神经网络软件操作手册Neurosolutions0.docx

《神经网络软件操作手册Neurosolutions0.docx》由会员分享，可在线阅读，更多相关《神经网络软件操作手册Neurosolutions0.docx（7页珍藏版）》请在三一文库上搜索。

1、1、打开Neurosolutions,进入以下界面Fr知独访的. BStlDSDAPni点击NS Excel 星 按钮，在 Excel加载项中出现 Neurosolutions项。汨文件幅辑阴柳图5施瓦 格式回工具数据 宣口口帮助 旦3陶URiw口直。鼻01弯W里工7141修 E宋体ABCDEFGHI12342、标记数据选定指标数据列(x1x5),点击 Neurosolutions 菜单选择 Tag DataT Column(s) As Input 选项，将(x1x5)标记为输入。选定指标数据列 y,点击Tag Data下Column(s) As Desired 选项，将y标记为输出。选定第

2、1到53个样本所在的行，点击 Tag Data下Row(s) As Training 选项，将其标记为训练集。最后选定第53到58个样本所在的行，点击Tag Data下Row(s) As Testing选项，将其标记为测试集。标记完成后界面如下。 MkrwOfl Ewcel - Dwrtt 1ABCHBF1d3*5y 12D. M67235W0.3472377B6QJ 的22390.副1丽95G.3 韭 503S29在3#项引34解50 3=47190BS70.7731803U. S1S121844 347ml60.3471 6355A 3号了8264844雨6Q. 34T75W4施0. 54

3、782641340. 日陷986910. 34T啾钟吕Q.3肝5门制10 34S144555d()7Cl羽科g打结a。31回题o. Balsas0. 34SlfidO2&山3肝日咽曲A 34附第艘6.34S35T490. 3480402230. B1CT877031 34B5?0390L 347999112仁的副第打74班g0,34789703o. 5A80203910.8132473S制的网嘴M34T? 82Ml1. 34&43SS8B40fi700. 34781 OBffT% S48-435G79凡 B1OT897020.34843S6790.3476319790. 34网若僦0. M83

4、26D570r SI 1222652Q. 34BM71660L54SZ4ffil8a S4E4S 理S4(*720. 382351530.34S437530_ BimO078&0. 34E&a80350.3猛。必95Q. 34的34前4沏3口 34兆也刻0, 35342320. 81826712Q. 3406773910.54B347196已 S4F598E3406174口, 34JS27035S0, S4769822&0P 8151-97138d. 34203550.347651373孔狷W曲海TB0 .涉P犯&(西0. 3473139210. 80fT824526S 3478038CL 3

5、476053830.4熊0. 3430261410. J48096&07o. etS461439S 3462225956制防93既2%”鹏如4 am0. 3-4S0S25E70. 273641520u B073S1222. 34E168110.3477T6787A的摩3不4即口 MT7559斫0.网睚Q1T3T0. 3%伺59560. 3裁花4354334门叼】9后M屏M叼然而d而9&. 3470435e0. 3116319&30-7&78430770.347277390.34肝跑知A 34&18&74解Q0. 3469281390. 3466106740. 8Q2J2214734714574

6、7.346354362爪 S48 0 48E84照110.346604440.60452818Cffi2143176 3犷您1390,34&012173。.才454打儆4020. 360391120. 346n1 能9口 7WL7S040. 34623970.345BU3471A4两3D. 346411BZ90. M6719211Q J 潮Q89290.34&T380170.34BP9我 4我5劭厘罪4燃4- 3兆蚪Ml0. 3457279220.S047201350.34&443196帆34E637818Q, 3*5824494陶5D. M55571120. 345324492Q, T9用4

7、44120. 3459412090L 3蜘网15& J/5W55Jd幽6D. M53741T2肥苗5&貂0. T9CM榔CL 345&6S027Q. 3452d?M3注3城划熬44O&870. M5T815490. 3458253900u791156522 3460172110l 345&5B161帆345阪鳖54陶gD,诃5893捌0.列钝99立1o. 7&1M7646Q. $45105霏5晒1神骂0, /39旷901/80950440. 3459907910. 7603430. 34619480.3457S3214A电4T 83、预处理数据点击 Neurosolutions 菜单，选择 P

