zhaojinghao
/
coal_materials
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
coal_materials/.ipynb_checkpoints/multi-task-NN-checkpoint.ipynb

62 KiB
Raw Blame History

In [1]:
import os
os.environ['CUDA_DEVICE_ORDER'] = 'PCB_BUS_ID'
os.environ['CUDA_VISIBLE_DEVICES'] = '0, 1'
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('./data/20240102/train_data.csv')
In [4]:
out_cols = [x for x in data.columns if '碳材料' in x]
In [5]:
out_cols
Out[5]:
['碳材料结构特征-比表面积', '碳材料结构特征-总孔体积', '碳材料结构特征-微孔体积', '碳材料结构特征-平均孔径']
In [6]:
train_data = data.dropna(subset=out_cols).fillna(0)
In [7]:
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
import tensorflow.keras.backend as K

2024-01-04 16:14:39.388684: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcudart.so.11.0
In [8]:
tf.test.is_gpu_available()

WARNING:tensorflow:From /tmp/ipykernel_43672/337460670.py:1: is_gpu_available (from tensorflow.python.framework.test_util) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.config.list_physical_devices('GPU')` instead.

2024-01-04 16:14:40.311876: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 AVX512F FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-01-04 16:14:40.319726: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcuda.so.1
2024-01-04 16:14:40.406804: E tensorflow/stream_executor/cuda/cuda_driver.cc:328] failed call to cuInit: CUDA_ERROR_INVALID_DEVICE: invalid device ordinal
2024-01-04 16:14:40.406829: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:169] retrieving CUDA diagnostic information for host: zhaojh-yv621
2024-01-04 16:14:40.406833: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:176] hostname: zhaojh-yv621
2024-01-04 16:14:40.406963: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:200] libcuda reported version is: 520.61.5
2024-01-04 16:14:40.406982: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:204] kernel reported version is: 520.61.5
2024-01-04 16:14:40.406985: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:310] kernel version seems to match DSO: 520.61.5
Out[8]:
False
In [9]:
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 [10]:
from tensorflow.keras import Model
In [11]:
from tensorflow.keras.initializers import Constant
In [12]:
# Custom loss layer
class CustomMultiLossLayer(layers.Layer):
    def __init__(self, nb_outputs=2, **kwargs):
        self.nb_outputs = nb_outputs
        self.is_placeholder = True
        super(CustomMultiLossLayer, self).__init__(**kwargs)
        
    def build(self, input_shape=None):
        # initialise log_vars
        self.log_vars = []
        for i in range(self.nb_outputs):
            self.log_vars += [self.add_weight(name='log_var' + str(i), shape=(1,),
                                              initializer=tf.initializers.he_normal(), trainable=True)]
        super(CustomMultiLossLayer, self).build(input_shape)

    def multi_loss(self, ys_true, ys_pred):
        assert len(ys_true) == self.nb_outputs and len(ys_pred) == self.nb_outputs
        loss = 0
        for y_true, y_pred, log_var in zip(ys_true, ys_pred, self.log_vars):
            mse = (y_true - y_pred) ** 2.
            pre = K.exp(-log_var[0])
            loss += tf.abs(tf.reduce_logsumexp(pre * mse + log_var[0], axis=-1))
        return K.mean(loss)

    def call(self, inputs):
        ys_true = inputs[:self.nb_outputs]
        ys_pred = inputs[self.nb_outputs:]
        loss = self.multi_loss(ys_true, ys_pred)
        self.add_loss(loss, inputs=inputs)
        # We won't actually use the output.
        return K.concatenate(inputs, -1)
In [13]:
num_heads, ff_dim = 1, 12
In [38]:
def get_prediction_model():
    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)

    bet = build_output(out, 'bet')
    mesco = build_output(out, 'mesco')
    micro = build_output(out, 'micro')
    avg = build_output(out, 'avg')

    bet = layers.Dense(1, activation='sigmoid', name='bet')(bet)
    mesco = layers.Dense(1, activation='sigmoid', name='mesco')(mesco)
    micro = layers.Dense(1, activation='sigmoid', name='micro')(micro)
    avg = layers.Dense(1, activation='sigmoid', name='avg')(avg)

    model = Model(inputs=[inputs], outputs=[bet, mesco, micro, avg])
    return model
In [39]:
def get_trainable_model(prediction_model):
    inputs = layers.Input(shape=(1,len(feature_cols)), name='input')
    bet, mesco, micro, avg = prediction_model(inputs)
    bet_real = layers.Input(shape=(1,), name='bet_real')
    mesco_real = layers.Input(shape=(1,), name='mesco_real')
    micro_real = layers.Input(shape=(1,), name='micro_real')
    avg_real = layers.Input(shape=(1,), name='avg_real')
    out = CustomMultiLossLayer(nb_outputs=4)([bet_real, mesco_real, micro_real, avg_real, bet, mesco, micro, avg])
    return Model([inputs, bet_real, mesco_real, micro_real, avg_real], out)
In [40]:
maxs = train_data.max()
mins = train_data.min()
for col in train_data.columns:
    if maxs[col] - mins[col] == 0:
        continue
    train_data[col] = (train_data[col] - mins[col]) / (maxs[col] - mins[col])
In [41]:
train_data
Out[41]:
热处理条件-热处理次数 热处理条件-是否是中温停留 第一次热处理-温度 第一次热处理-升温速率 第一次热处理-保留时间 第二次热处理-温度 第二次热处理-升温速率· 第二次热处理-保留时间 共碳化-是否是共碳化物质 共碳化-共碳化物质/沥青 ... 模板剂-种类_二氧化硅 模板剂-种类_氢氧化镁 模板剂-种类_氧化钙 模板剂-种类_氧化锌 模板剂-种类_氧化镁 模板剂-种类_氯化钠 模板剂-种类_氯化钾 模板剂-种类_碱式碳酸镁 模板剂-种类_碳酸钙 模板剂-种类_纤维素
0 0.0 0.0 0.166667 0.3 0.5 0.000000 0.0 0.000000 0.0 0.0 ... 0 0.0 1.0 0 0.0 0.0 0 0.0 0.0 0.0
1 0.0 0.0 0.333333 0.3 0.5 0.000000 0.0 0.000000 0.0 0.0 ... 0 0.0 1.0 0 0.0 0.0 0 0.0 0.0 0.0
2 0.0 0.0 0.333333 0.3 0.5 0.000000 0.0 0.000000 0.0 0.0 ... 0 0.0 1.0 0 0.0 0.0 0 0.0 0.0 0.0
3 0.0 0.0 0.333333 0.3 0.5 0.000000 0.0 0.000000 0.0 0.0 ... 0 0.0 1.0 0 0.0 0.0 0 0.0 0.0 0.0
4 1.0 0.0 0.166667 0.3 0.5 0.666667 0.5 0.666667 0.0 0.0 ... 0 0.0 0.0 0 0.0 0.0 0 1.0 0.0 0.0
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
144 0.0 0.0 0.333333 0.3 0.0 0.000000 0.0 0.000000 0.0 0.0 ... 0 0.0 0.0 0 0.0 0.0 0 0.0 0.0 0.0
145 0.0 0.0 0.500000 0.3 0.0 0.000000 0.0 0.000000 0.0 0.0 ... 0 0.0 0.0 0 0.0 0.0 0 0.0 0.0 0.0
146 0.0 0.0 0.666667 0.3 0.0 0.000000 0.0 0.000000 0.0 0.0 ... 0 0.0 0.0 0 0.0 0.0 0 0.0 0.0 0.0
147 0.0 0.0 0.500000 0.3 0.0 0.000000 0.0 0.000000 0.0 0.0 ... 0 0.0 0.0 0 0.0 0.0 0 0.0 0.0 0.0
148 0.0 0.0 0.500000 0.3 0.0 0.000000 0.0 0.000000 0.0 0.0 ... 0 0.0 0.0 0 0.0 0.0 0 0.0 0.0 0.0

123 rows × 42 columns

In [42]:
# feature_cols = [x for x in train_data.columns if x not in out_cols and '第二次' not in x]
feature_cols = [x for x in train_data.columns if x not in out_cols]
use_cols = feature_cols + out_cols
In [43]:
use_data = train_data.copy()
for col in use_cols:
    use_data[col] = use_data[col].astype('float32')
In [44]:
train, valid = train_test_split(use_data[use_cols], test_size=0.2, random_state=42, shuffle=True)
valid, test = train_test_split(valid, test_size=0.5, random_state=42, shuffle=True)
In [45]:
prediction_model = get_prediction_model()
trainable_model = get_trainable_model(prediction_model)
In [46]:
prediction_model.summary()

Model: "model_4"
__________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================
input (InputLayer)              [(None, 1, 38)]      0                                            
__________________________________________________________________________________________________
conv1d_3 (Conv1D)               (None, 1, 64)        2496        input[0][0]                      
__________________________________________________________________________________________________
bidirectional_3 (Bidirectional) (None, 1, 128)       66048       conv1d_3[0][0]                   
__________________________________________________________________________________________________
dense_28 (Dense)                (None, 1, 128)       16512       bidirectional_3[0][0]            
__________________________________________________________________________________________________
transformer_block_7 (Transforme (None, 1, 128)       202640      dense_28[0][0]                   
__________________________________________________________________________________________________
global_average_pooling1d_7 (Glo (None, 128)          0           transformer_block_7[0][0]        
__________________________________________________________________________________________________
dropout_23 (Dropout)            (None, 128)          0           global_average_pooling1d_7[0][0] 
__________________________________________________________________________________________________
dense_31 (Dense)                (None, 64)           8256        dropout_23[0][0]                 
__________________________________________________________________________________________________
tf.expand_dims_3 (TFOpLambda)   (None, 1, 64)        0           dense_31[0][0]                   
__________________________________________________________________________________________________
transformer_block_8 (Transforme (None, 1, 64)        52176       tf.expand_dims_3[0][0]           
__________________________________________________________________________________________________
transformer_block_9 (Transforme (None, 1, 64)        52176       tf.expand_dims_3[0][0]           
__________________________________________________________________________________________________
transformer_block_10 (Transform (None, 1, 64)        52176       tf.expand_dims_3[0][0]           
__________________________________________________________________________________________________
transformer_block_11 (Transform (None, 1, 64)        52176       tf.expand_dims_3[0][0]           
__________________________________________________________________________________________________
global_average_pooling1d_8 (Glo (None, 64)           0           transformer_block_8[0][0]        
__________________________________________________________________________________________________
global_average_pooling1d_9 (Glo (None, 64)           0           transformer_block_9[0][0]        
__________________________________________________________________________________________________
global_average_pooling1d_10 (Gl (None, 64)           0           transformer_block_10[0][0]       
__________________________________________________________________________________________________
global_average_pooling1d_11 (Gl (None, 64)           0           transformer_block_11[0][0]       
__________________________________________________________________________________________________
dense_34 (Dense)                (None, 32)           2080        global_average_pooling1d_8[0][0] 
__________________________________________________________________________________________________
dense_37 (Dense)                (None, 32)           2080        global_average_pooling1d_9[0][0] 
__________________________________________________________________________________________________
dense_40 (Dense)                (None, 32)           2080        global_average_pooling1d_10[0][0]
__________________________________________________________________________________________________
dense_43 (Dense)                (None, 32)           2080        global_average_pooling1d_11[0][0]
__________________________________________________________________________________________________
bet (Dense)                     (None, 1)            33          dense_34[0][0]                   
__________________________________________________________________________________________________
mesco (Dense)                   (None, 1)            33          dense_37[0][0]                   
__________________________________________________________________________________________________
micro (Dense)                   (None, 1)            33          dense_40[0][0]                   
__________________________________________________________________________________________________
avg (Dense)                     (None, 1)            33          dense_43[0][0]                   
==================================================================================================
Total params: 513,108
Trainable params: 513,108
Non-trainable params: 0
__________________________________________________________________________________________________
In [47]:
from tensorflow.keras import optimizers
from tensorflow.python.keras.utils.vis_utils import plot_model
In [48]:
X = np.expand_dims(train[feature_cols].values, axis=1)
Y = [x for x in train[out_cols].values.T]
Y_valid = [x for x in valid[out_cols].values.T]
In [49]:
from keras.callbacks import ReduceLROnPlateau
reduce_lr = ReduceLROnPlateau(monitor='val_loss', patience=10, mode='auto')
In [50]:
trainable_model.compile(optimizer='adam', loss=None)
hist = trainable_model.fit([X, Y[0], Y[1], Y[2], Y[3]], epochs=160, batch_size=8, verbose=1, 
                           validation_data=[np.expand_dims(valid[feature_cols].values, axis=1), Y_valid[0], Y_valid[1], Y_valid[2], Y_valid[3]],
                           callbacks=[reduce_lr]
                           )

2024-01-04 16:17:21.543163: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:176] None of the MLIR Optimization Passes are enabled (registered 2)
2024-01-04 16:17:21.562835: I tensorflow/core/platform/profile_utils/cpu_utils.cc:114] CPU Frequency: 2200000000 Hz

Epoch 1/160
13/13 [==============================] - 6s 103ms/step - loss: 5.1128 - val_loss: 4.4845
Epoch 2/160
13/13 [==============================] - 0s 30ms/step - loss: 4.4173 - val_loss: 4.3305
Epoch 3/160
13/13 [==============================] - 0s 32ms/step - loss: 4.0913 - val_loss: 4.4123
Epoch 4/160
13/13 [==============================] - 0s 32ms/step - loss: 4.0410 - val_loss: 4.3142
Epoch 5/160
13/13 [==============================] - 0s 33ms/step - loss: 3.9933 - val_loss: 4.5518
Epoch 6/160
13/13 [==============================] - 0s 29ms/step - loss: 4.1234 - val_loss: 4.3268
Epoch 7/160
13/13 [==============================] - 0s 35ms/step - loss: 4.0470 - val_loss: 4.2908
Epoch 8/160
13/13 [==============================] - 0s 32ms/step - loss: 4.1230 - val_loss: 4.4964
Epoch 9/160
13/13 [==============================] - 0s 33ms/step - loss: 3.8889 - val_loss: 4.0178
Epoch 10/160
13/13 [==============================] - 0s 33ms/step - loss: 3.6648 - val_loss: 3.9010
Epoch 11/160
13/13 [==============================] - 0s 32ms/step - loss: 3.7712 - val_loss: 3.9471
Epoch 12/160
13/13 [==============================] - 0s 32ms/step - loss: 3.5449 - val_loss: 3.8723
Epoch 13/160
13/13 [==============================] - 0s 26ms/step - loss: 3.3373 - val_loss: 3.8543
Epoch 14/160
13/13 [==============================] - 0s 28ms/step - loss: 3.5200 - val_loss: 3.8259
Epoch 15/160
13/13 [==============================] - 0s 28ms/step - loss: 3.5623 - val_loss: 3.8838
Epoch 16/160
13/13 [==============================] - 0s 29ms/step - loss: 3.3898 - val_loss: 3.8122
Epoch 17/160
13/13 [==============================] - 0s 35ms/step - loss: 3.2718 - val_loss: 3.8799
Epoch 18/160
13/13 [==============================] - 0s 32ms/step - loss: 3.3303 - val_loss: 3.7849
Epoch 19/160
13/13 [==============================] - 0s 26ms/step - loss: 3.2860 - val_loss: 3.7713
Epoch 20/160
13/13 [==============================] - 0s 34ms/step - loss: 3.2669 - val_loss: 3.5689
Epoch 21/160
13/13 [==============================] - 0s 34ms/step - loss: 3.2366 - val_loss: 3.5238
Epoch 22/160
13/13 [==============================] - 0s 33ms/step - loss: 3.0037 - val_loss: 3.6039
Epoch 23/160
13/13 [==============================] - 0s 32ms/step - loss: 3.2087 - val_loss: 3.5221
Epoch 24/160
13/13 [==============================] - 0s 32ms/step - loss: 3.0619 - val_loss: 3.5939
Epoch 25/160
13/13 [==============================] - 0s 32ms/step - loss: 3.0423 - val_loss: 3.2731
Epoch 26/160
13/13 [==============================] - 0s 33ms/step - loss: 3.0533 - val_loss: 3.2256
Epoch 27/160
13/13 [==============================] - 0s 32ms/step - loss: 3.0105 - val_loss: 3.2154
Epoch 28/160
13/13 [==============================] - 0s 33ms/step - loss: 2.9607 - val_loss: 3.2926
Epoch 29/160
13/13 [==============================] - 0s 28ms/step - loss: 3.0072 - val_loss: 3.5834
Epoch 30/160
13/13 [==============================] - 0s 28ms/step - loss: 2.9276 - val_loss: 3.1635
Epoch 31/160
13/13 [==============================] - 0s 28ms/step - loss: 2.8930 - val_loss: 3.1363
Epoch 32/160
13/13 [==============================] - 0s 28ms/step - loss: 2.7904 - val_loss: 3.0188
Epoch 33/160
13/13 [==============================] - 0s 35ms/step - loss: 2.6856 - val_loss: 2.9808
Epoch 34/160
13/13 [==============================] - 0s 32ms/step - loss: 2.6503 - val_loss: 3.0943
Epoch 35/160
13/13 [==============================] - 0s 32ms/step - loss: 2.5339 - val_loss: 2.9359
Epoch 36/160
13/13 [==============================] - 0s 33ms/step - loss: 2.5369 - val_loss: 2.9704
Epoch 37/160
13/13 [==============================] - 0s 32ms/step - loss: 2.5478 - val_loss: 2.9344
Epoch 38/160
13/13 [==============================] - 0s 32ms/step - loss: 2.7129 - val_loss: 2.8326
Epoch 39/160
13/13 [==============================] - 0s 33ms/step - loss: 2.3453 - val_loss: 2.8198
Epoch 40/160
13/13 [==============================] - 0s 33ms/step - loss: 2.4666 - val_loss: 2.7701
Epoch 41/160
13/13 [==============================] - 0s 32ms/step - loss: 2.3401 - val_loss: 2.7727
Epoch 42/160
13/13 [==============================] - 0s 28ms/step - loss: 2.4369 - val_loss: 2.7568
Epoch 43/160
13/13 [==============================] - 0s 28ms/step - loss: 2.2170 - val_loss: 2.6998
Epoch 44/160
13/13 [==============================] - 0s 33ms/step - loss: 2.1565 - val_loss: 2.6711
Epoch 45/160
13/13 [==============================] - 0s 28ms/step - loss: 2.1251 - val_loss: 2.6134
Epoch 46/160
13/13 [==============================] - 0s 30ms/step - loss: 2.1728 - val_loss: 2.6394
Epoch 47/160
13/13 [==============================] - 0s 28ms/step - loss: 2.0609 - val_loss: 2.6568
Epoch 48/160
13/13 [==============================] - 0s 28ms/step - loss: 2.0893 - val_loss: 2.6603
Epoch 49/160
13/13 [==============================] - 0s 28ms/step - loss: 2.0615 - val_loss: 2.6517
Epoch 50/160
13/13 [==============================] - 0s 33ms/step - loss: 2.0500 - val_loss: 2.6041
Epoch 51/160
13/13 [==============================] - 0s 28ms/step - loss: 1.9309 - val_loss: 2.6218
Epoch 52/160
13/13 [==============================] - 0s 28ms/step - loss: 1.9971 - val_loss: 2.4494
Epoch 53/160
13/13 [==============================] - 0s 35ms/step - loss: 1.8514 - val_loss: 2.3886
Epoch 54/160
13/13 [==============================] - 0s 33ms/step - loss: 1.7823 - val_loss: 2.5517
Epoch 55/160
13/13 [==============================] - 0s 32ms/step - loss: 1.7059 - val_loss: 2.3293
Epoch 56/160
13/13 [==============================] - 0s 33ms/step - loss: 1.7039 - val_loss: 2.3810
Epoch 57/160
13/13 [==============================] - 0s 34ms/step - loss: 1.6961 - val_loss: 2.4552
Epoch 58/160
13/13 [==============================] - 0s 31ms/step - loss: 1.7350 - val_loss: 2.3526
Epoch 59/160
13/13 [==============================] - 0s 32ms/step - loss: 1.5840 - val_loss: 2.2976
Epoch 60/160
13/13 [==============================] - 0s 28ms/step - loss: 1.4915 - val_loss: 2.3516
Epoch 61/160
13/13 [==============================] - 0s 28ms/step - loss: 1.5910 - val_loss: 2.2383
Epoch 62/160
13/13 [==============================] - 0s 29ms/step - loss: 1.6101 - val_loss: 2.1474
Epoch 63/160
13/13 [==============================] - 0s 34ms/step - loss: 1.4698 - val_loss: 2.1210
Epoch 64/160
13/13 [==============================] - 0s 31ms/step - loss: 1.4796 - val_loss: 2.0695
Epoch 65/160
13/13 [==============================] - 0s 33ms/step - loss: 1.4472 - val_loss: 1.9768
Epoch 66/160
13/13 [==============================] - 0s 33ms/step - loss: 1.6250 - val_loss: 2.1760
Epoch 67/160
13/13 [==============================] - 0s 33ms/step - loss: 1.4058 - val_loss: 2.0605
Epoch 68/160
13/13 [==============================] - 0s 33ms/step - loss: 1.4318 - val_loss: 2.1487
Epoch 69/160
13/13 [==============================] - 0s 28ms/step - loss: 1.3285 - val_loss: 1.8259
Epoch 70/160
13/13 [==============================] - 0s 28ms/step - loss: 1.4306 - val_loss: 1.7314
Epoch 71/160
13/13 [==============================] - 0s 28ms/step - loss: 1.3449 - val_loss: 1.7509
Epoch 72/160
13/13 [==============================] - 0s 32ms/step - loss: 1.2737 - val_loss: 1.8892
Epoch 73/160
13/13 [==============================] - 0s 32ms/step - loss: 1.3647 - val_loss: 1.8109
Epoch 74/160
13/13 [==============================] - 0s 28ms/step - loss: 1.3075 - val_loss: 1.8175
Epoch 75/160
13/13 [==============================] - 0s 28ms/step - loss: 1.2765 - val_loss: 1.7334
Epoch 76/160
13/13 [==============================] - 0s 28ms/step - loss: 1.2761 - val_loss: 1.7685
Epoch 77/160
13/13 [==============================] - 0s 28ms/step - loss: 1.1870 - val_loss: 1.7683
Epoch 78/160
13/13 [==============================] - 0s 28ms/step - loss: 1.1441 - val_loss: 1.8794
Epoch 79/160
13/13 [==============================] - 0s 28ms/step - loss: 1.1193 - val_loss: 1.9073
Epoch 80/160
13/13 [==============================] - 0s 35ms/step - loss: 1.1361 - val_loss: 1.8016
Epoch 81/160
13/13 [==============================] - 0s 26ms/step - loss: 1.0629 - val_loss: 1.8359
Epoch 82/160
13/13 [==============================] - 0s 29ms/step - loss: 1.1137 - val_loss: 1.9310
Epoch 83/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0947 - val_loss: 1.9212
Epoch 84/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0225 - val_loss: 1.9027
Epoch 85/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0892 - val_loss: 1.8943
Epoch 86/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0387 - val_loss: 1.9131
Epoch 87/160
13/13 [==============================] - 0s 34ms/step - loss: 1.0590 - val_loss: 1.8768
Epoch 88/160
13/13 [==============================] - 0s 27ms/step - loss: 1.0909 - val_loss: 1.8732
Epoch 89/160
13/13 [==============================] - 0s 34ms/step - loss: 1.0632 - val_loss: 1.8506
Epoch 90/160
13/13 [==============================] - 0s 26ms/step - loss: 1.0703 - val_loss: 1.8108
Epoch 91/160
13/13 [==============================] - 0s 33ms/step - loss: 1.0262 - val_loss: 1.8143
Epoch 92/160
13/13 [==============================] - 0s 27ms/step - loss: 1.0330 - val_loss: 1.8132
Epoch 93/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0440 - val_loss: 1.8156
Epoch 94/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0279 - val_loss: 1.8182
Epoch 95/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0713 - val_loss: 1.8191
Epoch 96/160
13/13 [==============================] - 0s 29ms/step - loss: 1.0401 - val_loss: 1.8187
Epoch 97/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0318 - val_loss: 1.8193
Epoch 98/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0306 - val_loss: 1.8233
Epoch 99/160
13/13 [==============================] - 0s 29ms/step - loss: 1.0348 - val_loss: 1.8268
Epoch 100/160
13/13 [==============================] - 0s 35ms/step - loss: 1.0165 - val_loss: 1.8276
Epoch 101/160
13/13 [==============================] - 0s 32ms/step - loss: 1.0826 - val_loss: 1.8275
Epoch 102/160
13/13 [==============================] - 0s 32ms/step - loss: 1.0631 - val_loss: 1.8269
Epoch 103/160
13/13 [==============================] - 0s 31ms/step - loss: 0.9980 - val_loss: 1.8268
Epoch 104/160
13/13 [==============================] - 0s 29ms/step - loss: 1.0245 - val_loss: 1.8270
Epoch 105/160
13/13 [==============================] - 0s 29ms/step - loss: 1.1240 - val_loss: 1.8272
Epoch 106/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0099 - val_loss: 1.8275
Epoch 107/160
13/13 [==============================] - 0s 32ms/step - loss: 1.0435 - val_loss: 1.8272
Epoch 108/160
13/13 [==============================] - 0s 32ms/step - loss: 1.0166 - val_loss: 1.8257
Epoch 109/160
13/13 [==============================] - 0s 33ms/step - loss: 1.0462 - val_loss: 1.8256
Epoch 110/160
13/13 [==============================] - 0s 29ms/step - loss: 1.1261 - val_loss: 1.8256
Epoch 111/160
13/13 [==============================] - 0s 29ms/step - loss: 1.0418 - val_loss: 1.8256
Epoch 112/160
13/13 [==============================] - 0s 27ms/step - loss: 0.9933 - val_loss: 1.8255
Epoch 113/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0758 - val_loss: 1.8255
Epoch 114/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0435 - val_loss: 1.8254
Epoch 115/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0265 - val_loss: 1.8254
Epoch 116/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0268 - val_loss: 1.8254
Epoch 117/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0709 - val_loss: 1.8254
Epoch 118/160
13/13 [==============================] - 0s 36ms/step - loss: 1.0304 - val_loss: 1.8253
Epoch 119/160
13/13 [==============================] - 0s 25ms/step - loss: 1.0074 - val_loss: 1.8254
Epoch 120/160
13/13 [==============================] - 0s 29ms/step - loss: 1.0375 - val_loss: 1.8254
Epoch 121/160
13/13 [==============================] - 0s 29ms/step - loss: 1.0196 - val_loss: 1.8254
Epoch 122/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0706 - val_loss: 1.8254
Epoch 123/160
13/13 [==============================] - 0s 29ms/step - loss: 1.0120 - val_loss: 1.8254
Epoch 124/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0962 - val_loss: 1.8254
Epoch 125/160
13/13 [==============================] - 0s 28ms/step - loss: 0.9942 - val_loss: 1.8254
Epoch 126/160
13/13 [==============================] - 0s 33ms/step - loss: 1.0419 - val_loss: 1.8254
Epoch 127/160
13/13 [==============================] - 0s 32ms/step - loss: 1.1072 - val_loss: 1.8254
Epoch 128/160
13/13 [==============================] - 0s 33ms/step - loss: 1.0153 - val_loss: 1.8254
Epoch 129/160
13/13 [==============================] - 0s 29ms/step - loss: 1.0324 - val_loss: 1.8254
Epoch 130/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0363 - val_loss: 1.8254
Epoch 131/160
13/13 [==============================] - 0s 35ms/step - loss: 1.0624 - val_loss: 1.8254
Epoch 132/160
13/13 [==============================] - 0s 34ms/step - loss: 1.1191 - val_loss: 1.8254
Epoch 133/160
13/13 [==============================] - 0s 32ms/step - loss: 1.0297 - val_loss: 1.8254
Epoch 134/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0494 - val_loss: 1.8254
Epoch 135/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0162 - val_loss: 1.8254
Epoch 136/160
13/13 [==============================] - 0s 32ms/step - loss: 0.9976 - val_loss: 1.8254
Epoch 137/160
13/13 [==============================] - 0s 32ms/step - loss: 1.0401 - val_loss: 1.8254
Epoch 138/160
13/13 [==============================] - 0s 28ms/step - loss: 0.9879 - val_loss: 1.8254
Epoch 139/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0398 - val_loss: 1.8254
Epoch 140/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0297 - val_loss: 1.8254
Epoch 141/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0344 - val_loss: 1.8254
Epoch 142/160
13/13 [==============================] - 0s 32ms/step - loss: 1.0372 - val_loss: 1.8254
Epoch 143/160
13/13 [==============================] - 0s 34ms/step - loss: 1.0513 - val_loss: 1.8254
Epoch 144/160
13/13 [==============================] - 0s 34ms/step - loss: 1.0447 - val_loss: 1.8254
Epoch 145/160
13/13 [==============================] - 0s 26ms/step - loss: 1.0532 - val_loss: 1.8254
Epoch 146/160
13/13 [==============================] - 0s 29ms/step - loss: 1.0670 - val_loss: 1.8254
Epoch 147/160
13/13 [==============================] - 0s 28ms/step - loss: 1.0499 - val_loss: 1.8254
Epoch 148/160
13/13 [==============================] - 0s 34ms/step - loss: 1.0295 - val_loss: 1.8254
Epoch 149/160
13/13 [==============================] - 0s 27ms/step - loss: 1.1065 - val_loss: 1.8254
Epoch 150/160
13/13 [==============================] - 0s 27ms/step - loss: 1.1115 - val_loss: 1.8254
Epoch 151/160
13/13 [==============================] - 0s 32ms/step - loss: 1.1047 - val_loss: 1.8254
Epoch 152/160
13/13 [==============================] - 0s 26ms/step - loss: 1.0098 - val_loss: 1.8254
Epoch 153/160
13/13 [==============================] - 0s 34ms/step - loss: 0.9954 - val_loss: 1.8254
Epoch 154/160
13/13 [==============================] - 0s 27ms/step - loss: 1.0815 - val_loss: 1.8254
Epoch 155/160
13/13 [==============================] - 0s 28ms/step - loss: 1.1248 - val_loss: 1.8254
Epoch 156/160
13/13 [==============================] - 0s 35ms/step - loss: 1.0116 - val_loss: 1.8254
Epoch 157/160
13/13 [==============================] - 0s 31ms/step - loss: 1.0502 - val_loss: 1.8254
Epoch 158/160
13/13 [==============================] - 0s 33ms/step - loss: 1.0578 - val_loss: 1.8254
Epoch 159/160
13/13 [==============================] - 0s 32ms/step - loss: 1.0356 - val_loss: 1.8254
Epoch 160/160
13/13 [==============================] - 0s 27ms/step - loss: 1.0162 - val_loss: 1.8254
In [27]:
rst = prediction_model.predict(np.expand_dims(test[feature_cols], axis=1))
rst
Out[27]:
[array([[0.00237915],
        [0.21877757],
        [0.00211415],
        [0.00235748],
        [0.2187585 ],
        [0.25013232],
        [0.00218698],
        [0.00213578],
        [0.0021604 ],
        [0.00213698],
        [0.00215602],
        [0.00211859],
        [0.00211859]], dtype=float32),
 array([[0.26313323],
        [0.43726084],
        [0.257788  ],
        [0.2622419 ],
        [0.43731606],
        [0.41615662],
        [0.2588436 ],
        [0.2605151 ],
        [0.2610975 ],
        [0.26035452],
        [0.25860977],
        [0.25888485],
        [0.2590733 ]], dtype=float32),
 array([[0.03315076],
        [0.43969226],
        [0.0066632 ],
        [0.0311569 ],
        [0.4396916 ],
        [0.46122804],
        [0.01751196],
        [0.0046435 ],
        [0.00397068],
        [0.00480857],
        [0.01166728],
        [0.00597936],
        [0.00580207]], dtype=float32),
 array([[0.2627051 ],
        [0.25722986],
        [0.30297792],
        [0.26330546],
        [0.25718838],
        [0.30138326],
        [0.27484083],
        [0.3198207 ],
        [0.3352574 ],
        [0.31860778],
        [0.28594404],
        [0.30712652],
        [0.3081199 ]], dtype=float32)]
In [28]:
[np.exp(K.get_value(log_var[0]))**0.5 for log_var in trainable_model.layers[-1].log_vars]
Out[28]:
[0.44671712167235617,
 0.995773503303174,
 0.8775154468883085,
 0.9863306026616467]
In [29]:
pred_rst = pd.DataFrame.from_records(np.squeeze(np.asarray(rst), axis=2).T, columns=out_cols)
In [30]:
real_rst = test[out_cols].copy()
In [31]:
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 [32]:
real_rst.columns
Out[32]:
Index(['碳材料结构特征-比表面积', '碳材料结构特征-总孔体积', '碳材料结构特征-微孔体积', '碳材料结构特征-平均孔径'], dtype='object')
In [33]:
y_pred_pm25 = pred_rst['碳材料结构特征-比表面积'].values.reshape(-1,)
y_pred_pm10 = pred_rst['碳材料结构特征-总孔体积'].values.reshape(-1,)
y_pred_so2 = pred_rst['碳材料结构特征-微孔体积'].values.reshape(-1,)
y_pred_no2 = pred_rst['碳材料结构特征-平均孔径'].values.reshape(-1,)
y_true_pm25 = real_rst['碳材料结构特征-比表面积'].values.reshape(-1,)
y_true_pm10 = real_rst['碳材料结构特征-总孔体积'].values.reshape(-1,)
y_true_so2 = real_rst['碳材料结构特征-微孔体积'].values.reshape(-1,)
y_true_no2 = real_rst['碳材料结构特征-平均孔径'].values.reshape(-1,)
In [34]:
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score, mean_absolute_percentage_error
In [35]:
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 [36]:
pm25_eva = print_eva(y_true_pm25, y_pred_pm25, tp='比表面积')
pm10_eva = print_eva(y_true_pm10, y_pred_pm10, tp='总孔体积')
so2_eva = print_eva(y_true_so2, y_pred_so2, tp='微孔体积')
nox_eva = print_eva(y_true_no2, y_pred_no2, tp='平均孔径')

COL: 比表面积, MSE: 2.59E+06,RMSE: 1609.7549,MAPE: 90.44 %,MAE: 1456.9479,R_2: -3.4577
COL: 总孔体积, MSE: 2.74E-01,RMSE: 0.5234,MAPE: 36.559999999999995 %,MAE: 0.4001,R_2: 0.1427
COL: 微孔体积, MSE: 1.45E-01,RMSE: 0.3802,MAPE: 77.27000000000001 %,MAE: 0.324,R_2: -2.0216
COL: 平均孔径, MSE: 1.44E+00,RMSE: 1.201,MAPE: 42.24 %,MAE: 1.0489,R_2: -0.0048
In [ ]:
In [ ]: