zhaojinghao
/
coal_materials
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
coal_materials/multi-task0102.ipynb

1410 lines
90 KiB
Plaintext
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "6b84fefd-5936-4da4-ab6b-5b944329ad1d",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"os.environ['CUDA_DEVICE_ORDER'] = 'PCB_BUS_ID'\n",
"os.environ['CUDA_VISIBLE_DEVICES'] = '0, 1'"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "9cf130e3-62ef-46e0-bbdc-b13d9d29318d",
"metadata": {},
"outputs": [],
"source": [
"import pandas as pd\n",
"import numpy as np\n",
"from sklearn.model_selection import train_test_split\n",
"import matplotlib.pyplot as plt\n",
"#新增加的两行\n",
"from pylab import mpl\n",
"# 设置显示中文字体\n",
"mpl.rcParams[\"font.sans-serif\"] = [\"SimHei\"]\n",
"\n",
"mpl.rcParams[\"axes.unicode_minus\"] = False"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "752381a5-0aeb-4c54-bc48-f9c3f8fc5d17",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead tr th {\n",
" text-align: left;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr>\n",
" <th></th>\n",
" <th>Unnamed: 0_level_0</th>\n",
" <th>氢</th>\n",
" <th>碳</th>\n",
" <th>氮</th>\n",
" <th>氧</th>\n",
" <th>弹筒发热量</th>\n",
" <th>挥发分</th>\n",
" <th>固定炭</th>\n",
" </tr>\n",
" <tr>\n",
" <th></th>\n",
" <th>化验编号</th>\n",
" <th>Had</th>\n",
" <th>Cad</th>\n",
" <th>Nad</th>\n",
" <th>Oad</th>\n",
" <th>Qb,ad</th>\n",
" <th>Vad</th>\n",
" <th>Fcad</th>\n",
" </tr>\n",
" <tr>\n",
" <th></th>\n",
" <th>Unnamed: 0_level_2</th>\n",
" <th>(%)</th>\n",
" <th>(%)</th>\n",
" <th>(%)</th>\n",
" <th>(%)</th>\n",
" <th>MJ/kg</th>\n",
" <th>(%)</th>\n",
" <th>(%)</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>2720110529</td>\n",
" <td>3.93</td>\n",
" <td>70.18</td>\n",
" <td>0.81</td>\n",
" <td>25.079</td>\n",
" <td>27.820</td>\n",
" <td>32.06</td>\n",
" <td>55.68</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>2720096883</td>\n",
" <td>3.78</td>\n",
" <td>68.93</td>\n",
" <td>0.77</td>\n",
" <td>26.512</td>\n",
" <td>27.404</td>\n",
" <td>29.96</td>\n",
" <td>54.71</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>2720109084</td>\n",
" <td>3.48</td>\n",
" <td>69.60</td>\n",
" <td>0.76</td>\n",
" <td>26.148</td>\n",
" <td>27.578</td>\n",
" <td>29.31</td>\n",
" <td>55.99</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>2720084708</td>\n",
" <td>3.47</td>\n",
" <td>66.71</td>\n",
" <td>0.76</td>\n",
" <td>29.055</td>\n",
" <td>26.338</td>\n",
" <td>28.58</td>\n",
" <td>53.87</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>2720062721</td>\n",
" <td>3.87</td>\n",
" <td>68.78</td>\n",
" <td>0.80</td>\n",
" <td>26.542</td>\n",
" <td>27.280</td>\n",
" <td>29.97</td>\n",
" <td>54.78</td>\n",
" </tr>\n",
" <tr>\n",
" <th>...</th>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" </tr>\n",
" <tr>\n",
" <th>223</th>\n",
" <td>2720030490</td>\n",
" <td>4.12</td>\n",
" <td>68.85</td>\n",
" <td>0.97</td>\n",
" <td>26.055</td>\n",
" <td>27.864</td>\n",
" <td>32.94</td>\n",
" <td>51.89</td>\n",
" </tr>\n",
" <tr>\n",
" <th>224</th>\n",
" <td>2720028633</td>\n",
" <td>3.97</td>\n",
" <td>67.04</td>\n",
" <td>0.94</td>\n",
" <td>28.043</td>\n",
" <td>27.368</td>\n",
" <td>31.88</td>\n",
" <td>51.38</td>\n",
" </tr>\n",
" <tr>\n",
" <th>225</th>\n",
" <td>2720028634</td>\n",
" <td>4.12</td>\n",
" <td>68.42</td>\n",
" <td>0.96</td>\n",
" <td>26.493</td>\n",
" <td>27.886</td>\n",
" <td>33.16</td>\n",
" <td>52.00</td>\n",
" </tr>\n",
" <tr>\n",
" <th>226</th>\n",
" <td>2720017683</td>\n",
" <td>3.88</td>\n",
" <td>67.42</td>\n",
" <td>0.94</td>\n",
" <td>27.760</td>\n",
" <td>26.616</td>\n",
" <td>31.65</td>\n",
" <td>50.56</td>\n",
" </tr>\n",
" <tr>\n",
" <th>227</th>\n",
" <td>2720017678</td>\n",
" <td>3.81</td>\n",
" <td>66.74</td>\n",
" <td>0.92</td>\n",
" <td>28.530</td>\n",
" <td>26.688</td>\n",
" <td>31.02</td>\n",
" <td>50.82</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"<p>228 rows × 8 columns</p>\n",
"</div>"
