10_loss: 0.0868 - val_so2_loss: 0.1005 - val_no2_loss: 0.1525 - val_o3_loss: 0.1500 - val_co_loss: 0.0870 - val_pm25_mse: 0.0172 - val_pm10_mse: 0.0184 - val_so2_mse: 0.0232 - val_no2_mse: 0.0405 - val_o3_mse: 0.0446 - val_co_mse: 0.0173
Epoch 10/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1309 - pm25_loss: 0.0207 - pm10_loss: 0.0155 - so2_loss: 0.0171 - no2_loss: 0.0321 - o3_loss: 0.0236 - co_loss: 0.0219 - pm25_mse: 9.3529e-04 - pm10_mse: 5.6009e-04 - so2_mse: 8.5888e-04 - no2_mse: 0.0019 - o3_mse: 0.0011 - co_mse: 9.0392e-04 - val_loss: 0.6582 - val_pm25_loss: 0.0830 - val_pm10_loss: 0.0843 - val_so2_loss: 0.1017 - val_no2_loss: 0.1429 - val_o3_loss: 0.1571 - val_co_loss: 0.0892 - val_pm25_mse: 0.0170 - val_pm10_mse: 0.0176 - val_so2_mse: 0.0235 - val_no2_mse: 0.0375 - val_o3_mse: 0.0485 - val_co_mse: 0.0179
Epoch 11/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1283 - pm25_loss: 0.0203 - pm10_loss: 0.0152 - so2_loss: 0.0167 - no2_loss: 0.0314 - o3_loss: 0.0232 - co_loss: 0.0215 - pm25_mse: 9.1349e-04 - pm10_mse: 5.3300e-04 - so2_mse: 8.1375e-04 - no2_mse: 0.0018 - o3_mse: 0.0010 - co_mse: 8.8792e-04 - val_loss: 0.6643 - val_pm25_loss: 0.0828 - val_pm10_loss: 0.0846 - val_so2_loss: 0.1023 - val_no2_loss: 0.1540 - val_o3_loss: 0.1519 - val_co_loss: 0.0886 - val_pm25_mse: 0.0170 - val_pm10_mse: 0.0177 - val_so2_mse: 0.0237 - val_no2_mse: 0.0417 - val_o3_mse: 0.0457 - val_co_mse: 0.0176
Epoch 12/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1259 - pm25_loss: 0.0198 - pm10_loss: 0.0150 - so2_loss: 0.0165 - no2_loss: 0.0309 - o3_loss: 0.0226 - co_loss: 0.0211 - pm25_mse: 8.6479e-04 - pm10_mse: 5.1626e-04 - so2_mse: 8.0258e-04 - no2_mse: 0.0018 - o3_mse: 9.9486e-04 - co_mse: 8.5558e-04 - val_loss: 0.6619 - val_pm25_loss: 0.0837 - val_pm10_loss: 0.0858 - val_so2_loss: 0.1036 - val_no2_loss: 0.1451 - val_o3_loss: 0.1550 - val_co_loss: 0.0887 - val_pm25_mse: 0.0176 - val_pm10_mse: 0.0181 - val_so2_mse: 0.0240 - val_no2_mse: 0.0388 - val_o3_mse: 0.0471 - val_co_mse: 0.0180
Epoch 13/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1235 - pm25_loss: 0.0196 - pm10_loss: 0.0146 - so2_loss: 0.0162 - no2_loss: 0.0303 - o3_loss: 0.0222 - co_loss: 0.0206 - pm25_mse: 8.5601e-04 - pm10_mse: 4.9180e-04 - so2_mse: 7.6929e-04 - no2_mse: 0.0017 - o3_mse: 9.6504e-04 - co_mse: 8.1659e-04 - val_loss: 0.6584 - val_pm25_loss: 0.0828 - val_pm10_loss: 0.0853 - val_so2_loss: 0.1023 - val_no2_loss: 0.1457 - val_o3_loss: 0.1557 - val_co_loss: 0.0867 - val_pm25_mse: 0.0174 - val_pm10_mse: 0.0180 - val_so2_mse: 0.0236 - val_no2_mse: 0.0387 - val_o3_mse: 0.0477 - val_co_mse: 0.0174
Epoch 14/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1210 - pm25_loss: 0.0191 - pm10_loss: 0.0143 - so2_loss: 0.0158 - no2_loss: 0.0299 - o3_loss: 0.0217 - co_loss: 0.0203 - pm25_mse: 8.1625e-04 - pm10_mse: 4.6297e-04 - so2_mse: 7.2483e-04 - no2_mse: 0.0017 - o3_mse: 9.1613e-04 - co_mse: 7.8936e-04 - val_loss: 0.6591 - val_pm25_loss: 0.0826 - val_pm10_loss: 0.0869 - val_so2_loss: 0.1019 - val_no2_loss: 0.1494 - val_o3_loss: 0.1498 - val_co_loss: 0.0885 - val_pm25_mse: 0.0173 - val_pm10_mse: 0.0185 - val_so2_mse: 0.0235 - val_no2_mse: 0.0403 - val_o3_mse: 0.0441 - val_co_mse: 0.0180
Epoch 15/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1196 - pm25_loss: 0.0190 - pm10_loss: 0.0140 - so2_loss: 0.0158 - no2_loss: 0.0294 - o3_loss: 0.0213 - co_loss: 0.0200 - pm25_mse: 8.0990e-04 - pm10_mse: 4.4497e-04 - so2_mse: 7.2605e-04 - no2_mse: 0.0016 - o3_mse: 8.9130e-04 - co_mse: 7.7173e-04 - val_loss: 0.6597 - val_pm25_loss: 0.0829 - val_pm10_loss: 0.0857 - val_so2_loss: 0.1023 - val_no2_loss: 0.1473 - val_o3_loss: 0.1530 - val_co_loss: 0.0884 - val_pm25_mse: 0.0173 - val_pm10_mse: 0.0181 - val_so2_mse: 0.0237 - val_no2_mse: 0.0398 - val_o3_mse: 0.0462 - val_co_mse: 0.0179
Epoch 16/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1177 - pm25_loss: 0.0185 - pm10_loss: 0.0138 - so2_loss: 0.0154 - no2_loss: 0.0291 - o3_loss: 0.0211 - co_loss: 0.0197 - pm25_mse: 7.7055e-04 - pm10_mse: 4.3773e-04 - so2_mse: 6.9458e-04 - no2_mse: 0.0016 - o3_mse: 8.6953e-04 - co_mse: 7.4449e-04 - val_loss: 0.6606 - val_pm25_loss: 0.0818 - val_pm10_loss: 0.0847 - val_so2_loss: 0.1006 - val_no2_loss: 0.1530 - val_o3_loss: 0.1511 - val_co_loss: 0.0894 - val_pm25_mse: 0.0168 - val_pm10_mse: 0.0178 - val_so2_mse: 0.0232 - val_no2_mse: 0.0413 - val_o3_mse: 0.0455 - val_co_mse: 0.0176
Epoch 17/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1163 - pm25_loss: 0.0185 - pm10_loss: 0.0137 - so2_loss: 0.0152 - no2_loss: 0.0288 - o3_loss: 0.0207 - co_loss: 0.0195 - pm25_mse: 7.7383e-04 - pm10_mse: 4.2540e-04 - so2_mse: 6.6704e-04 - no2_mse: 0.0016 - o3_mse: 8.3579e-04 - co_mse: 7.3199e-04 - val_loss: 0.6612 - val_pm25_loss: 0.0831 - val_pm10_loss: 0.0843 - val_so2_loss: 0.1018 - val_no2_loss: 0.1472 - val_o3_loss: 0.1559 - val_co_loss: 0.0889 - val_pm25_mse: 0.0174 - val_pm10_mse: 0.0177 - val_so2_mse: 0.0236 - val_no2_mse: 0.0395 - val_o3_mse: 0.0481 - val_co_mse: 0.0179
Epoch 18/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1147 - pm25_loss: 0.0181 - pm10_loss: 0.0135 - so2_loss: 0.0150 - no2_loss: 0.0284 - o3_loss: 0.0204 - co_loss: 0.0192 - pm25_mse: 7.2693e-04 - pm10_mse: 4.1564e-04 - so2_mse: 6.4517e-04 - no2_mse: 0.0015 - o3_mse: 8.1594e-04 - co_mse: 7.0265e-04 - val_loss: 0.6583 - val_pm25_loss: 0.0828 - val_pm10_loss: 0.0841 - val_so2_loss: 0.1002 - val_no2_loss: 0.1490 - val_o3_loss: 0.1546 - val_co_loss: 0.0875 - val_pm25_mse: 0.0170 - val_pm10_mse: 0.0176 - val_so2_mse: 0.0231 - val_no2_mse: 0.0400 - val_o3_mse: 0.0475 - val_co_mse: 0.0173
Epoch 19/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1139 - pm25_loss: 0.0180 - pm10_loss: 0.0133 - so2_loss: 0.0150 - no2_loss: 0.0284 - o3_loss: 0.0203 - co_loss: 0.0191 - pm25_mse: 7.2973e-04 - pm10_mse: 3.9422e-04 - so2_mse: 6.5502e-04 - no2_mse: 0.0015 - o3_mse: 8.0642e-04 - co_mse: 6.9577e-04 - val_loss: 0.6550 - val_pm25_loss: 0.0823 - val_pm10_loss: 0.0854 - val_so2_loss: 0.1022 - val_no2_loss: 0.1446 - val_o3_loss: 0.1525 - val_co_loss: 0.0880 - val_pm25_mse: 0.0172 - val_pm10_mse: 0.0181 - val_so2_mse: 0.0237 - val_no2_mse: 0.0380 - val_o3_mse: 0.0457 - val_co_mse: 0.0178
Epoch 20/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1125 - pm25_loss: 0.0177 - pm10_loss: 0.0132 - so2_loss: 0.0148 - no2_loss: 0.0280 - o3_loss: 0.0200 - co_loss: 0.0188 - pm25_mse: 7.0545e-04 - pm10_mse: 3.9227e-04 - so2_mse: 6.3487e-04 - no2_mse: 0.0015 - o3_mse: 7.8305e-04 - co_mse: 6.7808e-04 - val_loss: 0.6481 - val_pm25_loss: 0.0813 - val_pm10_loss: 0.0851 - val_so2_loss: 0.1032 - val_no2_loss: 0.1401 - val_o3_loss: 0.1514 - val_co_loss: 0.0870 - val_pm25_mse: 0.0167 - val_pm10_mse: 0.0181 - val_so2_mse: 0.0239 - val_no2_mse: 0.0358 - val_o3_mse: 0.0450 - val_co_mse: 0.0174
Epoch 21/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1119 - pm25_loss: 0.0177 - pm10_loss: 0.0131 - so2_loss: 0.0146 - no2_loss: 0.0279 - o3_loss: 0.0199 - co_loss: 0.0187 - pm25_mse: 7.0314e-04 - pm10_mse: 3.8956e-04 - so2_mse: 6.1240e-04 - no2_mse: 0.0015 - o3_mse: 7.7719e-04 - co_mse: 6.7488e-04 - val_loss: 0.6602 - val_pm25_loss: 0.0825 - val_pm10_loss: 0.0851 - val_so2_loss: 0.1020 - val_no2_loss: 0.1498 - val_o3_loss: 0.1531 - val_co_loss: 0.0876 - val_pm25_mse: 0.0171 - val_pm10_mse: 0.0180 - val_so2_mse: 0.0237 - val_no2_mse: 0.0403 - val_o3_mse: 0.0468 - val_co_mse: 0.0177
Epoch 22/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1100 - pm25_loss: 0.0173 - pm10_loss: 0.0128 - so2_loss: 0.0143 - no2_loss: 0.0275 - o3_loss: 0.0195 - co_loss: 0.0185 - pm25_mse: 6.6950e-04 - pm10_mse: 3.7277e-04 - so2_mse: 5.8999e-04 - no2_mse: 0.0014 - o3_mse: 7.4992e-04 - co_mse: 6.5288e-04 - val_loss: 0.6713 - val_pm25_loss: 0.0840 - val_pm10_loss: 0.0854 - val_so2_loss: 0.1017 - val_no2_loss: 0.1527 - val_o3_loss: 0.1566 - val_co_loss: 0.0910 - val_pm25_mse: 0.0179 - val_pm10_mse: 0.0181 - val_so2_mse: 0.0237 - val_no2_mse: 0.0427 - val_o3_mse: 0.0487 - val_co_mse: 0.0187
Epoch 23/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1097 - pm25_loss: 0.0173 - pm10_loss: 0.0128 - so2_loss: 0.0143 - no2_loss: 0.0275 - o3_loss: 0.0194 - co_loss: 0.0185 - pm25_mse: 6.7541e-04 - pm10_mse: 3.6521e-04 - so2_mse: 5.8931e-04 - no2_mse: 0.0014 - o3_mse: 7.4037e-04 - co_mse: 6.6391e-04 - val_loss: 0.6614 - val_pm25_loss: 0.0829 - val_pm10_loss: 0.0846 - val_so2_loss: 0.1011 - val_no2_loss: 0.1520 - val_o3_loss: 0.1504 - val_co_loss: 0.0903 - val_pm25_mse: 0.0172 - val_pm10_mse: 0.0178 - val_so2_mse: 0.0234 - val_no2_mse: 0.0412 - val_o3_mse: 0.0449 - val_co_mse: 0.0181
Epoch 24/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1084 - pm25_loss: 0.0171 - pm10_loss: 0.0126 - so2_loss: 0.0142 - no2_loss: 0.0272 - o3_loss: 0.0192 - co_loss: 0.0182 - pm25_mse: 6.5768e-04 - pm10_mse: 3.5788e-04 - so2_mse: 5.7106e-04 - no2_mse: 0.0014 - o3_mse: 7.2528e-04 - co_mse: 6.3424e-04 - val_loss: 0.6572 - val_pm25_loss: 0.0836 - val_pm10_loss: 0.0853 - val_so2_loss: 0.1021 - val_no2_loss: 0.1426 - val_o3_loss: 0.1561 - val_co_loss: 0.0874 - val_pm25_mse: 0.0176 - val_pm10_mse: 0.0181 - val_so2_mse: 0.0236 - val_no2_mse: 0.0380 - val_o3_mse: 0.0482 - val_co_mse: 0.0178
Epoch 25/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1078 - pm25_loss: 0.0170 - pm10_loss: 0.0125 - so2_loss: 0.0140 - no2_loss: 0.0272 - o3_loss: 0.0190 - co_loss: 0.0181 - pm25_mse: 6.5638e-04 - pm10_mse: 3.4962e-04 - so2_mse: 5.6850e-04 - no2_mse: 0.0014 - o3_mse: 7.0940e-04 - co_mse: 6.3257e-04 - val_loss: 0.6582 - val_pm25_loss: 0.0826 - val_pm10_loss: 0.0857 - val_so2_loss: 0.1022 - val_no2_loss: 0.1497 - val_o3_loss: 0.1503 - val_co_loss: 0.0876 - val_pm25_mse: 0.0172 - val_pm10_mse: 0.0182 - val_so2_mse: 0.0237 - val_no2_mse: 0.0408 - val_o3_mse: 0.0448 - val_co_mse: 0.0177
