体积模量预测

$$K=-V\frac{\partial p}{\partial V}$$

体积模量 （K）也称为不可压缩量，是材料对于表面四周压强产生形变程度的度量。

定义：产生单位相对体积收缩所需的压强，在SI单位制中的基本单位是帕斯卡。

数据含义

• G_Reuss：多晶材料的剪切模量下界
• G_VRH：G_Reuss与G_Voigt的平均值
• G_Voigt：多晶材料的剪切模量上界
• K_Reuss：多晶材料的体积模量下界
• K_VRH：K_Reuss与K_Voigt的平均值
• K_Voigt：多晶材料的体积模量上界
• cif：可选：结构的描述字符串
• compliance_tensor：描述弹性行为的张量
• elastic_anisotropy：材料弹性方向依赖性的度量，度量总是>= 0
• elastic_tensor：描述对应于IEEE 方向的弹性行为的张量，对称于晶体结构
• elastic_tensor_original：描述弹性行为的张量，非对称的，对应于POSCAR 常规标准单元方向
• formula：材料的化学组成
• kpoint_density：可选：计算中的采样参数
• material_id：材料的Materials Project ID
• nsites：计算单胞的原子数
• poisson_ratio 描述对负载的横向响应
• poscar：可选：POSCAR数据
• space_group：材料晶体结构的空间群
• structure：pandas 系列定义了材料的结构
• volume：以立方埃为单位的晶胞体积，对于超晶胞计算，这个量是指整个超晶胞的体积。

导入和处理数据

#导入数据
from matminer.datasets import load_dataset
data=load_dataset('elastic_tensor_2015',data_home='.')

data.head()

	material_id	formula	nsites	space_group	volume	structure	elastic_anisotropy	G_Reuss	G_VRH	G_Voigt	K_Reuss	K_VRH	K_Voigt	poisson_ratio	compliance_tensor	elastic_tensor	elastic_tensor_original	cif	kpoint_density	poscar
0	mp-10003	Nb4CoSi	12	124	194.419802	[[0.94814328 2.07280467 2.5112 ] Nb, [5.273...	0.030688	96.844535	97.141604	97.438674	194.267623	194.268884	194.270146	0.285701	[[0.004385293093993, -0.0016070693558990002, -...	[[311.33514638650246, 144.45092552856926, 126....	[[311.33514638650246, 144.45092552856926, 126....	#\#CIF1.1\n###################################...	7000	Nb8 Co2 Si2\n1.0\n6.221780 0.000000 0.000000\n...
1	mp-10010	Al(CoSi)2	5	164	61.987320	[[0. 0. 0.] Al, [1.96639263 1.13529553 0.75278...	0.266910	93.939650	96.252006	98.564362	173.647763	175.449907	177.252050	0.268105	[[0.0037715428949660003, -0.000844229828709, -...	[[306.93357350984974, 88.02634955100905, 105.6...	[[306.93357350984974, 88.02634955100905, 105.6...	#\#CIF1.1\n###################################...	7000	Al1 Co2 Si2\n1.0\n3.932782 0.000000 0.000000\n...
2	mp-10015	SiOs	2	221	25.952539	[[1.480346 1.480346 1.480346] Si, [0. 0. 0.] Os]	0.756489	120.962289	130.112955	139.263621	295.077545	295.077545	295.077545	0.307780	[[0.0019959391925840004, -0.000433146670736000...	[[569.5291276937579, 157.8517489654999, 157.85...	[[569.5291276937579, 157.8517489654999, 157.85...	#\#CIF1.1\n###################################...	7000	Si1 Os1\n1.0\n2.960692 0.000000 0.000000\n0.00...
3	mp-10021	Ga	4	63	76.721433	[[0. 1.09045794 0.84078375] Ga, [0. ...	2.376805	12.205989	15.101901	17.997812	49.025963	49.130670	49.235377	0.360593	[[0.021647143908635, -0.005207263618160001, -0...	[[69.28798774976904, 34.7875015216915, 37.3877...	[[70.13259066665267, 40.60474945058445, 37.387...	#\#CIF1.1\n###################################...	7000	Ga4\n1.0\n2.803229 0.000000 0.000000\n0.000000...
4	mp-10025	SiRu2	12	62	160.300999	[[1.0094265 4.24771709 2.9955487 ] Si, [3.028...	0.196930	100.110773	101.947798	103.784823	255.055257	256.768081	258.480904	0.324682	[[0.00410214297725, -0.001272204332729, -0.001...	[[349.3767766177825, 186.67131003104407, 176.4...	[[407.4791016459293, 176.4759188081947, 213.83...	#\#CIF1.1\n###################################...	7000	Si4 Ru8\n1.0\n4.037706 0.000000 0.000000\n0.00...