8、reprocess DataT Randomize Rows 选项，完成数据处理, Excel 出现 sheet1Randomize.4、建立BP神经网络模型选择Excel的sheetl工作表界面，点击Neurosolutions菜单，选择 Create/Open Network下的New Custom Network选项，出现以下界面。y文件 小IREQ!） ffiAQ）格式明）CD）苦口底J怦助aI口二0匕&4/&z ft H U HI B 碎K2S -&NeuroSolutions 卜DiincnsioiK Inc.I The NeumIHMu曲曲m Em 卜国训m皿l-QND.JDEA

9、A3J2-377-5I44Fs 352.377-900?EjndJ: tnfb（nd c0*. nd. C4H1Thank oufor installing the Evaluation Edition of NeuroSoliilians This software will help you gain a be（ler understanding of neural networics by illustrating key concepts with numerous Iiyb, interactive demos In additiont you will have 1he abili

10、ty Io train 国nd teat neural nelwgrks with your Qi dala （some 话别ricti口口嚣 apply）. Obtain 疝，unrestrictedl lic-ftezincti?i4068040635mis40675OrdAtr 3k*403 40679. 4078, 40710. 40703料695 虫避的 W网 40697. 406T7. 40693 406T6 虫：IT IT 40705. 40709 4Q网 40632 40719 407064Q沏 40637点击迪逅二I按钮,出现模型参数设置界面。, Mfcrowft EkeH

11、曲确 1Ii5呼町EdiiiHem AlBuildeiGengrifeXiTSd Fnd F*也P 曲力如lor M受uroj HatworkJor duKirELTiiri VietYrkFrlucipal Cmjionent Aoalysa s 里EEP/GBWEW Ik-kSelf-OrcaDiziixe F电自tore N&p JTet Tin/-Lg器 E事CUsrenR NetworX O-CTITT bD.t 4-tviarkCAEFIS Metwork (Juzzy Lag: c) Support Veetbi Na.chineMWtU 野穹*(MLPs) art liyerc

12、 d fta dfoiwsrd networks typic ally t：aiiisd with, static b耽pi印蝇，tig These networks have found thcsrNeural MndelWelcaffie Id thegNeuralBuild.e3,. SLartkigwith your dal4P this tool 碗11 声dk youthiough the process of designing and lamag a neurd network There 掘e ntany diflFcient types of neural iifliwot

13、fcs, hul most cia b b el8sMwd belonging to one of the majorparidigpis lifted t-o th left. Eich pandagm will have advantages 触id 出的dy5ML- depending on. yourpmticuloj axplicahMi. TKalp将设置隐层层数。出现界面如下。设置1个隐层层数,K&lf口口 |4点击区堂n按钮,对隐层参数进行设定，出现界面如下。设置隐层处理单元为为0.7,都为系统默认值。点击4个，激活函数SigmoidAxon,学习规则选择 Momentum,步

14、长为1,冲量二立二1|按钮,对输出层进行参数设定，出现界面如下。TieurrilBuflderOutput LayerThis pBJiBlis used to specify the parameters n layer ofprocessing elements (TEs).NewQSoluhckns sunudations are vectoi based for efficiency. This implies that- each layer GonlaiiLE a vector ofPEs and that 由甘 pajameters seketed apply to ihe en

15、tiie Tactor. The puuiitUrs： uc dependent on th neutfd nw 金 L but 皿 re-qiuire a nonltneatty function 加 specify theO.CEE步长为0.1,冲量为0.7,都为系统默认值。点击仁全3按钮，设置最大迭代次数，出现界面如下。色 IMeurdlBuildti1 xSupervised LearningIIIAiiimum EpochsThe M axbnutn Ep ochs Held epeciHes how many iterations (over the training s e) w

16、ill b e don? if n-o other mtenon kicks in. Tht Errof Chang# box contains the patainelers|booc甲煦Threshol|0 01,P* lininwTraining: St,IncrmDtn lucresLoad Best on Teused to tenttmate the tidning base d on me an 叫uand eiTDf.The NeuralBuildei has MSE terttunahoiiIfei ght UpdateC On-LineBatchActivat-B d by