Epoch 26/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1067 - pm25_loss: 0.0168 - pm10_loss: 0.0124 - so2_loss: 0.0139 - no2_loss: 0.0269 - o3_loss: 0.0188 - co_loss: 0.0180 - pm25_mse: 6.3574e-04 - pm10_mse: 3.4226e-04 - so2_mse: 5.5284e-04 - no2_mse: 0.0014 - o3_mse: 6.9490e-04 - co_mse: 6.2231e-04 - val_loss: 0.6668 - val_pm25_loss: 0.0838 - val_pm10_loss: 0.0848 - val_so2_loss: 0.1002 - val_no2_loss: 0.1538 - val_o3_loss: 0.1555 - val_co_loss: 0.0887 - val_pm25_mse: 0.0176 - val_pm10_mse: 0.0179 - val_so2_mse: 0.0231 - val_no2_mse: 0.0426 - val_o3_mse: 0.0479 - val_co_mse: 0.0180
Epoch 27/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1060 - pm25_loss: 0.0167 - pm10_loss: 0.0122 - so2_loss: 0.0138 - no2_loss: 0.0268 - o3_loss: 0.0186 - co_loss: 0.0179 - pm25_mse: 6.2988e-04 - pm10_mse: 3.3928e-04 - so2_mse: 5.4696e-04 - no2_mse: 0.0014 - o3_mse: 6.8858e-04 - co_mse: 6.1392e-04 - val_loss: 0.6545 - val_pm25_loss: 0.0829 - val_pm10_loss: 0.0849 - val_so2_loss: 0.1032 - val_no2_loss: 0.1422 - val_o3_loss: 0.1550 - val_co_loss: 0.0863 - val_pm25_mse: 0.0172 - val_pm10_mse: 0.0179 - val_so2_mse: 0.0239 - val_no2_mse: 0.0374 - val_o3_mse: 0.0476 - val_co_mse: 0.0176
Epoch 28/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1055 - pm25_loss: 0.0165 - pm10_loss: 0.0122 - so2_loss: 0.0137 - no2_loss: 0.0267 - o3_loss: 0.0187 - co_loss: 0.0177 - pm25_mse: 6.1787e-04 - pm10_mse: 3.3690e-04 - so2_mse: 5.3732e-04 - no2_mse: 0.0014 - o3_mse: 6.8977e-04 - co_mse: 6.0403e-04 - val_loss: 0.6602 - val_pm25_loss: 0.0831 - val_pm10_loss: 0.0852 - val_so2_loss: 0.1013 - val_no2_loss: 0.1478 - val_o3_loss: 0.1541 - val_co_loss: 0.0887 - val_pm25_mse: 0.0174 - val_pm10_mse: 0.0181 - val_so2_mse: 0.0234 - val_no2_mse: 0.0401 - val_o3_mse: 0.0472 - val_co_mse: 0.0179
Epoch 29/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1051 - pm25_loss: 0.0165 - pm10_loss: 0.0121 - so2_loss: 0.0136 - no2_loss: 0.0266 - o3_loss: 0.0186 - co_loss: 0.0177 - pm25_mse: 6.2427e-04 - pm10_mse: 3.3626e-04 - so2_mse: 5.2777e-04 - no2_mse: 0.0014 - o3_mse: 6.8583e-04 - co_mse: 5.9979e-04 - val_loss: 0.6590 - val_pm25_loss: 0.0832 - val_pm10_loss: 0.0868 - val_so2_loss: 0.1012 - val_no2_loss: 0.1461 - val_o3_loss: 0.1534 - val_co_loss: 0.0882 - val_pm25_mse: 0.0177 - val_pm10_mse: 0.0186 - val_so2_mse: 0.0235 - val_no2_mse: 0.0392 - val_o3_mse: 0.0470 - val_co_mse: 0.0181
Epoch 30/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1042 - pm25_loss: 0.0164 - pm10_loss: 0.0121 - so2_loss: 0.0135 - no2_loss: 0.0264 - o3_loss: 0.0183 - co_loss: 0.0175 - pm25_mse: 6.1534e-04 - pm10_mse: 3.2614e-04 - so2_mse: 5.3001e-04 - no2_mse: 0.0014 - o3_mse: 6.6255e-04 - co_mse: 5.9483e-04 - val_loss: 0.6618 - val_pm25_loss: 0.0831 - val_pm10_loss: 0.0846 - val_so2_loss: 0.1023 - val_no2_loss: 0.1482 - val_o3_loss: 0.1549 - val_co_loss: 0.0886 - val_pm25_mse: 0.0175 - val_pm10_mse: 0.0178 - val_so2_mse: 0.0238 - val_no2_mse: 0.0405 - val_o3_mse: 0.0475 - val_co_mse: 0.0181
Epoch 31/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1035 - pm25_loss: 0.0162 - pm10_loss: 0.0119 - so2_loss: 0.0134 - no2_loss: 0.0264 - o3_loss: 0.0181 - co_loss: 0.0175 - pm25_mse: 5.9792e-04 - pm10_mse: 3.1864e-04 - so2_mse: 5.2179e-04 - no2_mse: 0.0013 - o3_mse: 6.4994e-04 - co_mse: 5.8687e-04 - val_loss: 0.6585 - val_pm25_loss: 0.0826 - val_pm10_loss: 0.0848 - val_so2_loss: 0.0991 - val_no2_loss: 0.1517 - val_o3_loss: 0.1508 - val_co_loss: 0.0894 - val_pm25_mse: 0.0171 - val_pm10_mse: 0.0180 - val_so2_mse: 0.0228 - val_no2_mse: 0.0414 - val_o3_mse: 0.0456 - val_co_mse: 0.0180
Epoch 32/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1028 - pm25_loss: 0.0161 - pm10_loss: 0.0118 - so2_loss: 0.0132 - no2_loss: 0.0263 - o3_loss: 0.0180 - co_loss: 0.0174 - pm25_mse: 5.9557e-04 - pm10_mse: 3.1492e-04 - so2_mse: 4.9868e-04 - no2_mse: 0.0013 - o3_mse: 6.4688e-04 - co_mse: 5.7651e-04 - val_loss: 0.6626 - val_pm25_loss: 0.0828 - val_pm10_loss: 0.0844 - val_so2_loss: 0.1020 - val_no2_loss: 0.1491 - val_o3_loss: 0.1566 - val_co_loss: 0.0877 - val_pm25_mse: 0.0172 - val_pm10_mse: 0.0178 - val_so2_mse: 0.0237 - val_no2_mse: 0.0410 - val_o3_mse: 0.0484 - val_co_mse: 0.0179
Epoch 33/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1023 - pm25_loss: 0.0161 - pm10_loss: 0.0117 - so2_loss: 0.0133 - no2_loss: 0.0261 - o3_loss: 0.0180 - co_loss: 0.0172 - pm25_mse: 5.9043e-04 - pm10_mse: 3.0794e-04 - so2_mse: 5.0748e-04 - no2_mse: 0.0013 - o3_mse: 6.4401e-04 - co_mse: 5.7312e-04 - val_loss: 0.6591 - val_pm25_loss: 0.0831 - val_pm10_loss: 0.0845 - val_so2_loss: 0.1015 - val_no2_loss: 0.1461 - val_o3_loss: 0.1533 - val_co_loss: 0.0906 - val_pm25_mse: 0.0172 - val_pm10_mse: 0.0178 - val_so2_mse: 0.0236 - val_no2_mse: 0.0391 - val_o3_mse: 0.0469 - val_co_mse: 0.0184