要尝试预测 K _ VRH 和 G _ VRH (分别是体积模量和剪切模量的 Voight-Reuss-Hill 平均值)和elastic_anisotropy。

删除无关列：volume，nsites，compliance_tensor，elastic_tensor，elastic_tensor_original，G_Reuss，G_Voigt，K_Reuss，K_Voigt，cif，kpoint_density ，poscar

deleted_col=['volume','nsites','compliance_tensor','elastic_tensor','elastic_tensor_original','G_Reuss','G_Voigt','K_Reuss','K_Voigt','cif','kpoint_density','poscar']
data=data.drop(deleted_col,axis=1)

data.head()

	material_id	formula	space_group	structure	elastic_anisotropy	G_VRH	K_VRH	poisson_ratio
0	mp-10003	Nb4CoSi	124	[[0.94814328 2.07280467 2.5112 ] Nb, [5.273...	0.030688	97.141604	194.268884	0.285701
1	mp-10010	Al(CoSi)2	164	[[0. 0. 0.] Al, [1.96639263 1.13529553 0.75278...	0.266910	96.252006	175.449907	0.268105
2	mp-10015	SiOs	221	[[1.480346 1.480346 1.480346] Si, [0. 0. 0.] Os]	0.756489	130.112955	295.077545	0.307780
3	mp-10021	Ga	63	[[0. 1.09045794 0.84078375] Ga, [0. ...	2.376805	15.101901	49.130670	0.360593
4	mp-10025	SiRu2	62	[[1.0094265 4.24771709 2.9955487 ] Si, [3.028...	0.196930	101.947798	256.768081	0.324682

data.describe()

	space_group	elastic_anisotropy	G_VRH	K_VRH	poisson_ratio
count	1181.000000	1181.000000	1181.000000	1181.000000	1181.000000
mean	163.403895	2.145013	67.543145	136.259661	0.287401
std	65.040733	19.140097	44.579408	72.886978	0.062177
min	4.000000	0.000005	2.722175	6.476135	0.042582
25%	124.000000	0.145030	34.117959	76.435350	0.249159
50%	193.000000	0.355287	59.735163	130.382766	0.290198
75%	221.000000	0.923117	91.332142	189.574194	0.328808
max	229.000000	397.297866	522.921225	435.661487	0.467523

数据添加描述符

我们正在寻找输入(材料的组成和晶体结构)和输出(弹性性质，如 K _ VRH，G _ VRH 和elastic_anisotropy)之间的关系。为了找到这样的关系，我们需要“特征化”输入数据，使它们成为有意义地表示底层物理量的数字。例如，材料组成的一个“特征”或“描述符”，如 Nb4CoSi，将是化合物中元素的鲍林标准差电负性(按化学计量加权)。这个量值较高的组分离子较多，值较低的组分趋向于共价或离子。晶体结构的一个描述可能是位点的平均配位数; 更高的配位数表示更多的键，因此可能表示更硬的材料。使用 matminer，我们可以开始使用可用的描述符库生成数百个可能的描述符。数据挖掘技术可以使用可用的输出数据作为指导，帮助缩小与目标问题最相关的描述符的范围。

