H2O中的隨機森林算法介紹及其項目實戰(python實現)python
包的引入:from h2o.estimators.random_forest import H2ORandomForestEstimator算法
H2ORandomForestEstimator 的經常使用方法和參數介紹:框架
(一)建模方法: dom
model =H2ORandomForestEstimator(ntrees=n,max_depth =m)測試
model.train(x=random_pv.names,y='Catrgory',training_frame=trainData)spa
經過trainData來構建隨機森林模型,model.train中的trainData:訓練集,x:預測變量名稱,y:預測 響應變量的名稱rest
(二)預測方法:blog
pre_tag=H2ORandomForestEstimator.predict(model ,test_data) 利用訓練好的模型來對測試集進行預測,其中的model:訓練好的模型, test_data:測試集。ci
(三)算法參數說明:pandas
(1)ntrees:構建模型時要生成的樹的棵樹。
(2)max_depth :每棵樹的最大深度。
項目要求:
題目一: 利用train.csv中的數據,經過H2O框架中的隨機森林算法構建分類模型,而後利用模型對 test.csv中的數據進行預測,並計算分類的準確度進而評價模型的分類效果;經過調節參 數,觀察分類準確度的變化狀況。 注:準確度=預測正確的數佔樣本數的比例
題目二: 經過H2o Flow 的隨機森林算法,用同題目一中所用一樣的訓練數據和參數,構建模型; 參看模型中特徵的重要性程度,從中選取前8個特徵,再去訓練模型,並從新預測結果, 進而計算分類的準確度。
需求完成內容:2個題目的代碼,認爲最好的準確度的輸出值和test數據與預測結果合併 後的數據集,命名爲predict.csv
python實現代碼以下:
(1) 題目一:
#手動進行調節參數獲得最好的準確率 import pandas as pd import numpy as np import matplotlib.pyplot as plt import h2o h2o.init() from h2o.estimators.random_forest import H2ORandomForestEstimator from __future__ import division df=h2o.import_file('train.csv') trainData=df[2:] model=H2ORandomForestEstimator(ntrees=6,max_depth =16) model.train(x=trainData.names,y='Catrgory',training_frame=trainData) df2=h2o.import_file('test.csv') test_data=df2[2:] pre_tag=H2ORandomForestEstimator.predict(model ,test_data) predict=df2.concat(pre_tag) dfnew=predict[predict['Catrgory']==predict['predict']] Precision=dfnew.nrow/predict.nrow print(Precision) h2o.download_csv(predict,'predict.csv')
運行結果最好爲87.0833%-6-16,以下
#for循環進行調節參數獲得最好的準確率
import pandas as pd import numpy as np import matplotlib.pyplot as plt import h2o h2o.init() from h2o.estimators.random_forest import H2ORandomForestEstimator from __future__ import division df=h2o.import_file('train.csv') trainData=df[2:] df2=h2o.import_file('test.csv') test_data=df2[2:] Precision=0 nt=0 md=0 for i in range(1,50): for j in range(1,50): model=H2ORandomForestEstimator(ntrees=i,max_depth =j) model.train(x=trainData.names,y='Catrgory',training_frame=trainData) pre_tag=H2ORandomForestEstimator.predict(model ,test_data) predict=df2.concat(pre_tag) dfnew=predict[predict['Catrgory']==predict['predict']] p=dfnew.nrow/predict.nrow if Precision<p: Precision=p nt=i md=j print(Precision) print(i) print(j) h2o.download_csv(predict,'predict.csv')
運行結果最好爲87.5%-49-49,以下
(2)題目二:建模以下,以後挑出排名前8的特徵進行再次建模
#手動調節參數獲得最大準確率 import pandas as pd import numpy as np import matplotlib.pyplot as plt import h2o h2o.init() from h2o.estimators.random_forest import H2ORandomForestEstimator from __future__ import division df=h2o.import_file('train.csv') trainData=df[['Average_speed','r_a','r_b','v_a','v_d','Average_RPM','Variance_speed','v_c','Catrgory']] df2=h2o.import_file('test.csv') test_data=df2[['Average_speed','r_a','r_b','v_a','v_d','Average_RPM','Variance_speed','v_c','Catrgory']] model=H2ORandomForestEstimator(ntrees=5,max_depth =18) model.train(x=trainData.names,y='Catrgory',training_frame=trainData) pre_tag=H2ORandomForestEstimator.predict(model ,test_data) predict=df2.concat(pre_tag) dfnew=predict[predict['Catrgory']==predict['predict']] Precision=dfnew.nrow/predict.nrow print(Precision) h2o.download_csv(predict,'predict.csv')
運行結果最好爲87.5%-5-18,以下
#for循環調節參數獲得最大正確率 import pandas as pd import numpy as np import matplotlib.pyplot as plt import h2o h2o.init() from h2o.estimators.random_forest import H2ORandomForestEstimator from __future__ import division df=h2o.import_file('train.csv') trainData=df[['Average_speed','r_a','r_b','v_a','v_d','Average_RPM','Variance_speed','v_c','Catrgory']] df2=h2o.import_file('test.csv') test_data=df2[['Average_speed','r_a','r_b','v_a','v_d','Average_RPM','Variance_speed','v_c','Catrgory']] Precision=0 nt=0 md=0 for i in range(1,50): for j in range(1,50): model=H2ORandomForestEstimator(ntrees=i,max_depth =j) model.train(x=trainData.names,y='Catrgory',training_frame=trainData) pre_tag=H2ORandomForestEstimator.predict(model ,test_data) predict=df2.concat(pre_tag) dfnew=predict[predict['Catrgory']==predict['predict']] p=dfnew.nrow/predict.nrow if Precision<p: Precision=p nt=i md=j print(Precision) print(i) print(j) h2o.download_csv(predict,'predict.csv')
運行結果最好爲87.5%-49-49,以下