Epoch 34/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1019 - pm25_loss: 0.0160 - pm10_loss: 0.0117 - so2_loss: 0.0131 - no2_loss: 0.0261 - o3_loss: 0.0179 - co_loss: 0.0171 - pm25_mse: 5.8674e-04 - pm10_mse: 3.1048e-04 - so2_mse: 4.9479e-04 - no2_mse: 0.0013 - o3_mse: 6.3890e-04 - co_mse: 5.7154e-04 - val_loss: 0.6655 - val_pm25_loss: 0.0831 - val_pm10_loss: 0.0850 - val_so2_loss: 0.1017 - val_no2_loss: 0.1480 - val_o3_loss: 0.1576 - val_co_loss: 0.0901 - val_pm25_mse: 0.0172 - val_pm10_mse: 0.0180 - val_so2_mse: 0.0236 - val_no2_mse: 0.0404 - val_o3_mse: 0.0495 - val_co_mse: 0.0183
Epoch 35/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1011 - pm25_loss: 0.0159 - pm10_loss: 0.0117 - so2_loss: 0.0131 - no2_loss: 0.0258 - o3_loss: 0.0177 - co_loss: 0.0170 - pm25_mse: 5.8263e-04 - pm10_mse: 3.0657e-04 - so2_mse: 4.9633e-04 - no2_mse: 0.0013 - o3_mse: 6.2868e-04 - co_mse: 5.5694e-04 - val_loss: 0.6539 - val_pm25_loss: 0.0825 - val_pm10_loss: 0.0839 - val_so2_loss: 0.1018 - val_no2_loss: 0.1436 - val_o3_loss: 0.1540 - val_co_loss: 0.0880 - val_pm25_mse: 0.0169 - val_pm10_mse: 0.0177 - val_so2_mse: 0.0236 - val_no2_mse: 0.0381 - val_o3_mse: 0.0473 - val_co_mse: 0.0175
Epoch 36/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1007 - pm25_loss: 0.0158 - pm10_loss: 0.0115 - so2_loss: 0.0129 - no2_loss: 0.0258 - o3_loss: 0.0177 - co_loss: 0.0169 - pm25_mse: 5.7141e-04 - pm10_mse: 3.0444e-04 - so2_mse: 4.7820e-04 - no2_mse: 0.0013 - o3_mse: 6.2529e-04 - co_mse: 5.5519e-04 - val_loss: 0.6573 - val_pm25_loss: 0.0825 - val_pm10_loss: 0.0848 - val_so2_loss: 0.1033 - val_no2_loss: 0.1448 - val_o3_loss: 0.1542 - val_co_loss: 0.0876 - val_pm25_mse: 0.0172 - val_pm10_mse: 0.0180 - val_so2_mse: 0.0241 - val_no2_mse: 0.0388 - val_o3_mse: 0.0473 - val_co_mse: 0.0178
Epoch 37/50
690/690 [==============================] - 12s 17ms/step - loss: 0.1006 - pm25_loss: 0.0158 - pm10_loss: 0.0115 - so2_loss: 0.0129 - no2_loss: 0.0258 - o3_loss: 0.0176 - co_loss: 0.0169 - pm25_mse: 5.7186e-04 - pm10_mse: 2.9683e-04 - so2_mse: 4.8355e-04 - no2_mse: 0.0013 - o3_mse: 6.2410e-04 - co_mse: 5.5257e-04 - val_loss: 0.6605 - val_pm25_loss: 0.0830 - val_pm10_loss: 0.0850 - val_so2_loss: 0.1005 - val_no2_loss: 0.1450 - val_o3_loss: 0.1586 - val_co_loss: 0.0884 - val_pm25_mse: 0.0174 - val_pm10_mse: 0.0180 - val_so2_mse: 0.0232 - val_no2_mse: 0.0394 - val_o3_mse: 0.0500 - val_co_mse: 0.0178
Epoch 38/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0998 - pm25_loss: 0.0156 - pm10_loss: 0.0114 - so2_loss: 0.0129 - no2_loss: 0.0256 - o3_loss: 0.0174 - co_loss: 0.0169 - pm25_mse: 5.5708e-04 - pm10_mse: 2.9609e-04 - so2_mse: 4.8605e-04 - no2_mse: 0.0013 - o3_mse: 6.0751e-04 - co_mse: 5.5119e-04 - val_loss: 0.6543 - val_pm25_loss: 0.0828 - val_pm10_loss: 0.0851 - val_so2_loss: 0.1004 - val_no2_loss: 0.1472 - val_o3_loss: 0.1509 - val_co_loss: 0.0879 - val_pm25_mse: 0.0173 - val_pm10_mse: 0.0181 - val_so2_mse: 0.0233 - val_no2_mse: 0.0401 - val_o3_mse: 0.0453 - val_co_mse: 0.0177
Epoch 39/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0996 - pm25_loss: 0.0156 - pm10_loss: 0.0113 - so2_loss: 0.0128 - no2_loss: 0.0257 - o3_loss: 0.0174 - co_loss: 0.0168 - pm25_mse: 5.6576e-04 - pm10_mse: 2.9270e-04 - so2_mse: 4.7426e-04 - no2_mse: 0.0013 - o3_mse: 6.0135e-04 - co_mse: 5.4368e-04 - val_loss: 0.6670 - val_pm25_loss: 0.0826 - val_pm10_loss: 0.0842 - val_so2_loss: 0.1003 - val_no2_loss: 0.1550 - val_o3_loss: 0.1559 - val_co_loss: 0.0891 - val_pm25_mse: 0.0173 - val_pm10_mse: 0.0177 - val_so2_mse: 0.0233 - val_no2_mse: 0.0431 - val_o3_mse: 0.0483 - val_co_mse: 0.0181
Epoch 40/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0989 - pm25_loss: 0.0154 - pm10_loss: 0.0112 - so2_loss: 0.0127 - no2_loss: 0.0255 - o3_loss: 0.0173 - co_loss: 0.0167 - pm25_mse: 5.5163e-04 - pm10_mse: 2.8335e-04 - so2_mse: 4.5740e-04 - no2_mse: 0.0013 - o3_mse: 6.0434e-04 - co_mse: 5.3955e-04 - val_loss: 0.6561 - val_pm25_loss: 0.0830 - val_pm10_loss: 0.0844 - val_so2_loss: 0.1002 - val_no2_loss: 0.1470 - val_o3_loss: 0.1537 - val_co_loss: 0.0879 - val_pm25_mse: 0.0173 - val_pm10_mse: 0.0178 - val_so2_mse: 0.0232 - val_no2_mse: 0.0394 - val_o3_mse: 0.0472 - val_co_mse: 0.0176
Epoch 41/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0984 - pm25_loss: 0.0154 - pm10_loss: 0.0112 - so2_loss: 0.0126 - no2_loss: 0.0254 - o3_loss: 0.0172 - co_loss: 0.0166 - pm25_mse: 5.4599e-04 - pm10_mse: 2.8575e-04 - so2_mse: 4.6223e-04 - no2_mse: 0.0013 - o3_mse: 5.9550e-04 - co_mse: 5.3020e-04 - val_loss: 0.6543 - val_pm25_loss: 0.0826 - val_pm10_loss: 0.0849 - val_so2_loss: 0.1017 - val_no2_loss: 0.1467 - val_o3_loss: 0.1512 - val_co_loss: 0.0872 - val_pm25_mse: 0.0173 - val_pm10_mse: 0.0180 - val_so2_mse: 0.0237 - val_no2_mse: 0.0391 - val_o3_mse: 0.0457 - val_co_mse: 0.0176
Epoch 42/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0984 - pm25_loss: 0.0154 - pm10_loss: 0.0112 - so2_loss: 0.0125 - no2_loss: 0.0254 - o3_loss: 0.0173 - co_loss: 0.0165 - pm25_mse: 5.5078e-04 - pm10_mse: 2.8158e-04 - so2_mse: 4.5397e-04 - no2_mse: 0.0013 - o3_mse: 5.9908e-04 - co_mse: 5.2796e-04 - val_loss: 0.6594 - val_pm25_loss: 0.0825 - val_pm10_loss: 0.0842 - val_so2_loss: 0.1007 - val_no2_loss: 0.1482 - val_o3_loss: 0.1552 - val_co_loss: 0.0885 - val_pm25_mse: 0.0171 - val_pm10_mse: 0.0177 - val_so2_mse: 0.0234 - val_no2_mse: 0.0404 - val_o3_mse: 0.0478 - val_co_mse: 0.0180