17、 defauli. T terminate the training strictly bastd on thewHipd 七 | W |设置最大迭代次数为 6000次，选定MSE按钮，设置标准误差参数为0.01,选择批量更新Batch按钮。点击广I按钮,出现以下界面。点击按钮,BP神经网络模型设立完成，出现界面如下。绘 fibgront Ipok 出tHpO 0 急 像电 诙 云 度！ 毕Rtf*amgnNBOMmW江电CHVF-IGiiTidingQ*UNrCnhrHf点击丑气I按钮，得至U Trainltraining Report和Trainltraining MSE 两张工作表，结果

18、如下。5、训练神经网络选择 Excel 的 sheetIRandomized 工作表界面，点击 Neurosolutions 菜单，选择 Train Network 下的Train选项，出现以下界面。在Train name中填入名称为train1 ,迭代此处为 6000次，点击二EZZI按钮，出现以下界Ctitput LocatkinTrlaJ Name:| franlTrailing OptwnsNumber of Epoch*| 6000-回 Randonue Initial WeightsCross Validation Teinwiitbn Termrate after | LOO e

19、pochs 同口 irprav-DniEritFar Ulas研Nation prcblBnis, rnake desses eveniy weightedHelpOK I CanedEvaluation biode DescnptmIhank ywj for hstalirg the cvakidtkxi version of NeiroSaliitiorK For Escel. Ths software is identacal to the reikase version of hJetira5c1utlon for Excel 劭tEt for the Fdkmng 匕利冶EMr；Li

20、mitations Mcimufn of 3 outputs& Maximum of 30 inputs-Maximum of 30口 exemars-Mftximufn of 30QQ tiriput$ k 珈的叩匕You iw u$e the ev-duetisn wson tf this stftwre Itw of charge For &0 days. F y?u waukl livto purchaw an unreftrlttiEd verslan of NeuroSdUtung fcr Exaelj ple-ase caritad: Meur口Dirnensn ta place

21、 an QT*JOrder Info；Cor*att NeuroDirncnfignHeurdSoliitiMiK rfireMiboardflEyji Luation ModisTrilnMSE versus Epoch0.060.050.04Training MSE6001199 1798 2397 2996 3595 4194 4793 5392 5991Epoch0.030.020.0101Best Network Training Epoch #6000Minimum MSE0.00221929Final MSE0.002219296、测试神经网络 选择Excel的sheet1Ran

22、domized工作表界面，点击 Neurosolutions菜单，选择 Test下的test 选项，出现以下界面。在Trial name中填入名称为 test1,在Dataset to Test中选择 Training ,点击1HI按钮,出现以下界面。E 汨 uabon Mode DcriptianThar杰 you forthe eveiuion version ofNeuroSokitions fw Ek6sL This software is identic to the rddase varsim of MeuroSditians Mr Ztel except For the faHo

23、vdng Imltatians：Umitathns- Mau册5 印 3 OUtpLAfi- Max hi urn of 30 inputs- MaMimijnn of 300 exemplars- Maamim oF 3000 (inputs； x exemptars)You may use the evaluation vetsian of th 缶 5 oft Here free of charge far 60 das. If ytxi add I融 to purche wi unrKtrktedl verSicn of MEurg&kAnm for ExceL pkase： cont

24、act HsuroOimertsion to place -an order.Order InFo i Cort.act MauroDimensKjn CIkwPerformanceyMSE1.59298E-07NMSE0.049432915MAE0.000334578Min Abs Error1.11036E-06Max Abs Error0.000899092r0.975037668点击Close按钮，得到 Testltesting Report和Testltesting MSE两张工作表，结果如下。Desired Output and Actual Network Output0.349yy Output7、预测次日收盘价点击Neurosolutions菜单，点击 Tag Data下Row(s) As Production的选项，确定要预测的样本。然后点击 Neurosolutions菜单下 Apply Production Dataset的最终得到预测数据。0. 343767650. 344063048Q. 343938760. 344078318