添加基于组合的特性

Matminer 中的一个主要类别的特征使用化学成份来特征化输入数据。让我们向 DataFrame 添加一些基于组合的特性

第一步是让一个列表示化学成份作为 pymatgen 组合对象。做到这一点的一种方法是使用 matminer 中的转换 conversions 将 String 组合(我们之前的formula列)转换为 pymatgen 组合。

from matminer.featurizers.conversions import StrToComposition
data=StrToComposition().featurize_dataframe(df=data,col_id='formula')
data.head()

StrToComposition:   0%|          | 0/1181 [00:00<?, ?it/s]

	material_id	formula	space_group	structure	elastic_anisotropy	G_VRH	K_VRH	poisson_ratio	composition
0	mp-10003	Nb4CoSi	124	[[0.94814328 2.07280467 2.5112 ] Nb, [5.273...	0.030688	97.141604	194.268884	0.285701	(Nb, Co, Si)
1	mp-10010	Al(CoSi)2	164	[[0. 0. 0.] Al, [1.96639263 1.13529553 0.75278...	0.266910	96.252006	175.449907	0.268105	(Al, Co, Si)
2	mp-10015	SiOs	221	[[1.480346 1.480346 1.480346] Si, [0. 0. 0.] Os]	0.756489	130.112955	295.077545	0.307780	(Si, Os)
3	mp-10021	Ga	63	[[0. 1.09045794 0.84078375] Ga, [0. ...	2.376805	15.101901	49.130670	0.360593	(Ga)
4	mp-10025	SiRu2	62	[[1.0094265 4.24771709 2.9955487 ] Si, [3.028...	0.196930	101.947798	256.768081	0.324682	(Si, Ru)

StrToComposition.featurize_dataframe

执行数据转换并动态设置目标列。

Args:

• df (Pandas.DataFrame): 包含输入数据的数据框架。.
• col_id (str or list of str): 包含要特征化的对象的列标签。如果特征化函数需要多个输入，则可以是多个标签

使用 matminer 中的一个特性向 DataFrame 添加一组描述符。

from matminer.featurizers.composition import ElementProperty
ep_feat=ElementProperty.from_preset(preset_name='magpie')#从预设字符串返回 ElementProperty
data=ep_feat.featurize_dataframe(data,'composition')#将composition转为特征
data.head()

ElementProperty:   0%|          | 0/1181 [00:00<?, ?it/s]

	material_id	formula	space_group	structure	elastic_anisotropy	G_VRH	K_VRH	poisson_ratio	composition	MagpieData minimum Number	...	MagpieData range GSmagmom	MagpieData mean GSmagmom	MagpieData avg_dev GSmagmom	MagpieData mode GSmagmom	MagpieData minimum SpaceGroupNumber	MagpieData maximum SpaceGroupNumber	MagpieData range SpaceGroupNumber	MagpieData mean SpaceGroupNumber	MagpieData avg_dev SpaceGroupNumber	MagpieData mode SpaceGroupNumber
0	mp-10003	Nb4CoSi	124	[[0.94814328 2.07280467 2.5112 ] Nb, [5.273...	0.030688	97.141604	194.268884	0.285701	(Nb, Co, Si)	14.0	...	1.548471	0.258079	0.430131	0.0	194.0	229.0	35.0	222.833333	9.611111	229.0
1	mp-10010	Al(CoSi)2	164	[[0. 0. 0.] Al, [1.96639263 1.13529553 0.75278...	0.266910	96.252006	175.449907	0.268105	(Al, Co, Si)	13.0	...	1.548471	0.619388	0.743266	0.0	194.0	227.0	33.0	213.400000	15.520000	194.0
2	mp-10015	SiOs	221	[[1.480346 1.480346 1.480346] Si, [0. 0. 0.] Os]	0.756489	130.112955	295.077545	0.307780	(Si, Os)	14.0	...	0.000000	0.000000	0.000000	0.0	194.0	227.0	33.0	210.500000	16.500000	194.0
3	mp-10021	Ga	63	[[0. 1.09045794 0.84078375] Ga, [0. ...	2.376805	15.101901	49.130670	0.360593	(Ga)	31.0	...	0.000000	0.000000	0.000000	0.0	64.0	64.0	0.0	64.000000	0.000000	64.0
4	mp-10025	SiRu2	62	[[1.0094265 4.24771709 2.9955487 ] Si, [3.028...	0.196930	101.947798	256.768081	0.324682	(Si, Ru)	14.0	...	0.000000	0.000000	0.000000	0.0	194.0	227.0	33.0	205.000000	14.666667	194.0