Epoch 43/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0978 - pm25_loss: 0.0152 - pm10_loss: 0.0111 - so2_loss: 0.0125 - no2_loss: 0.0254 - o3_loss: 0.0171 - co_loss: 0.0165 - pm25_mse: 5.4125e-04 - pm10_mse: 2.8138e-04 - so2_mse: 4.5162e-04 - no2_mse: 0.0013 - o3_mse: 5.8738e-04 - co_mse: 5.3130e-04 - val_loss: 0.6612 - val_pm25_loss: 0.0834 - val_pm10_loss: 0.0843 - val_so2_loss: 0.1005 - val_no2_loss: 0.1519 - val_o3_loss: 0.1522 - val_co_loss: 0.0889 - val_pm25_mse: 0.0176 - val_pm10_mse: 0.0178 - val_so2_mse: 0.0233 - val_no2_mse: 0.0419 - val_o3_mse: 0.0466 - val_co_mse: 0.0181
Epoch 44/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0975 - pm25_loss: 0.0152 - pm10_loss: 0.0111 - so2_loss: 0.0125 - no2_loss: 0.0253 - o3_loss: 0.0171 - co_loss: 0.0165 - pm25_mse: 5.3677e-04 - pm10_mse: 2.7633e-04 - so2_mse: 4.5067e-04 - no2_mse: 0.0013 - o3_mse: 5.8541e-04 - co_mse: 5.2256e-04 - val_loss: 0.6570 - val_pm25_loss: 0.0839 - val_pm10_loss: 0.0841 - val_so2_loss: 0.1006 - val_no2_loss: 0.1469 - val_o3_loss: 0.1518 - val_co_loss: 0.0899 - val_pm25_mse: 0.0176 - val_pm10_mse: 0.0178 - val_so2_mse: 0.0234 - val_no2_mse: 0.0394 - val_o3_mse: 0.0462 - val_co_mse: 0.0181
Epoch 45/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0973 - pm25_loss: 0.0152 - pm10_loss: 0.0110 - so2_loss: 0.0124 - no2_loss: 0.0252 - o3_loss: 0.0171 - co_loss: 0.0164 - pm25_mse: 5.3606e-04 - pm10_mse: 2.7196e-04 - so2_mse: 4.4353e-04 - no2_mse: 0.0012 - o3_mse: 5.8985e-04 - co_mse: 5.2791e-04 - val_loss: 0.6557 - val_pm25_loss: 0.0816 - val_pm10_loss: 0.0843 - val_so2_loss: 0.0979 - val_no2_loss: 0.1528 - val_o3_loss: 0.1484 - val_co_loss: 0.0907 - val_pm25_mse: 0.0167 - val_pm10_mse: 0.0179 - val_so2_mse: 0.0226 - val_no2_mse: 0.0410 - val_o3_mse: 0.0441 - val_co_mse: 0.0180
Epoch 46/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0973 - pm25_loss: 0.0153 - pm10_loss: 0.0110 - so2_loss: 0.0123 - no2_loss: 0.0252 - o3_loss: 0.0170 - co_loss: 0.0164 - pm25_mse: 5.4535e-04 - pm10_mse: 2.7624e-04 - so2_mse: 4.4083e-04 - no2_mse: 0.0012 - o3_mse: 5.7941e-04 - co_mse: 5.2028e-04 - val_loss: 0.6552 - val_pm25_loss: 0.0823 - val_pm10_loss: 0.0839 - val_so2_loss: 0.0989 - val_no2_loss: 0.1488 - val_o3_loss: 0.1520 - val_co_loss: 0.0893 - val_pm25_mse: 0.0170 - val_pm10_mse: 0.0176 - val_so2_mse: 0.0228 - val_no2_mse: 0.0401 - val_o3_mse: 0.0463 - val_co_mse: 0.0180
Epoch 47/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0962 - pm25_loss: 0.0151 - pm10_loss: 0.0108 - so2_loss: 0.0123 - no2_loss: 0.0249 - o3_loss: 0.0170 - co_loss: 0.0162 - pm25_mse: 5.2758e-04 - pm10_mse: 2.6199e-04 - so2_mse: 4.3608e-04 - no2_mse: 0.0012 - o3_mse: 5.7838e-04 - co_mse: 5.0790e-04 - val_loss: 0.6524 - val_pm25_loss: 0.0818 - val_pm10_loss: 0.0847 - val_so2_loss: 0.1009 - val_no2_loss: 0.1477 - val_o3_loss: 0.1507 - val_co_loss: 0.0866 - val_pm25_mse: 0.0170 - val_pm10_mse: 0.0180 - val_so2_mse: 0.0234 - val_no2_mse: 0.0396 - val_o3_mse: 0.0455 - val_co_mse: 0.0174
Epoch 48/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0965 - pm25_loss: 0.0150 - pm10_loss: 0.0109 - so2_loss: 0.0123 - no2_loss: 0.0252 - o3_loss: 0.0168 - co_loss: 0.0164 - pm25_mse: 5.2843e-04 - pm10_mse: 2.6876e-04 - so2_mse: 4.3716e-04 - no2_mse: 0.0013 - o3_mse: 5.7539e-04 - co_mse: 5.1675e-04 - val_loss: 0.6573 - val_pm25_loss: 0.0820 - val_pm10_loss: 0.0839 - val_so2_loss: 0.1001 - val_no2_loss: 0.1474 - val_o3_loss: 0.1548 - val_co_loss: 0.0891 - val_pm25_mse: 0.0171 - val_pm10_mse: 0.0176 - val_so2_mse: 0.0232 - val_no2_mse: 0.0398 - val_o3_mse: 0.0478 - val_co_mse: 0.0180
Epoch 49/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0961 - pm25_loss: 0.0150 - pm10_loss: 0.0108 - so2_loss: 0.0122 - no2_loss: 0.0251 - o3_loss: 0.0169 - co_loss: 0.0162 - pm25_mse: 5.2412e-04 - pm10_mse: 2.6725e-04 - so2_mse: 4.2533e-04 - no2_mse: 0.0012 - o3_mse: 5.7784e-04 - co_mse: 5.0353e-04 - val_loss: 0.6569 - val_pm25_loss: 0.0825 - val_pm10_loss: 0.0847 - val_so2_loss: 0.0997 - val_no2_loss: 0.1492 - val_o3_loss: 0.1517 - val_co_loss: 0.0892 - val_pm25_mse: 0.0172 - val_pm10_mse: 0.0180 - val_so2_mse: 0.0230 - val_no2_mse: 0.0406 - val_o3_mse: 0.0459 - val_co_mse: 0.0178
Epoch 50/50
690/690 [==============================] - 12s 17ms/step - loss: 0.0956 - pm25_loss: 0.0149 - pm10_loss: 0.0107 - so2_loss: 0.0122 - no2_loss: 0.0250 - o3_loss: 0.0167 - co_loss: 0.0161 - pm25_mse: 5.1678e-04 - pm10_mse: 2.6319e-04 - so2_mse: 4.3059e-04 - no2_mse: 0.0012 - o3_mse: 5.6103e-04 - co_mse: 5.0597e-04 - val_loss: 0.6567 - val_pm25_loss: 0.0831 - val_pm10_loss: 0.0838 - val_so2_loss: 0.1020 - val_no2_loss: 0.1449 - val_o3_loss: 0.1542 - val_co_loss: 0.0887 - val_pm25_mse: 0.0174 - val_pm10_mse: 0.0175 - val_so2_mse: 0.0238 - val_no2_mse: 0.0390 - val_o3_mse: 0.0477 - val_co_mse: 0.0182

In [29]:

rst = model.predict(np.expand_dims(test_data[feature_cols], axis=1))
rst

Out[29]:

[array([[0.02576166],
        [0.17945182],
        [0.06094658],
        ...,
        [0.05389276],
        [0.09377471],
        [0.13254392]], dtype=float32),
 array([[0.03676841],
        [0.12384248],
        [0.04926959],
        ...,
        [0.08627176],
        [0.08261123],
        [0.07822514]], dtype=float32),
 array([[0.00334924],
        [0.03361022],
        [0.00760874],
        ...,
        [0.01287785],
        [0.0179143 ],
        [0.01522398]], dtype=float32),
 array([[0.02466571],
        [0.24751806],
        [0.28095278],
        ...,
        [0.45145005],
        [0.08254912],
        [0.4848558 ]], dtype=float32),
 array([[0.27359694],
        [0.32201412],
        [0.11083704],
        ...,
        [0.02645177],
        [0.5149532 ],
        [0.04898548]], dtype=float32),
 array([[0.0195317 ],
        [0.18470812],
        [0.07515424],
        ...,
        [0.0780178 ],
        [0.06473574],
        [0.12362501]], dtype=float32)]

In [30]:

pred_rst = pd.DataFrame.from_records(np.squeeze(np.asarray(rst), axis=2).T, columns=out_cols)

In [31]:

real_rst = test_data[out_cols].copy()

In [32]:

for col in out_cols:
    pred_rst[col] = pred_rst[col] * (maxs[col] - mins[col]) + mins[col]
    real_rst[col] = real_rst[col] * (maxs[col] - mins[col]) + mins[col]

In [33]:

y_pred_pm25 = pred_rst['PM2.5'].values.reshape(-1,)
y_pred_pm10 = pred_rst['PM10'].values.reshape(-1,)
y_pred_so2 = pred_rst['SO2'].values.reshape(-1,)
y_pred_no2 = pred_rst['NO2'].values.reshape(-1,)
y_pred_o3 = pred_rst['O3'].values.reshape(-1,)
y_pred_co = pred_rst['CO'].values.reshape(-1,)
y_true_pm25 = real_rst['PM2.5'].values.reshape(-1,)
y_true_pm10 = real_rst['PM10'].values.reshape(-1,)
y_true_so2 = real_rst['SO2'].values.reshape(-1,)
y_true_no2 = real_rst['NO2'].values.reshape(-1,)
y_true_o3 = real_rst['O3'].values.reshape(-1,)
y_true_co = real_rst['CO'].values.reshape(-1,)

In [34]:

from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score, mean_absolute_percentage_error