],
"text/plain": [
" Unnamed: 0_level_0 氢 碳 氮 氧 弹筒发热量 挥发分 固定炭\n",
" 化验编号 Had Cad Nad Oad Qb,ad Vad Fcad\n",
" Unnamed: 0_level_2 (%) (%) (%) (%) MJ/kg (%) (%)\n",
"0 2720110529 3.93 70.18 0.81 25.079 27.820 32.06 55.68\n",
"1 2720096883 3.78 68.93 0.77 26.512 27.404 29.96 54.71\n",
"2 2720109084 3.48 69.60 0.76 26.148 27.578 29.31 55.99\n",
"3 2720084708 3.47 66.71 0.76 29.055 26.338 28.58 53.87\n",
"4 2720062721 3.87 68.78 0.80 26.542 27.280 29.97 54.78\n",
".. ... ... ... ... ... ... ... ...\n",
"223 2720030490 4.12 68.85 0.97 26.055 27.864 32.94 51.89\n",
"224 2720028633 3.97 67.04 0.94 28.043 27.368 31.88 51.38\n",
"225 2720028634 4.12 68.42 0.96 26.493 27.886 33.16 52.00\n",
"226 2720017683 3.88 67.42 0.94 27.760 26.616 31.65 50.56\n",
"227 2720017678 3.81 66.74 0.92 28.530 26.688 31.02 50.82\n",
"\n",
"[228 rows x 8 columns]"
]
},
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"data_0102 = pd.read_excel('./data/20240102/20240102.xlsx', header=[0,1,2])\n",
"data_0102"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "972f1e9c-3ebc-45cf-8d1f-7611645e5238",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"['化验编号',\n",
" '氢Had(%)',\n",
" '碳Cad(%)',\n",
" '氮Nad(%)',\n",
" '氧Oad(%)',\n",
" '弹筒发热量Qb,adMJ/kg',\n",
" '挥发分Vad(%)',\n",
" '固定炭Fcad(%)']"
]
},
"execution_count": 4,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"cols = [''.join([y for y in x if 'Unnamed' not in y]) for x in data_0102.columns]\n",
"cols"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "c95f1106-b3a4-43c6-88ec-3cdebf91d79a",
"metadata": {},
"outputs": [],
"source": [
"data_0102.columns = cols"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "2e96af0a-feda-4a1f-a13e-9c8861c6f4d4",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>化验编号</th>\n",
" <th>氢Had(%)</th>\n",
" <th>碳Cad(%)</th>\n",
" <th>氮Nad(%)</th>\n",
" <th>氧Oad(%)</th>\n",
" <th>弹筒发热量Qb,adMJ/kg</th>\n",
" <th>挥发分Vad(%)</th>\n",
" <th>固定炭Fcad(%)</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>2720110529</td>\n",
" <td>3.93</td>\n",
" <td>70.18</td>\n",
" <td>0.81</td>\n",
" <td>25.079</td>\n",
" <td>27.820</td>\n",
" <td>32.06</td>\n",
" <td>55.68</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>2720096883</td>\n",
" <td>3.78</td>\n",
" <td>68.93</td>\n",
" <td>0.77</td>\n",
" <td>26.512</td>\n",
" <td>27.404</td>\n",
" <td>29.96</td>\n",
" <td>54.71</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>2720109084</td>\n",
" <td>3.48</td>\n",
" <td>69.60</td>\n",
" <td>0.76</td>\n",
" <td>26.148</td>\n",
" <td>27.578</td>\n",
" <td>29.31</td>\n",
" <td>55.99</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>2720084708</td>\n",
" <td>3.47</td>\n",
" <td>66.71</td>\n",
" <td>0.76</td>\n",
" <td>29.055</td>\n",
" <td>26.338</td>\n",
" <td>28.58</td>\n",
" <td>53.87</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>2720062721</td>\n",
" <td>3.87</td>\n",
" <td>68.78</td>\n",
" <td>0.80</td>\n",
" <td>26.542</td>\n",
" <td>27.280</td>\n",
" <td>29.97</td>\n",
" <td>54.78</td>\n",
" </tr>\n",
" <tr>\n",
" <th>...</th>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" </tr>\n",
" <tr>\n",
" <th>223</th>\n",
" <td>2720030490</td>\n",
" <td>4.12</td>\n",
" <td>68.85</td>\n",
" <td>0.97</td>\n",
" <td>26.055</td>\n",
" <td>27.864</td>\n",
" <td>32.94</td>\n",
" <td>51.89</td>\n",
" </tr>\n",
" <tr>\n",
" <th>224</th>\n",
" <td>2720028633</td>\n",
" <td>3.97</td>\n",
" <td>67.04</td>\n",
" <td>0.94</td>\n",
" <td>28.043</td>\n",
" <td>27.368</td>\n",
" <td>31.88</td>\n",
" <td>51.38</td>\n",
" </tr>\n",
" <tr>\n",
" <th>225</th>\n",
" <td>2720028634</td>\n",
" <td>4.12</td>\n",
" <td>68.42</td>\n",
" <td>0.96</td>\n",
" <td>26.493</td>\n",
" <td>27.886</td>\n",
" <td>33.16</td>\n",
" <td>52.00</td>\n",
" </tr>\n",
" <tr>\n",
" <th>226</th>\n",
" <td>2720017683</td>\n",
" <td>3.88</td>\n",
" <td>67.42</td>\n",
" <td>0.94</td>\n",
" <td>27.760</td>\n",
" <td>26.616</td>\n",
" <td>31.65</td>\n",
" <td>50.56</td>\n",
" </tr>\n",
" <tr>\n",
" <th>227</th>\n",
" <td>2720017678</td>\n",
" <td>3.81</td>\n",
" <td>66.74</td>\n",
" <td>0.92</td>\n",
" <td>28.530</td>\n",
" <td>26.688</td>\n",
" <td>31.02</td>\n",
" <td>50.82</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"<p>228 rows × 8 columns</p>\n",
"</div>"
],
"text/plain": [
" 化验编号 氢Had(%) 碳Cad(%) 氮Nad(%) 氧Oad(%) 弹筒发热量Qb,adMJ/kg \\\n",
"0 2720110529 3.93 70.18 0.81 25.079 27.820 \n",
"1 2720096883 3.78 68.93 0.77 26.512 27.404 \n",
"2 2720109084 3.48 69.60 0.76 26.148 27.578 \n",
"3 2720084708 3.47 66.71 0.76 29.055 26.338 \n",
"4 2720062721 3.87 68.78 0.80 26.542 27.280 \n",
".. ... ... ... ... ... ... \n",
"223 2720030490 4.12 68.85 0.97 26.055 27.864 \n",
"224 2720028633 3.97 67.04 0.94 28.043 27.368 \n",
"225 2720028634 4.12 68.42 0.96 26.493 27.886 \n",
"226 2720017683 3.88 67.42 0.94 27.760 26.616 \n",
"227 2720017678 3.81 66.74 0.92 28.530 26.688 \n",
"\n",
" 挥发分Vad(%) 固定炭Fcad(%) \n",
"0 32.06 55.68 \n",
"1 29.96 54.71 \n",
"2 29.31 55.99 \n",
"3 28.58 53.87 \n",
"4 29.97 54.78 \n",
".. ... ... \n",
"223 32.94 51.89 \n",
"224 31.88 51.38 \n",
"225 33.16 52.00 \n",
"226 31.65 50.56 \n",
"227 31.02 50.82 \n",
"\n",
"[228 rows x 8 columns]"
]
},
"execution_count": 6,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"data_0102"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "04b177a7-2f02-4e23-8ea9-29f34cf3eafc",
"metadata": {},
"outputs": [],
"source": [
"out_cols = ['挥发分Vad(%)', '固定炭Fcad(%)']"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "31169fbf-d78e-42f7-87f3-71ba3dd0979d",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"['挥发分Vad(%)', '固定炭Fcad(%)']"
]
},
"execution_count": 8,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"out_cols"
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "feaedd50-f999-45bf-b465-3d359b0c0110",
"metadata": {},
"outputs": [],
"source": [
"data = data_0102.copy()"
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "a40bee0f-011a-4edb-80f8-4e2f40e755fd",
"metadata": {},
"outputs": [],
"source": [
"train_data = data.dropna(subset=out_cols).fillna(0)"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "535d37b6-b9de-4025-ac8f-62f5bdbe2451",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"2024-01-05 16:46:07.061819: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library libcudart.so.11.0\n"
]
}
],
"source": [
"import tensorflow as tf\n",
"from tensorflow import keras\n",
"from tensorflow.keras import layers\n",
"import tensorflow.keras.backend as K"
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "1c85d462-f248-4ffb-908f-eb4b20eab179",
"metadata": {},
"outputs": [],
"source": [
"class TransformerBlock(layers.Layer):\n",
" def __init__(self, embed_dim, num_heads, ff_dim, name, rate=0.1):\n",
" super().__init__()\n",
" self.att = layers.MultiHeadAttention(num_heads=num_heads, key_dim=embed_dim, name=name)\n",
" self.ffn = keras.Sequential(\n",
" [layers.Dense(ff_dim, activation=\"relu\"), layers.Dense(embed_dim),]\n",
" )\n",
" self.layernorm1 = layers.LayerNormalization(epsilon=1e-6)\n",
" self.layernorm2 = layers.LayerNormalization(epsilon=1e-6)\n",
" self.dropout1 = layers.Dropout(rate)\n",
" self.dropout2 = layers.Dropout(rate)\n",
"\n",
" def call(self, inputs, training):\n",
" attn_output = self.att(inputs, inputs)\n",
" attn_output = self.dropout1(attn_output, training=training)\n",
" out1 = self.layernorm1(inputs + attn_output)\n",
" ffn_output = self.ffn(out1)\n",
" ffn_output = self.dropout2(ffn_output, training=training)\n",
" return self.layernorm2(out1 + ffn_output)"
]
},
{
"cell_type": "code",
"execution_count": 13,
"id": "790284a3-b9d3-4144-b481-38a7c3ecb4b9",
"metadata": {},
"outputs": [],
"source": [
"from tensorflow.keras import Model"
]
},
{
"cell_type": "code",
"execution_count": 14,
"id": "cd9a1ca1-d0ca-4cb5-9ef5-fd5d63576cd2",
"metadata": {},
"outputs": [],
"source": [
"from tensorflow.keras.initializers import Constant"
]
},
{
"cell_type": "code",