5 rows × 141 columns

另外，请注意，每个特性化工具还有一个citations()函数，它告诉您在哪里可以找到有关特性化工具的更多信息。

ep_feat.citations()

['@article{ward_agrawal_choudary_wolverton_2016, title={A general-purpose machine learning framework for predicting properties of inorganic materials}, volume={2}, DOI={10.1038/npjcompumats.2017.28}, number={1}, journal={npj Computational Materials}, author={Ward, Logan and Agrawal, Ankit and Choudhary, Alok and Wolverton, Christopher}, year={2016}}']

添加更多基于组合的特性

除了 ElementProperty 之外，还有很多基于组合的特性可以在 matmin.featurizer.composition找到。让我们试试ElectronegativityDiff 功能，这需要知道各种元素的氧化状态的组成。目前还没有这方面的信息，但是我们可以使用conversions包来尝试猜测氧化状态，然后将ElectronegativityDiff功能应用到本列。

from matminer.featurizers.conversions import CompositionToOxidComposition
from matminer.featurizers.composition import OxidationStates

data=CompositionToOxidComposition().featurize_dataframe(data,'composition')
data.head()

CompositionToOxidComposition:   0%|          | 0/1181 [00:00<?, ?it/s]

	material_id	formula	space_group	structure	elastic_anisotropy	G_VRH	K_VRH	poisson_ratio	composition	MagpieData minimum Number	...	MagpieData mean GSmagmom	MagpieData avg_dev GSmagmom	MagpieData mode GSmagmom	MagpieData minimum SpaceGroupNumber	MagpieData maximum SpaceGroupNumber	MagpieData range SpaceGroupNumber	MagpieData mean SpaceGroupNumber	MagpieData avg_dev SpaceGroupNumber	MagpieData mode SpaceGroupNumber	composition_oxid
0	mp-10003	Nb4CoSi	124	[[0.94814328 2.07280467 2.5112 ] Nb, [5.273...	0.030688	97.141604	194.268884	0.285701	(Nb, Co, Si)	14.0	...	0.258079	0.430131	0.0	194.0	229.0	35.0	222.833333	9.611111	229.0	(Nb0+, Co0+, Si0+)
1	mp-10010	Al(CoSi)2	164	[[0. 0. 0.] Al, [1.96639263 1.13529553 0.75278...	0.266910	96.252006	175.449907	0.268105	(Al, Co, Si)	13.0	...	0.619388	0.743266	0.0	194.0	227.0	33.0	213.400000	15.520000	194.0	(Al3+, Co2+, Co3+, Si4-)
2	mp-10015	SiOs	221	[[1.480346 1.480346 1.480346] Si, [0. 0. 0.] Os]	0.756489	130.112955	295.077545	0.307780	(Si, Os)	14.0	...	0.000000	0.000000	0.0	194.0	227.0	33.0	210.500000	16.500000	194.0	(Si4-, Os4+)
3	mp-10021	Ga	63	[[0. 1.09045794 0.84078375] Ga, [0. ...	2.376805	15.101901	49.130670	0.360593	(Ga)	31.0	...	0.000000	0.000000	0.0	64.0	64.0	0.0	64.000000	0.000000	64.0	(Ga0+)
4	mp-10025	SiRu2	62	[[1.0094265 4.24771709 2.9955487 ] Si, [3.028...	0.196930	101.947798	256.768081	0.324682	(Si, Ru)	14.0	...	0.000000	0.000000	0.0	194.0	227.0	33.0	205.000000	14.666667	194.0	(Si4-, Ru2+)

5 rows × 142 columns

os_feat=OxidationStates()
data=os_feat.featurize_dataframe(data,'composition_oxid')
data.head()

OxidationStates:   0%|          | 0/1181 [00:00<?, ?it/s]

	material_id	formula	space_group	structure	elastic_anisotropy	G_VRH	K_VRH	poisson_ratio	composition	MagpieData minimum Number	...	MagpieData maximum SpaceGroupNumber	MagpieData range SpaceGroupNumber	MagpieData mean SpaceGroupNumber	MagpieData avg_dev SpaceGroupNumber	MagpieData mode SpaceGroupNumber	composition_oxid	minimum oxidation state	maximum oxidation state	range oxidation state	std_dev oxidation state
0	mp-10003	Nb4CoSi	124	[[0.94814328 2.07280467 2.5112 ] Nb, [5.273...	0.030688	97.141604	194.268884	0.285701	(Nb, Co, Si)	14.0	...	229.0	35.0	222.833333	9.611111	229.0	(Nb0+, Co0+, Si0+)	0	0	0	0.000000
1	mp-10010	Al(CoSi)2	164	[[0. 0. 0.] Al, [1.96639263 1.13529553 0.75278...	0.266910	96.252006	175.449907	0.268105	(Al, Co, Si)	13.0	...	227.0	33.0	213.400000	15.520000	194.0	(Al3+, Co2+, Co3+, Si4-)	-4	3	7	3.872983
2	mp-10015	SiOs	221	[[1.480346 1.480346 1.480346] Si, [0. 0. 0.] Os]	0.756489	130.112955	295.077545	0.307780	(Si, Os)	14.0	...	227.0	33.0	210.500000	16.500000	194.0	(Si4-, Os4+)	-4	4	8	5.656854
3	mp-10021	Ga	63	[[0. 1.09045794 0.84078375] Ga, [0. ...	2.376805	15.101901	49.130670	0.360593	(Ga)	31.0	...	64.0	0.0	64.000000	0.000000	64.0	(Ga0+)	0	0	0	0.000000
4	mp-10025	SiRu2	62	[[1.0094265 4.24771709 2.9955487 ] Si, [3.028...	0.196930	101.947798	256.768081	0.324682	(Si, Ru)	14.0	...	227.0	33.0	205.000000	14.666667	194.0	(Si4-, Ru2+)	-4	2	6	4.242641

5 rows × 146 columns

添加一些基于结构的特性

并不是所有的l featurizers都起作用。Matminer 还可以分析晶体结构并对其进行特征描述。让我们从添加一些简单的密度特性开始。

from matminer.featurizers.structure import DensityFeaturesdf_feat=DensityFeatures()
data=df_feat.featurize_dataframe(data,'structure')#输入structure列
data.head()

DensityFeatures:   0%|          | 0/1181 [00:00<?, ?it/s]

	material_id	formula	space_group	structure	elastic_anisotropy	G_VRH	K_VRH	poisson_ratio	composition	MagpieData minimum Number	...	MagpieData avg_dev SpaceGroupNumber	MagpieData mode SpaceGroupNumber	composition_oxid	minimum oxidation state	maximum oxidation state	range oxidation state	std_dev oxidation state	density	vpa	packing fraction
0	mp-10003	Nb4CoSi	124	[[0.94814328 2.07280467 2.5112 ] Nb, [5.273...	0.030688	97.141604	194.268884	0.285701	(Nb, Co, Si)	14.0	...	9.611111	229.0	(Nb0+, Co0+, Si0+)	0	0	0	0.000000	7.834556	16.201654	0.688834
1	mp-10010	Al(CoSi)2	164	[[0. 0. 0.] Al, [1.96639263 1.13529553 0.75278...	0.266910	96.252006	175.449907	0.268105	(Al, Co, Si)	13.0	...	15.520000	194.0	(Al3+, Co2+, Co3+, Si4-)	-4	3	7	3.872983	5.384968	12.397466	0.644386
2	mp-10015	SiOs	221	[[1.480346 1.480346 1.480346] Si, [0. 0. 0.] Os]	0.756489	130.112955	295.077545	0.307780	(Si, Os)	14.0	...	16.500000	194.0	(Si4-, Os4+)	-4	4	8	5.656854	13.968635	12.976265	0.569426
3	mp-10021	Ga	63	[[0. 1.09045794 0.84078375] Ga, [0. ...	2.376805	15.101901	49.130670	0.360593	(Ga)	31.0	...	0.000000	64.0	(Ga0+)	0	0	0	0.000000	6.036267	19.180359	0.479802
4	mp-10025	SiRu2	62	[[1.0094265 4.24771709 2.9955487 ] Si, [3.028...	0.196930	101.947798	256.768081	0.324682	(Si, Ru)	14.0	...	14.666667	194.0	(Si4-, Ru2+)	-4	2	6	4.242641	9.539514	13.358418	0.598395

5 rows × 149 columns

查看新添加的了什么特性

df_feat.feature_labels()

['density', 'vpa', 'packing fraction']

使用机器学习模型

定义输入数据和输出数据

现在，我们将使用 K _ VRH (体积模量)作为输出。

对于输入，我们将使用我们生成的所有特性。也就是说，除了输出数据和非数值列(如composition和structure.)之外的所有内容。

y=data['K_VRH'].values
del_col=['G_VRH','formula','material_id','elastic_anisotropy','poisson_ratio','composition','composition_oxid','K_VRH','structure']
X=data.drop(del_col,axis=1).values

X.shape

(1181, 140)

使用 scikit-learn 建立一个线性回归模型

from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_squared_error
import numpy as nplr=LinearRegression()lr.fit(X,y)

LinearRegression()

print('training R2=%.3f'%lr.score(X,y))

training R2=0.928

print('training RMSE:%.3f'%np.sqrt(mean_squared_error(y,lr.predict(X))))

training RMSE:19.587

这看起来是合理的，因为线性回归是一个简单(高偏见)的模型。但是，要真正验证我们没有过度拟合，我们需要检查交叉验证得分，而不是拟合得分。

from sklearn.model_selection import KFold,cross_val_score#使用10折交叉和验证
crossvalidation=KFold(n_splits=10,shuffle=True,random_state=1)
scores=cross_val_score(lr,X,y,scoring='neg_mean_squared_error',cv=crossvalidation)rmse_scores=[np.sqrt(abs(s)) for s in scores]
r2_scores=cross_val_score(lr,X,y,scoring='r2',cv=crossvalidation)

print(' mean R2:%.3f'%np.mean(r2_scores))

 mean R2:0.903

print('mean RMSE:%.3f'%np.mean(rmse_scores))

mean RMSE:22.377

import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.model_selection import cross_val_predict

sns.scatterplot(y,cross_val_predict(lr,X,y,cv=crossvalidation),palette=current_palette)
plt.plot(np.arange(0,400,1),np.arange(0,400,1),'r--')

E:\Anaconda\lib\site-packages\seaborn\_decorators.py:36: FutureWarning: Pass the following variables as keyword args: x, y. From version 0.12, the only valid positional argument will be `data`, and passing other arguments without an explicit keyword will result in an error or misinterpretation.warnings.warn([<matplotlib.lines.Line2D at 0x1e13781cb80>]

尝试一个随机森林模型

from sklearn.ensemble import RandomForestRegressorrf=RandomForestRegressor(n_estimators=50,random_state=1)rf.fit(X,y)
print('training R2=%.3f'%rf.score(X,y))

training R2=0.989

print('traing RMSE:%.3f'%np.sqrt(mean_squared_error(y,rf.predict(X))))

traing RMSE:7.687

#使用10折交叉和验证
crossvalidation=KFold(n_splits=10,shuffle=True,random_state=1)
scores=cross_val_score(rf,X,y,scoring='neg_mean_squared_error',cv=crossvalidation)rmse_scores=[np.sqrt(abs(s)) for s in scores]
r2_scores=cross_val_score(rf,X,y,scoring='r2',cv=crossvalidation)

print(' mean R2:%.3f'%np.mean(r2_scores))

 mean R2:0.924

print('mean RMSE:%.3f'%np.mean(rmse_scores))

mean RMSE:19.277

cm = plt.cm.get_cmap('RdYlBu')
plt.scatter(y,cross_val_predict(rf,X,y,cv=crossvalidation),cmap=cm,c=data['poisson_ratio'])
plt.plot(np.arange(0,400,1),np.arange(0,400,1),'r--')
plt.colorbar()

<matplotlib.colorbar.Colorbar at 0x1e1356f21c0>

#划分训练集和测试集
from sklearn.model_selection import train_test_split
X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.2,random_state=1)

rf_reg = RandomForestRegressor(n_estimators=50, random_state=1)
rf_reg.fit(X_train, y_train)

RandomForestRegressor(n_estimators=50, random_state=1)

print('training R2:%.3f'%rf_reg.score(X_train,y_train))

training R2:0.987

print('training RMSE:%.3f'%np.sqrt(mean_squared_error(y_train,rf_reg.predict(X_train))))

training RMSE:8.259

print('testing R2:%.3f'%rf_reg.score(X_test,y_test))

testing R2:0.942

print('test RMSE:%.3f'%np.sqrt(mean_squared_error(y_test,rf_reg.predict(X_test))))

test RMSE:16.928

sns.histplot(x=y_train-rf_reg.predict(X_train),stat='probability',color='r',label='train')
sns.histplot(x=y_test-rf_reg.predict(X_test),stat='probability',color='b',label='test')
plt.legend(loc='best')

<matplotlib.legend.Legend at 0x1e135a1a040>

观察随机森林模型使用的最重要的特性是什么。

importances=rf.feature_importances_
inclued=data.drop(del_col,axis=1).columns

inclued

Index(['space_group', 'MagpieData minimum Number', 'MagpieData maximum Number','MagpieData range Number', 'MagpieData mean Number','MagpieData avg_dev Number', 'MagpieData mode Number','MagpieData minimum MendeleevNumber','MagpieData maximum MendeleevNumber','MagpieData range MendeleevNumber',...'MagpieData mean SpaceGroupNumber','MagpieData avg_dev SpaceGroupNumber','MagpieData mode SpaceGroupNumber', 'minimum oxidation state','maximum oxidation state', 'range oxidation state','std_dev oxidation state', 'density', 'vpa', 'packing fraction'],dtype='object', length=140)

b = a[i:j:s]这种格式呢，i,j与上面的一样，但s表示步进，缺省为1.

所以a[i:j:1]相当于a[i:j]

当s<0时：i缺省时，默认为-1； j缺省时，默认为-len(a)-1

所以a[::-1]相当于 a[-1:-len(a)-1:-1]，也就是从最后一个元素到第一个元素复制一遍。

indices=np.argsort(importances)[::-1]

chart=sns.barplot(x=inclued[indices][0:15],y=importances[indices][0:15],palette='BuGn_r')
chart.set_xticklabels(chart.get_xticklabels(), rotation=90, horizontalalignment='right')

[Text(0, 0, 'MagpieData mean MeltingT'),Text(1, 0, 'vpa'),Text(2, 0, 'MagpieData minimum MeltingT'),Text(3, 0, 'density'),Text(4, 0, 'MagpieData maximum MeltingT'),Text(5, 0, 'MagpieData mean GSvolume_pa'),Text(6, 0, 'MagpieData minimum Column'),Text(7, 0, 'MagpieData maximum GSvolume_pa'),Text(8, 0, 'MagpieData mean Electronegativity'),Text(9, 0, 'MagpieData mode Column'),Text(10, 0, 'MagpieData mode NValence'),Text(11, 0, 'MagpieData mean NUnfilled'),Text(12, 0, 'MagpieData range MendeleevNumber'),Text(13, 0, 'MagpieData minimum MendeleevNumber'),Text(14, 0, 'packing fraction')]

在随机森林模型中，与熔点和每个原子/密度的体积有关的特征是最重要的。