In [59]:

def print_eva(y_true, y_pred, tp):
    MSE = mean_squared_error(y_true, y_pred)
    RMSE = np.sqrt(MSE)
    MAE = mean_absolute_error(y_true, y_pred)
    MAPE = mean_absolute_percentage_error(y_true, y_pred)
    R_2 = r2_score(y_true, y_pred)
    print(f"COL: {tp}, MSE: {format(MSE, '.2E')}", end=',')
    print(f'RMSE: {round(RMSE, 4)}', end=',')
    print(f'MAPE: {round(MAPE, 4) * 100} %', end=',')
    print(f'MAE: {round(MAE, 4)}', end=',')
    print(f'R_2: {round(R_2, 4)}')
    return [MSE, RMSE, MAE, MAPE, R_2]

In [62]:

pm25_eva = print_eva(y_true_pm25, y_pred_pm25, tp='pm25')
pm10_eva = print_eva(y_true_pm10, y_pred_pm10, tp='pm10')
so2_eva = print_eva(y_true_so2, y_pred_so2, tp='so2')
nox_eva = print_eva(y_true_no2, y_pred_no2, tp='no2')
o3_eva = print_eva(y_true_o3, y_pred_o3, tp='o3')
co_eva = print_eva(y_true_co, y_pred_co, tp='co')

COL: pm25, MSE: 1.09E+02,RMSE: 10.422200202941895,MAPE: 12.559999525547028 %,MAE: 6.230599880218506,R_2: 0.9559
COL: pm10, MSE: 3.40E+02,RMSE: 18.428300857543945,MAPE: 11.68999969959259 %,MAE: 12.01930046081543,R_2: 0.9395
COL: so2, MSE: 1.35E+02,RMSE: 11.639599800109863,MAPE: 20.329999923706055 %,MAE: 6.62060022354126,R_2: 0.9567
COL: no2, MSE: 2.97E+01,RMSE: 5.448800086975098,MAPE: 10.360000282526016 %,MAE: 3.8643999099731445,R_2: 0.9526
COL: o3, MSE: 4.93E+01,RMSE: 7.019100189208984,MAPE: 15.330000221729279 %,MAE: 4.6255998611450195,R_2: 0.9802
COL: co, MSE: 1.92E-02,RMSE: 0.1386999934911728,MAPE: 8.969999849796295 %,MAE: 0.0982000008225441,R_2: 0.9616

In [64]:

pd.DataFrame.from_records([pm25_eva, pm10_eva, so2_eva, nox_eva, o3_eva,co_eva], columns=['MSE', 'RMSE', 'MAE', 'MAPE', 'R_2'], index=['PM25', 'PM10', 'SO2', 'NO2', 'O3', 'CO'])

Out[64]:

	MSE	RMSE	MAE	MAPE	R_2
PM25	108.621880	10.422182	6.230620	0.125599	0.955881
PM10	339.600830	18.428261	12.019309	0.116882	0.939480
SO2	135.480988	11.639630	6.620558	0.203316	0.956686
NO2	29.689177	5.448778	3.864434	0.103608	0.952610
O3	49.268276	7.019136	4.625617	0.153265	0.980247
CO	0.019225	0.138656	0.098185	0.089745	0.961569

In [58]:

plt.figure(figsize=(16, 9))
plt.plot(pred_rst['PM10'].values[50:150], 'o-', label='pred')
plt.plot(real_rst['PM10'].values[50:150], '*-', label='real')
plt.legend(loc='best')

Out[58]:

<matplotlib.legend.Legend at 0x7fe5c01bc6d0>

In [74]:

model.get_weights()[-2].shape

Out[74]:

(32, 1)

In [39]:

real_rst

Out[39]:

	PM2.5	PM10	SO2	NO2	O3	CO
date
2020-02-21 17:00:00	15.0	51.000000	3.0	8.000000	87.000008	0.370
2017-07-08 03:00:00	90.0	166.000000	32.0	49.000000	99.000000	1.560
2020-07-21 05:00:00	33.0	71.000000	10.0	56.000000	32.000000	0.790
2018-06-25 15:00:00	70.0	121.999992	16.0	24.000000	165.000015	1.070
2018-02-27 12:00:00	205.0	303.000000	66.0	62.000000	92.000008	1.990
...	...	...	...	...	...	...
2016-03-21 04:00:00	131.0	218.999985	47.0	52.999996	29.000000	1.510
2016-11-16 13:00:00	251.0	382.500000	234.0	105.000000	23.500002	2.915
2018-06-13 06:00:00	30.0	123.000008	18.0	76.000000	14.000000	0.890
2020-05-15 15:00:00	36.0	94.000000	14.0	22.000000	144.000000	0.500
2020-10-15 21:00:00	67.0	103.000008	15.0	92.000000	12.000001	1.200

2451 rows × 6 columns

In [40]:

from statistics import mean
import matplotlib.pyplot as plt
from sklearn.metrics import explained_variance_score,r2_score, median_absolute_error, mean_squared_error, mean_absolute_error
from scipy import stats
import numpy as np
from matplotlib import rcParams
config = {"font.size": 32,"mathtext.fontset":'stix'}
rcParams.update(config)

In [41]:

config = {"font.size": 32,"mathtext.fontset":'stix'}
rcParams.update(config)

In [42]:

def scatter_out_1(x, y, label, name): ## x,y为两个需要做对比分析的两个量。
    # ==========计算评价指标==========
    BIAS = mean(x - y)
    MSE = mean_squared_error(x, y)
    RMSE = np.power(MSE, 0.5)
    R2 = r2_score(x, y)
    MAE = mean_absolute_error(x, y)
    EV = explained_variance_score(x, y)
    print('==========算法评价指标==========')
    print('Explained Variance(EV):', '%.3f' % (EV))
    print('Mean Absolute Error(MAE):', '%.3f' % (MAE))
    print('Mean squared error(MSE):', '%.3f' % (MSE))
    print('Root Mean Squard Error(RMSE):', '%.3f' % (RMSE))
    print('R_squared:', '%.3f' % (R2))
    # ===========Calculate the point density==========
    xy = np.vstack([x, y])
    z = stats.gaussian_kde(xy)(xy)
    # ===========Sort the points by density, so that the densest points are plotted last===========
    idx = z.argsort()
    x, y, z = x[idx], y[idx], z[idx]
    def best_fit_slope_and_intercept(xs, ys):
        m = (((mean(xs) * mean(ys)) - mean(xs * ys)) / ((mean(xs) * mean(xs)) - mean(xs * xs)))
        b = mean(ys) - m * mean(xs)
        return m, b
    m, b = best_fit_slope_and_intercept(x, y)
    regression_line = []
    for a in x:
        regression_line.append((m * a) + b)
    fig,ax=plt.subplots(figsize=(12,9),dpi=400)
    scatter=ax.scatter(x,y,marker='o',c=z*100,s=15,label='LST',cmap='Spectral_r')
    cbar=plt.colorbar(scatter,shrink=1,orientation='vertical',extend='both',pad=0.015,aspect=30,label='Frequency')
    min_value = min(min(x), min(y))
    max_value = max(max(x), max(y))