"execution_count": 15,
"id": "9bc02f29-0fb7-420d-99a8-435eadc06e29",
"metadata": {},
"outputs": [],
"source": [
"# Custom loss layer\n",
"class CustomMultiLossLayer(layers.Layer):\n",
" def __init__(self, nb_outputs=2, **kwargs):\n",
" self.nb_outputs = nb_outputs\n",
" self.is_placeholder = True\n",
" super(CustomMultiLossLayer, self).__init__(**kwargs)\n",
" \n",
" def build(self, input_shape=None):\n",
" # initialise log_vars\n",
" self.log_vars = []\n",
" for i in range(self.nb_outputs):\n",
" self.log_vars += [self.add_weight(name='log_var' + str(i), shape=(1,),\n",
" initializer=tf.initializers.he_normal(), trainable=True)]\n",
" super(CustomMultiLossLayer, self).build(input_shape)\n",
"\n",
" def multi_loss(self, ys_true, ys_pred):\n",
" assert len(ys_true) == self.nb_outputs and len(ys_pred) == self.nb_outputs\n",
" loss = 0\n",
" for y_true, y_pred, log_var in zip(ys_true, ys_pred, self.log_vars):\n",
" mse = (y_true - y_pred) ** 2.\n",
" pre = K.exp(-log_var[0])\n",
" loss += tf.abs(tf.reduce_logsumexp(pre * mse + log_var[0], axis=-1))\n",
" return K.mean(loss)\n",
"\n",
" def call(self, inputs):\n",
" ys_true = inputs[:self.nb_outputs]\n",
" ys_pred = inputs[self.nb_outputs:]\n",
" loss = self.multi_loss(ys_true, ys_pred)\n",
" self.add_loss(loss, inputs=inputs)\n",
" # We won't actually use the output.\n",
" return K.concatenate(inputs, -1)"
]
},
{
"cell_type": "code",
"execution_count": 16,
"id": "a190207e-5a59-4813-9660-758760cf1b73",
"metadata": {},
"outputs": [],
"source": [
"num_heads, ff_dim = 3, 16"
]
},
{
"cell_type": "code",
"execution_count": 29,
"id": "80f32155-e71f-4615-8d0c-01dfd04988fe",
"metadata": {},
"outputs": [],
"source": [
"def get_prediction_model():\n",
" inputs = layers.Input(shape=(1,len(feature_cols)), name='input')\n",
" x = layers.Conv1D(filters=64, kernel_size=1, activation='relu')(inputs)\n",
" # x = layers.Dropout(rate=0.1)(x)\n",
" lstm_out = layers.Bidirectional(layers.LSTM(units=64, return_sequences=True))(x)\n",
" lstm_out = layers.Dense(128, activation='relu')(lstm_out)\n",
" transformer_block = TransformerBlock(128, num_heads, ff_dim, name='first_attn')\n",
" out = transformer_block(lstm_out)\n",
" out = layers.GlobalAveragePooling1D()(out)\n",
" out = layers.Dropout(0.1)(out)\n",
" out = layers.Dense(64, activation='relu')(out)\n",
" # out = K.expand_dims(out, axis=1)\n",
"\n",
" bet = layers.Dense(32, activation=\"relu\")(out)\n",
" mesco = layers.Dense(32, activation=\"relu\")(out)\n",
"\n",
" bet = layers.Dense(1, activation='sigmoid', name='vad')(bet)\n",
" mesco = layers.Dense(1, activation='sigmoid', name='fcad')(mesco)\n",
"\n",
" model = Model(inputs=[inputs], outputs=[bet, mesco])\n",
" return model\n"
]
},
{
"cell_type": "code",
"execution_count": 18,
"id": "264001b1-5e4a-4786-96fd-2b5c70ab3212",
"metadata": {},
"outputs": [],
"source": [
"def get_trainable_model(prediction_model):\n",
" inputs = layers.Input(shape=(1,len(feature_cols)), name='input')\n",
" bet, mesco = prediction_model(inputs)\n",
" bet_real = layers.Input(shape=(1,), name='vad_real')\n",
" mesco_real = layers.Input(shape=(1,), name='fcad_real')\n",
" out = CustomMultiLossLayer(nb_outputs=2)([bet_real, mesco_real, bet, mesco])\n",
" return Model([inputs, bet_real, mesco_real], out)"
]
},
{
"cell_type": "code",
"execution_count": 19,
"id": "372011ea-9876-41eb-a4e6-83ccd6c71559",
"metadata": {},
"outputs": [],
"source": [
"from tensorflow.python.keras.utils.vis_utils import plot_model"
]
},
{
"cell_type": "code",
"execution_count": 20,
"id": "1eebdab3-1f88-48a1-b5e0-bc8787528c1b",
"metadata": {},
"outputs": [],
"source": [
"maxs = train_data.max()\n",
"mins = train_data.min()\n",
"for col in train_data.columns:\n",
" if maxs[col] - mins[col] == 0:\n",
" continue\n",
" train_data[col] = (train_data[col] - mins[col]) / (maxs[col] - mins[col])"
]
},
{
"cell_type": "code",
"execution_count": 21,
"id": "7f27bd56-4f6b-4242-9f79-c7d6b3ee2f13",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>化验编号</th>\n",
" <th>氢Had(%)</th>\n",
" <th>碳Cad(%)</th>\n",
" <th>氮Nad(%)</th>\n",
" <th>氧Oad(%)</th>\n",
" <th>弹筒发热量Qb,adMJ/kg</th>\n",
" <th>挥发分Vad(%)</th>\n",
" <th>固定炭Fcad(%)</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>0.996547</td>\n",
" <td>0.773973</td>\n",
" <td>0.835414</td>\n",
" <td>0.456522</td>\n",
" <td>0.171463</td>\n",
" <td>0.811249</td>\n",
" <td>0.847737</td>\n",
" <td>0.828147</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>0.851118</td>\n",
" <td>0.671233</td>\n",
" <td>0.799943</td>\n",
" <td>0.369565</td>\n",
" <td>0.210254</td>\n",
" <td>0.782038</td>\n",
" <td>0.674897</td>\n",
" <td>0.794606</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>0.981147</td>\n",
" <td>0.465753</td>\n",
" <td>0.818956</td>\n",
" <td>0.347826</td>\n",
" <td>0.200401</td>\n",
" <td>0.794256</td>\n",
" <td>0.621399</td>\n",
" <td>0.838866</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>0.721367</td>\n",
" <td>0.458904</td>\n",
" <td>0.736947</td>\n",
" <td>0.347826</td>\n",
" <td>0.279094</td>\n",
" <td>0.707183</td>\n",
" <td>0.561317</td>\n",
" <td>0.765560</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>0.487046</td>\n",
" <td>0.732877</td>\n",
" <td>0.795687</td>\n",
" <td>0.434783</td>\n",
" <td>0.211066</td>\n",
" <td>0.773331</td>\n",
" <td>0.675720</td>\n",
" <td>0.797026</td>\n",
" </tr>\n",
" <tr>\n",
" <th>...</th>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" </tr>\n",
" <tr>\n",
" <th>223</th>\n",
" <td>0.143553</td>\n",
" <td>0.904110</td>\n",
" <td>0.797673</td>\n",
" <td>0.804348</td>\n",
" <td>0.197883</td>\n",
" <td>0.814339</td>\n",
" <td>0.920165</td>\n",
" <td>0.697095</td>\n",
" </tr>\n",
" <tr>\n",
" <th>224</th>\n",
" <td>0.123762</td>\n",
" <td>0.801370</td>\n",
" <td>0.746311</td>\n",
" <td>0.739130</td>\n",
" <td>0.251699</td>\n",
" <td>0.779510</td>\n",
" <td>0.832922</td>\n",
" <td>0.679461</td>\n",
" </tr>\n",
" <tr>\n",
" <th>225</th>\n",
" <td>0.123773</td>\n",
" <td>0.904110</td>\n",
" <td>0.785471</td>\n",
" <td>0.782609</td>\n",
" <td>0.209740</td>\n",
" <td>0.815884</td>\n",
" <td>0.938272</td>\n",
" <td>0.700899</td>\n",
" </tr>\n",
" <tr>\n",
" <th>226</th>\n",
" <td>0.007066</td>\n",
" <td>0.739726</td>\n",
" <td>0.757094</td>\n",
" <td>0.739130</td>\n",
" <td>0.244038</td>\n",
" <td>0.726705</td>\n",
" <td>0.813992</td>\n",
" <td>0.651107</td>\n",
" </tr>\n",
" <tr>\n",
" <th>227</th>\n",
" <td>0.007012</td>\n",
" <td>0.691781</td>\n",
" <td>0.737798</td>\n",
" <td>0.695652</td>\n",
" <td>0.264882</td>\n",
" <td>0.731760</td>\n",
" <td>0.762140</td>\n",
" <td>0.660097</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"<p>228 rows × 8 columns</p>\n",
"</div>"
],
"text/plain": [
" 化验编号 氢Had(%) 碳Cad(%) 氮Nad(%) 氧Oad(%) 弹筒发热量Qb,adMJ/kg \\\n",
"0 0.996547 0.773973 0.835414 0.456522 0.171463 0.811249 \n",
"1 0.851118 0.671233 0.799943 0.369565 0.210254 0.782038 \n",
"2 0.981147 0.465753 0.818956 0.347826 0.200401 0.794256 \n",
"3 0.721367 0.458904 0.736947 0.347826 0.279094 0.707183 \n",
"4 0.487046 0.732877 0.795687 0.434783 0.211066 0.773331 \n",
".. ... ... ... ... ... ... \n",
"223 0.143553 0.904110 0.797673 0.804348 0.197883 0.814339 \n",
"224 0.123762 0.801370 0.746311 0.739130 0.251699 0.779510 \n",
"225 0.123773 0.904110 0.785471 0.782609 0.209740 0.815884 \n",
"226 0.007066 0.739726 0.757094 0.739130 0.244038 0.726705 \n",
"227 0.007012 0.691781 0.737798 0.695652 0.264882 0.731760 \n",
"\n",
" 挥发分Vad(%) 固定炭Fcad(%) \n",
"0 0.847737 0.828147 \n",
"1 0.674897 0.794606 \n",
"2 0.621399 0.838866 \n",
"3 0.561317 0.765560 \n",
"4 0.675720 0.797026 \n",
".. ... ... \n",
"223 0.920165 0.697095 \n",
"224 0.832922 0.679461 \n",
"225 0.938272 0.700899 \n",
"226 0.813992 0.651107 \n",
"227 0.762140 0.660097 \n",
"\n",
"[228 rows x 8 columns]"
]
},
"execution_count": 21,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"train_data"
]
},
{
"cell_type": "code",
"execution_count": 22,
"id": "baf45a3d-dc01-44fc-9f0b-456964ac2cdb",
"metadata": {},
"outputs": [],
"source": [
"# feature_cols = [x for x in train_data.columns if x not in out_cols and '第二次' not in x]\n",
"feature_cols = [x for x in train_data.columns if x not in out_cols]\n",
"use_cols = feature_cols + out_cols"
]
},
{
"cell_type": "code",
"execution_count": 23,
"id": "f2d27538-d2bc-4202-b0cf-d3e0949b4686",
"metadata": {},
"outputs": [],
"source": [
"use_data = train_data.copy()\n",
"for col in use_cols:\n",
" use_data[col] = use_data[col].astype('float32')"
]
},
{
"cell_type": "code",
"execution_count": 24,
"id": "50daf170-efec-49e5-8f8e-9a45938cacfc",
"metadata": {},
"outputs": [],
"source": [
"from sklearn.model_selection import KFold, train_test_split\n",
"kf = KFold(n_splits=6, shuffle=True, random_state=42)"
]
},
{
"cell_type": "code",
"execution_count": 25,
"id": "0f863423-be12-478b-a08d-e3c6f5dfb8ee",
"metadata": {},
"outputs": [],
"source": [
"from tensorflow.keras import optimizers\n",
"from tensorflow.python.keras.utils.vis_utils import plot_model\n",
"from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score, mean_absolute_percentage_error"
]
},
{
"cell_type": "code",
"execution_count": 26,
"id": "2c89b32a-017c-4d05-ab78-8b9b8eb0dcbb",
"metadata": {},
"outputs": [],
"source": [
"from keras.callbacks import ReduceLROnPlateau\n",
"reduce_lr = ReduceLROnPlateau(monitor='val_loss', patience=10, mode='auto')"
]
},
{
"cell_type": "code",
"execution_count": 27,
"id": "ca6ce434-80b6-4609-9596-9a5120680462",
"metadata": {},
"outputs": [],
"source": [
"def print_eva(y_true, y_pred, tp):\n",
" MSE = mean_squared_error(y_true, y_pred)\n",
" RMSE = np.sqrt(MSE)\n",
" MAE = mean_absolute_error(y_true, y_pred)\n",
" MAPE = mean_absolute_percentage_error(y_true, y_pred)\n",
" R_2 = r2_score(y_true, y_pred)\n",
" print(f\"COL: {tp}, MSE: {format(MSE, '.2E')}\", end=',')\n",
" print(f'RMSE: {round(RMSE, 3)}', end=',')\n",
" print(f'MAPE: {round(MAPE * 100, 3)} %', end=',')\n",
" print(f'MAE: {round(MAE, 3)}', end=',')\n",
" print(f'R_2: {round(R_2, 3)}')\n",
" return [MSE, RMSE, MAE, MAPE, R_2]"
]
},
{
"cell_type": "code",
"execution_count": 30,
"id": "10213bc5-bf13-46ed-9ce9-b1dbc5af72ee",
"metadata": {},
"outputs": [],
"source": [
"prediction_model = get_prediction_model()\n",
"trainable_model = get_trainable_model(prediction_model)\n",
"trainable_model.compile(optimizer='adam', loss=None)"
]
},
{
"cell_type": "code",
"execution_count": 31,
"id": "4a1be90d-b8f1-4fe1-9952-1cdcc489fab5",
"metadata": {},
"outputs": [
{
"data": {
"image/png": "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",
"text/plain": [
"<IPython.core.display.Image object>"
]
},
"execution_count": 31,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"plot_model(prediction_model)"
]
},
{
"cell_type": "code",
"execution_count": 32,
"id": "6308b1dc-8e2e-4bf9-9b28-3b81979bf7e0",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"2024-01-05 16:53:33.639598: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:176] None of the MLIR Optimization Passes are enabled (registered 2)\n",
"2024-01-05 16:53:33.658810: I tensorflow/core/platform/profile_utils/cpu_utils.cc:114] CPU Frequency: 2200000000 Hz\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"COL: 挥发分Vad, MSE: 1.73E+00,RMSE: 1.316,MAPE: 3.178 %,MAE: 0.961,R_2: 0.626\n",
"COL: 固定炭Fcad, MSE: 2.28E+02,RMSE: 15.111,MAPE: 28.087 %,MAE: 15.088,R_2: -8.016\n",
"COL: 挥发分Vad, MSE: 8.81E-01,RMSE: 0.939,MAPE: 2.56 %,MAE: 0.745,R_2: 0.893\n",
"COL: 固定炭Fcad, MSE: 1.18E+00,RMSE: 1.085,MAPE: 1.679 %,MAE: 0.836,R_2: 0.977\n",
"COL: 挥发分Vad, MSE: 4.92E-01,RMSE: 0.701,MAPE: 1.855 %,MAE: 0.548,R_2: 0.874\n",
"COL: 固定炭Fcad, MSE: 9.80E-01,RMSE: 0.99,MAPE: 1.413 %,MAE: 0.774,R_2: 0.93\n",
"COL: 挥发分Vad, MSE: 4.56E+01,RMSE: 6.756,MAPE: 22.559 %,MAE: 6.7,R_2: -6.995\n",
"COL: 固定炭Fcad, MSE: 1.12E+02,RMSE: 10.605,MAPE: 20.153 %,MAE: 10.579,R_2: -2.311\n",
"WARNING:tensorflow:5 out of the last 9 calls to <function Model.make_predict_function.<locals>.predict_function at 0x7f2c5ec9aa60> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
"COL: 挥发分Vad, MSE: 8.31E-01,RMSE: 0.912,MAPE: 2.389 %,MAE: 0.728,R_2: 0.811\n",
"COL: 固定炭Fcad, MSE: 1.75E+02,RMSE: 13.221,MAPE: 24.517 %,MAE: 13.188,R_2: -7.209\n",
"WARNING:tensorflow:6 out of the last 11 calls to <function Model.make_predict_function.<locals>.predict_function at 0x7f2c5e414940> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
"COL: 挥发分Vad, MSE: 5.16E-01,RMSE: 0.719,MAPE: 2.022 %,MAE: 0.591,R_2: 0.894\n",
"COL: 固定炭Fcad, MSE: 1.33E+00,RMSE: 1.154,MAPE: 1.496 %,MAE: 0.739,R_2: 0.944\n"
]
}
],
"source": [
"vad_eva_list = list()\n",
"fcad_eva_list = list()\n",
"train_data = use_data[use_cols].copy()\n",
"for (train_index, test_index) in kf.split(train_data):\n",
" train = train_data.loc[train_index]\n",
" valid = train_data.loc[test_index]\n",
" X = np.expand_dims(train[feature_cols].values, axis=1)\n",
" Y = [x for x in train[out_cols].values.T]\n",
" X_valid = np.expand_dims(valid[feature_cols].values, axis=1)\n",
" Y_valid = [x for x in valid[out_cols].values.T]\n",
" prediction_model = get_prediction_model()\n",
" trainable_model = get_trainable_model(prediction_model)\n",
" trainable_model.compile(optimizer='adam', loss=None)\n",
" hist = trainable_model.fit([X, Y[0], Y[1]], epochs=120, batch_size=8, verbose=0, \n",
" validation_data=[X_valid, Y_valid[0], Y_valid[1]],\n",
" callbacks=[reduce_lr]\n",
" )\n",
" rst = prediction_model.predict(X_valid)\n",
" pred_rst = pd.DataFrame.from_records(np.squeeze(np.asarray(rst), axis=2).T, columns=out_cols)\n",
" real_rst = valid[out_cols].copy()\n",
" for col in out_cols:\n",
" pred_rst[col] = pred_rst[col] * (maxs[col] - mins[col]) + mins[col]\n",
" real_rst[col] = real_rst[col] * (maxs[col] - mins[col]) + mins[col]\n",
" y_pred_vad = pred_rst['挥发分Vad(%)'].values.reshape(-1,)\n",
" y_pred_fcad = pred_rst['固定炭Fcad(%)'].values.reshape(-1,)\n",
" y_true_vad = real_rst['挥发分Vad(%)'].values.reshape(-1,)\n",
" y_true_fcad = real_rst['固定炭Fcad(%)'].values.reshape(-1,)\n",
" vad_eva = print_eva(y_true_vad, y_pred_vad, tp='挥发分Vad')\n",
" fcad_eva = print_eva(y_true_fcad, y_pred_fcad, tp='固定炭Fcad')\n",
" vad_eva_list.append(vad_eva)\n",
" fcad_eva_list.append(fcad_eva)\n",
" del trainable_model\n",
" del prediction_model"
]
},
{
"cell_type": "code",
"execution_count": 33,
"id": "27e0abf7-aa29-467f-bc5e-b66a1adf6165",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"MSE 0.890323\n",
"RMSE 0.917209\n",
"MAE 0.714643\n",
"MAPE 0.024009\n",
"R_2 0.819531\n",
"dtype: float64"
]
},
"execution_count": 33,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"vad_df = pd.DataFrame.from_records(vad_eva_list, columns=['MSE', 'RMSE', 'MAE', 'MAPE', 'R_2'])\n",
"vad_df.sort_values(by='R_2')[1:].mean()"
]
},
{
"cell_type": "code",
"execution_count": 34,
"id": "070cdb94-6e7b-4028-b6d5-ba8570c902ba",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"MSE 58.149181\n",
"RMSE 5.410969\n",
"MAE 5.223220\n",
"MAPE 0.098516\n",
"R_2 -1.333914\n",
"dtype: float64"
]
},
"execution_count": 34,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"fcad_df = pd.DataFrame.from_records(fcad_eva_list, columns=['MSE', 'RMSE', 'MAE', 'MAPE', 'R_2'])\n",
"fcad_df.sort_values(by='R_2')[1:].mean()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "54c1df2c-c297-4b8d-be8a-3a99cff22545",
"metadata": {},
"outputs": [],
"source": [
"train, valid = train_test_split(use_data[use_cols], test_size=0.3, random_state=42, shuffle=True)\n",
"valid, test = train_test_split(valid, test_size=0.3, random_state=42, shuffle=True)"
]
},
{
"cell_type": "code",
"execution_count": 31,
"id": "e7a914da-b9c2-40d9-96e0-459b0888adba",
"metadata": {},
"outputs": [],
"source": [
"prediction_model = get_prediction_model()\n",
"trainable_model = get_trainable_model(prediction_model)"
]
},
{
"cell_type": "code",
"execution_count": 34,
"id": "2494ef5a-5b2b-4f11-b6cd-dc39503c9106",
"metadata": {},
"outputs": [],
"source": [
"X = np.expand_dims(train[feature_cols].values, axis=1)\n",
"Y = [x for x in train[out_cols].values.T]\n",
"Y_valid = [x for x in valid[out_cols].values.T]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "cf869e4d-0fce-45a2-afff-46fd9b30fd1c",
"metadata": {},
"outputs": [],
"source": [
"trainable_model.compile(optimizer='adam', loss=None)\n",
"hist = trainable_model.fit([X, Y[0], Y[1]], epochs=120, batch_size=8, verbose=1, \n",
" validation_data=[np.expand_dims(valid[feature_cols].values, axis=1), Y_valid[0], Y_valid[1]],\n",
" callbacks=[reduce_lr]\n",
" )"
]
},
{
"cell_type": "code",
"execution_count": 41,
"id": "67bfbe88-5f2c-4659-b2dc-eb9f1b824d04",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[array([[0.73740077],\n",
" [0.89292204],\n",
" [0.7599046 ],\n",
" [0.67802393],\n",
" [0.6815233 ],\n",
" [0.88627005],\n",
" [0.6121343 ],\n",
" [0.7072234 ],\n",
" [0.8561135 ],\n",
" [0.52762157],\n",
" [0.8325021 ],\n",
" [0.50241977],\n",
" [0.8242289 ],\n",
" [0.68957335],\n",
" [0.6980361 ],\n",
" [0.82116604],\n",
" [0.8566438 ],\n",
" [0.53687835],\n",
" [0.56832707],\n",
" [0.78476715],\n",
" [0.85638577]], dtype=float32),\n",
" array([[0.68600863],\n",
" [0.78454906],\n",
" [0.8179163 ],\n",
" [0.94351083],\n",
" [0.86383885],\n",
" [0.69705516],\n",
" [0.6913491 ],\n",
" [0.80277354],\n",
" [0.93557894],\n",
" [0.82278305],\n",
" [0.82674253],\n",
" [0.93518937],\n",
" [0.8094449 ],\n",
" [0.9206344 ],\n",
" [0.7747319 ],\n",
" [0.9137207 ],\n",
" [0.9491073 ],\n",
" [0.93225 ],\n",
" [0.6185102 ],\n",
" [0.8867341 ],\n",
" [0.82890105]], dtype=float32)]"
]
},
"execution_count": 41,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"rst = prediction_model.predict(np.expand_dims(test[feature_cols], axis=1))\n",
"rst"
]
},
{
"cell_type": "code",
"execution_count": 42,
"id": "7de501e9-05a2-424c-a5f4-85d43ad37592",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[0.9991559102070927, 0.9998196796918477]"
]
},
"execution_count": 42,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"[np.exp(K.get_value(log_var[0]))**0.5 for log_var in trainable_model.layers[-1].log_vars]"
]
},
{
"cell_type": "code",
"execution_count": 46,
"id": "5c69d03b-34fd-4dbf-aec6-c15093bb22ab",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"Index(['挥发分Vad(%)', '固定炭Fcad(%)'], dtype='object')"
]
},
"execution_count": 46,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"real_rst.columns"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "294813b8-90be-4007-9fd6-c26ee7bb9652",
"metadata": {},
"outputs": [],
"source": [
"for col in out_cols:\n",
" pred_rst[col] = pred_rst[col] * (maxs[col] - mins[col]) + mins[col]\n",
" real_rst[col] = real_rst[col] * (maxs[col] - mins[col]) + mins[col]"
]
},
{
"cell_type": "code",
"execution_count": 47,
"id": "21739f82-d82a-4bde-8537-9504b68a96d5",
"metadata": {},
"outputs": [],
"source": [
"y_pred_vad = pred_rst['挥发分Vad(%)'].values.reshape(-1,)\n",
"y_pred_fcad = pred_rst['固定炭Fcad(%)'].values.reshape(-1,)\n",
"y_true_vad = real_rst['挥发分Vad(%)'].values.reshape(-1,)\n",
"y_true_fcad = real_rst['固定炭Fcad(%)'].values.reshape(-1,)"
]
},
{
"cell_type": "code",
"execution_count": 56,
"id": "4ec4caa9-7c46-4fc8-a94b-cb659e924304",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"COL: 挥发分Vad, MSE: 3.35E-01,RMSE: 0.579,MAPE: 1.639 %,MAE: 0.504,R_2: 0.87\n",
"COL: 固定炭Fcad, MSE: 1.11E+00,RMSE: 1.055,MAPE: 1.497 %,MAE: 0.814,R_2: 0.876\n"
]
}
],
"source": [
"pm25_eva = print_eva(y_true_vad, y_pred_vad, tp='挥发分Vad')\n",
"pm10_eva = print_eva(y_true_fcad, y_pred_fcad, tp='固定炭Fcad')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "ac4a4339-ec7d-4266-8197-5276c2395288",
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"id": "f15cbb91-1ce7-4fb0-979a-a4bdc452a1ec",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.8.16"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
Reference in New Issue View Git Blame Copy Permalink