    plt.plot([min_value-5,max_value+5],[min_value-5,max_value+5],'black',lw=1.5)  # 画的1:1线，线的颜色为black，线宽为0.8
    plt.plot(x,regression_line,'red',lw=1.5)      # 预测与实测数据之间的回归线
    plt.axis([min_value-5,max_value+5,min_value-5,max_value+5])  # 设置线的范围
    plt.xlabel('Measured %s' % label)
    plt.ylabel('Retrived %s' % label)
    # plt.xticks(fontproperties='Times New Roman')
    # plt.yticks(fontproperties='Times New Roman')


    plt.text(min_value-5 + (max_value-min_value) * 0.05, int(max_value * 0.95), '$N=%.f$' % len(y)) # text的位置需要根据x,y的大小范围进行调整。
    plt.text(min_value-5 + (max_value-min_value) * 0.05, int(max_value * 0.88), '$R^2=%.2f$' % R2)
    plt.text(min_value-5 + (max_value-min_value) * 0.05, int(max_value * 0.81), '$RMSE=%.2f$' % RMSE)
    plt.xlim(min_value-5,max_value+5)                                  # 设置x坐标轴的显示范围
    plt.ylim(min_value-5,max_value+5)                                  # 设置y坐标轴的显示范围
    # file_name = name.split('(')[0].strip()
    plt.savefig(f'./figure/{name}.png',dpi=800,bbox_inches='tight',pad_inches=0)
    plt.show()

In [50]:

scatter_out_1(real_rst['PM10'].values, pred_rst['PM10'].values, label='$PM_{10}\ (\mu g/m^3$)', name='PM10')

==========算法评价指标==========
Explained Variance(EV): 0.941
Mean Absolute Error(MAE): 12.019
Mean squared error(MSE): 339.601
Root Mean Squard Error(RMSE): 18.428
R_squared: 0.939

In [51]:

scatter_out_1(real_rst['PM2.5'].values, pred_rst['PM2.5'].values, label='$PM_{2.5} (\mu g/m^3$)', name='PM25')

==========算法评价指标==========
Explained Variance(EV): 0.956
Mean Absolute Error(MAE): 6.231
Mean squared error(MSE): 108.622
Root Mean Squard Error(RMSE): 10.422
R_squared: 0.956

In [52]:

scatter_out_1(real_rst['SO2'].values, pred_rst['SO2'].values, label='$SO_2\ (\mu g/m^3)$', name='SO2')

==========算法评价指标==========
Explained Variance(EV): 0.957
Mean Absolute Error(MAE): 6.621
Mean squared error(MSE): 135.481
Root Mean Squard Error(RMSE): 11.640
R_squared: 0.957

In [53]:

scatter_out_1(real_rst['NO2'].values, pred_rst['NO2'].values, label='$NO_2\ (\mu g/m^3)$', name='NO2')

==========算法评价指标==========
Explained Variance(EV): 0.956
Mean Absolute Error(MAE): 3.864
Mean squared error(MSE): 29.689
Root Mean Squard Error(RMSE): 5.449
R_squared: 0.953

In [54]:

scatter_out_1(real_rst['O3'], pred_rst['O3'], label='$O_3 \ (\mu g/m^3)$', name='O3')

==========算法评价指标==========
Explained Variance(EV): 0.981
Mean Absolute Error(MAE): 4.626
Mean squared error(MSE): 49.268
Root Mean Squard Error(RMSE): 7.019
R_squared: 0.980

In [48]:

def scatter_out_2(x, y, name): ## x,y为两个需要做对比分析的两个量。
    # ==========计算评价指标==========
    BIAS = mean(x - y)
    MSE = mean_squared_error(x, y)
    RMSE = np.power(MSE, 0.5)
    R2 = r2_score(x, y)
    MAE = mean_absolute_error(x, y)
    EV = explained_variance_score(x, y)
    print('==========算法评价指标==========')
    print('Explained Variance(EV):', '%.3f' % (EV))
    print('Mean Absolute Error(MAE):', '%.3f' % (MAE))
    print('Mean squared error(MSE):', '%.3f' % (MSE))
    print('Root Mean Squard Error(RMSE):', '%.3f' % (RMSE))
    print('R_squared:', '%.3f' % (R2))
    # ===========Calculate the point density==========
    xy = np.vstack([x, y])
    z = stats.gaussian_kde(xy)(xy)
    # ===========Sort the points by density, so that the densest points are plotted last===========
    idx = z.argsort()
    x, y, z = x[idx], y[idx], z[idx]
    def best_fit_slope_and_intercept(xs, ys):
        m = (((mean(xs) * mean(ys)) - mean(xs * ys)) / ((mean(xs) * mean(xs)) - mean(xs * xs)))
        b = mean(ys) - m * mean(xs)
        return m, b
    m, b = best_fit_slope_and_intercept(x, y)
    regression_line = []
    for a in x:
        regression_line.append((m * a) + b)
    fig,ax=plt.subplots(figsize=(12,9),dpi=400)
    scatter=ax.scatter(x,y,marker='o',c=z*100,s=15,label='LST',cmap='Spectral_r')
    cbar=plt.colorbar(scatter,shrink=1,orientation='vertical',extend='both',pad=0.015,aspect=30,label='frequency')

    plt.plot([0, 6], [0, 6],'black',lw=1.5)  # 画的1:1线，线的颜色为black，线宽为0.8
    plt.plot(x,regression_line,'red',lw=1.5)      # 预测与实测数据之间的回归线
    plt.axis([0,6,0,6])  # 设置线的范围
    plt.xlabel(f'Measured {name}')
    plt.ylabel(f'Retrived {name}')
    # plt.xticks(fontproperties='Times New Roman')
    # plt.yticks(fontproperties='Times New Roman')


    plt.text(0.3, 5.5, '$N=%.f$' % len(y)) # text的位置需要根据x,y的大小范围进行调整。
    plt.text(0.3, 5.0, '$R^2=%.2f$' % R2)
    plt.text(0.3, 4.5, '$RMSE=%.2f$' % RMSE)
    plt.xlim(0, 6)                                  # 设置x坐标轴的显示范围
    plt.ylim(0, 6)                                  # 设置y坐标轴的显示范围
    file_name = name.split('(')[0].strip()
    plt.savefig(f'./figure/CO.png',dpi=800,bbox_inches='tight',pad_inches=0)
    plt.show()

In [49]:

scatter_out_2(np.expm1(real_rst['CO'].values), np.expm1(pred_rst['CO'].values), name='$CO \ (mg/m^3)$')

==========算法评价指标==========
Explained Variance(EV): 0.902
Mean Absolute Error(MAE): 0.700
Mean squared error(MSE): 7.938
Root Mean Squard Error(RMSE): 2.818
R_squared: 0.902

findfont: Font family ['sans-serif'] not found. Falling back to DejaVu Sans.
findfont: Generic family 'sans-serif' not found because none of the following families were found: SimHei

In [ ]: