Federal gosudarstvennoe uchebnoe predpriyatie Chair of the System of Artificial Intelligence

4.2 Forecast forecast in Sosnovoborsk
140
Machine Translated by Google

---

Militia -organy vnutrennix del
otdelnyx politikov, obladayushchix otnositelnoy svobodoy deystviy, tayat v
information received in the commissions on the organization of promejutochnyh selections
Categories of work
prognozirovat, chto budet delat president, prime minister, leader
Sex - Paul: (male - 1, female - 0)
Worker -rabochiy
GKX - JKX
Leader - rukovoditel
presidential administration and political oligarchs, for example, Boris
poslevuzovskoe
Nowhere - nigde
vyborov 14.01.2001, provodimyx v gorode Sosnovoborske. Dannye sobrany i
Deputy - Byl li deputatom (da -1 net - 0)
Pensioner - pensioner
Kids - The number of children
institutes, nakonets, svoix corporate interests. Deystviya je
golosovaniya, izlojennym v mestnoy gazete «Rabochiy». Osnovnaya
Medicine - medicine
sebe gorazdo bolshe neopredelennosti. V silu etogo gorazdo legche
na neskolkix izbiratelnyx uchastkax.
Parameters: Birthday - Birth of God
Jobleess - useless
KZAP - KZAP
Employee - slujashchiy
dumskoy fraktsii ili gubernator, i trudnee proschitat povedenie figur iz
Place - Mesto rozhdeniya: (Krasnoyarskiy kray - 1, drugoe - 0)
CHP -chastnÿe predpriyatiya
Shool - obrazovanie Krasn -
rabota v Krasnoyarske
TEC - TETs
Berezovskiy and others.
Byla postavlena zadacha poluchit prognoz rezultatov vyborov, kotorye
sostoyatsya 23.06.2002 ÿÿ osnove dannyx sobrannyx po rezultatam
Livesosn - S kakogo goda projivaet v Sosnovoborke
Business - entrepreneur
grouped by biographies of candidates, articles and results
formalnyx i informalnykh pravil, pozitsiy otnositelno drugix
Familu - Semeynoe polojenie (ne jenat (zamujem) - 0, jenat (zamujem) - 1)
Place of work (sphere)
141
Machine Translated by Google

---

Livregion - Projivaet li kandidat na izbiratelnom uchastke na kotorom
EDIN - Edinstvo
company: (net-0, da - 1)
Bisiness - Podderjka small business
Grammat - Osvedomlennost, gramotnost v voprosax politiki (Da-1, net- 0)
company: (net-0, da - 1)
Hope - Hope and support
Syn - Privlechenie investment
Leaflet - Rasprostranyalis li listovki (da-1, net- 0)
Pension - Problems of pensioners
Publwork - Public organizations :( Da- 1, net - 0)
Quant2 - Number of votes in 2 rounds (14.01.01)
Roller - Televizionnye roliki s vystupleniyami avtoritetnyx lits
District - ÿ izbiratelnogo uchastka
Predvybornÿe obeshchaniya - tezisy v presse:
Municp - Blagoustroystvo goroda (Da-1, net-0)
Power - Voprosy raspredeleniya vlasti v gorode
Meeting - Vstrechi s izbiratelyami (da-1, net- 0)
Factor - Ozdorovlenie zavoda
KPRF - KPRF
Publicp - Polojitelnye publikatsii v presse v period predvybornoy
Welfare - Uluchshenie blagosostoyaniya naseleniya
Effect -Kommunikabelnost (umenie proizvesti vpechatlenie) (da-1, net- 0)
CHEST - Chest and Homeland
Public -Otritsatelnye publikatsii in the press in the pre-period
Narcotism - Problems of drug addiction
ballotiruetsya (da-1, net- 0)
Reconstruction -Reconstruction of the poor help
LDPR - LDPR
Quant1 - Number of votes in 1 round (14.01.01
Against1 - Number of votes against all in the first round
Ether -Television broadcast (da-1, net- 0)
city (da-1, net- 0)
Bonus - Nagrady: (net - 0, professional -1, boevye - 2)
Against1 -Chislo progolosovavshix against all in the second round
Newjob - ÿÿÿÿÿÿÿÿ ÿÿÿÿÿ ÿÿÿÿÿÿÿ ÿÿÿÿ
Amount - the number of candidates per plot
Political parties
Commynal - Reshenie kommunalnyx problem
Order - Obshchestvennyy poryadok
Renown - Izvestnost v gorode (ne izvesten -0, polojitelnaya izvestnost - 1, otritsatelnaya
-2)
142
Machine Translated by Google

---

The text, obÿyasnyayushchiy process resheniya neyronnoy setyu task. A poskolku
setey (NewroPro0.25), rabotayushchiy v srede MS Windows 95 ili MS Windows NT
4.0 i pozvolyayushchiy proizvodit sleduyushchie bazovÿe operatsii:
classification. 6.
Testing of neural networks on the data file, receipt
used traditional mathematical methods (regression)
znacheniy i dobavlenie novyx zapisey v bazu dannyx; soxranenie file
storage of significant data in a text file on disk;
trebuemaya tochnost prognoza. Tak je, zamechu, chto ot imeyushchixsya v nastoyashchee
nalichie vozmojnostey tselenapravlennogo uproshcheniya neyronnoy seti dlya
chislom sloev neyronov ot 1 do 10, chislom neyronov v sloe - do 100;
save the verbal description to the text file on the disk;
classification. Neyronnaya set mojet odnovremenno reshat kak neskolko
prognosis, setting neyronnoy seti.
sovokupnosti s postroeniem ee verbalnogo opisaniya pridaet
produktta vozmojno dlya resheniya lyuboy zadachi klassifikatsii ili prognoza,
kotoraya reshaetsya pri nalichii vÿborki dannyx i dlya resheniya kotoroy ranee
predstavlyayushchiy soboy manager obuchaemyx iskusstvennyx neyronnyx
zadach klassifikatsii, a takje odnovremenno zadach i prognozirovaniya, i
porojdeniya znaniy iz tablitsy dannyx. Under znaniyami zdes ponimaetsya
1. Creation of a neuroproject;
2. Podklyuchenie k neyroproektu fayla (bazy) dannyx v formatate dfb
(dBase, FoxBase, FoxPro, Clipper) ili db (Paradox); 3. Redaktirovanie file dannyx -
izmenenie sushchestvuyushchix
statisticheskoy informatsii o tochnosti resheniya zadachi; 7.
Vychislenie pokazateley znachimosti vxodnyx signalov seti,
analiz, neparametricheskaya statistika i drugie), odnako ne bÿla dostignuta
odnim iz preimushchestv neyronnyx setey yavlyaetsya vozmojnost resheniya
vremya neyrosetevyx programmnyx produktov dannyy produkt otlichaet
dannyx in drug formate; 4.
Addition to the project neyronnoy seti sloistoy architecture s
8. Uproshchenie neyronnoy seti; 9.
Generation and visualization of the verbal description of the neural network,
5. Obuchenie neyronnoy seti resheniyu zadachi prognozirovaniya ili
10. Vybor algorithm obucheniya, naznachenie trebuemoy tochnosti
posleduyushchey generatsii verbalnogo opisaniya.
A tak je nalichie razvityx vozmojnostey po uproshcheniyu seti v
zadach prognozirovaniya (prognozirovanie neskolkix chisel), tak i neskolko
For analysis dannyx bil vibran software product
NewroPro0.25 was selected due to the application of this program
predlagaemomu product new potrebitelskie svoystva - vozmojnost
143
Machine Translated by Google

---

stroki; proportional to uchastkam). The results are some
This is done because not all input signals are set in the synapse
seti.
platforme. -
Set mojet priobresti logicheski prozrachnuyu structure. Apparently, chto
sets of signals to zadavaemogo user.
(Table 2 and 3)
- Neyronnuyu set bolee prosto mojno budet realizovat na apparatnoy
algorithm resheniya kotoryx ne izvesten), to dannyy tekst predlojit odin iz
Mojno dostatochno silno uprostit set bez uxudsheniya tochnosti
mojno popÿtatsya postroit algorithm resheniya setyu zadachi na osnove
neobxodimy dlya pravilnogo resheniya zadachi setyu.
neformalizovannyx zadach klassifikatsii i prognoza (tex zadach, yavnyy
pochti nevozmojno ponyat, kak obuchennaya neyronnaya set reshaet zadachu.
After uproshcheniya neyronnaya set stanovitsya dostatochno obozrimoy i
neodnorodnyx vxodov neyronov seti. 5.
Ravnomernoe uproshchenie seti - sokrashchenie chisla prixodyashchix na neurony
set yavlyayutsya
sleduyushchie: - Sokrashchaetsya chislo vxodnyx signalov set. Esli correctly reshit zadachu
Prichem bÿlo primeneno 2 sposoba razbivki dannyx (chetnÿe i nechetnÿe
mojno na osnove menshego nabora vxodnyx dannyx, to eto mojet v
Dlya uproshcheniya neyronnoy seti prodelÿvalis sleduyushchie operatsii:
1. Sokrashchenie chisla vxodnyx signalov - udalenie naimenee znachimyx
algorithms resheniya takoy zadachi.
Execution of the course project was divided into 2 stages:
resheniya zadachi. Osnovnymi rezultatami provedeniya protsessa uproshcheniya
(obuchayushchuyu and testovuyu).
graficheskogo predstavleniya ili verbalnogo opisaniya struktury seti.
3. Sokrashchenie chisla synapsov - udalenie naimenee znachimyx sinapsov seti.
4. Sokrashchenie chisla neodnorodnyx vxodov - udalenie naimenee znachimyx
razlichiya (tablitsa 4.1)
Xarakteristika seti dlya oboix sposobov bÿla odinakova. After
testing, it was proved that the neuronal network has improved.
information.
vxodnyx signals (okazalo silnoe vliyanie na uluchshenie rezultata); 2.
Sokrashchenie chisla neyronov - udalenie naimenee znachimyx neyronov
dalneyshem sokratit vremennÿe i materialnye zatraty na sbor
144
Machine Translated by Google

---

place of work - JKX, and so it is provedenie televizionnyx efirov i pokaz
vzyaty vse dannÿe vyborov 14.01.01, a v kachestve testovoy vyborki dannÿe po
Layer 2 - 5 neurons
rollers, place of work did not have bolshoy znachimosti. At the
second stage of issledovaniya in quality obuchayushchey vÿborki bÿli
Bylo obucheno 20 setey for oboix variantov razbieniya dannyx. ÿÿÿÿÿÿÿÿ
ÿÿÿÿ ÿ ÿÿÿÿÿ ÿÿÿÿÿÿÿ ÿÿÿÿ ÿÿÿÿÿÿÿÿÿÿ, ÿ ÿÿÿÿÿÿ: Chislo vxodnyx poley -
57 Chislo vxodov seti - 57 Chislo vyxodnyx poley - 1 Chislo vyxodov seti -
1 Sloy 1 - 3 neyrona
neodnorodnyx vxodov neyronov k vÿdelennym znacheniyam.
parameters and analysis of results.
When training on the selection of "chetnÿe-nechetnÿe":
candidates on election 23.06.02.
Takje bÿlo provedeno uproshchenie seti, opredelenie znachimosti
6. Binary synapse set - privedenie znacheniy vesov synapsov i
Layer 3 - 7 neurons
vyborov (Prilojenie 1), provodimyx 14.01.01 v g. Sosnovoborske, byli
different.
When disassembling "chetnye-nechetnye" samym znachimym parametrom bÿlo
razbity na dve vÿborki testovuyu i obuchayushchuyu (prichem razbivka
Tsikl obucheniya
Shag Srednyaya
otsenka Pravilnÿe
resheniya - 32 of 32
Operations 2-6 did not affect the test results. It was determined
the importance of the parameters (Table 4 and 5). V
At the first stage of the study, the results were selected according to the results
zavisimosti ot sposoba razbivki dannyx znachimost parameters poluchilas
Tsikl obucheniya - 11 Shag -
0,03028921 Srednyaya
otsenka - 0
Pravilnÿe resheniya - 32 iz 32 Pri
obuchenii
pri vyborke «proportsionalno po
uchastkam»:
reklamnyx rollers. When
dismantling "on the plot" the most important parameter is shown
stroke; dannÿe delilis proportional to uchastkam).
-
9 - 0.1568877
- 0
provodilas po dvum printsipam: dannye delilis na chetnye i nechetnÿe
145
Machine Translated by Google

---

rezultatax kandidatov pod nomerom 1 i 3 daje usilenie etix pozitsiy
neskolko bolshe (ot 5 do11) chem pri razbivke po uchastkam (ot 4 do 10)
printsipy, chto i na pervom etape.
Testing results and test results after uproshcheniya
dannyx bolshe.
Vyvody
testirovanii
bolshee chislo golosov naberet kandidat pod nomerom 5 na 7 izbiratelnom
With uchetom togo, chto mestnaya televizionnaya company rekonstruiruetsya
veroyatnostyu na 5 izbiratelnom uchastke bolshe golosov naberet kandidat
«Chet-nechet» prakticheski vsegda umenshaetsya v 2 raza, krayne redko
sotsiologii »bÿli sobrany i proanalizirovany dannye po dvum
Neojidanno bolshaya znachimost na vtorom etape issledovaniya u
umenshaetsya (tablitsa 2, 3 sootvetstvenno)
Znachimost parametrov oboznachena v tablitse 4 i v tablitse 5. Na
vtorom etape issledovaniya v kachestve obuchayushchey vÿborki bÿli
sdelat sleduyushchie prognosis.
Analyzes the importance of parameters in the fact that it is selective
proizvoditsya s bolshey tochnostyu, chem pobeda. This and ponyatno - takix
Diapazon oshibok pri razbivke po chetnÿm i nechetnÿm strokam
candidates on election 23.06.02. Bylo obucheno 20 setey. Applied to you
Silno popravit polojenie kandidata mogut televizionnye efiry, lichnye vstrechi s
izbiratelyami, televizionnye roliki. Pravda na
The number of training cycles does not affect the number of shifts
privedeny v Tablitse 6, 7. Takje privedena znachimost parametrov.
According to the results of the second analysis, the analysis can be assumed that
positively not mentioned.
uchastke, na 8 izbiratelnom uchastke kandidat pod nomerom 4, i s nebolshoy
At sokrashchenii chisla vxodnyx signalov chislo oshibok pri razbivke
With uchetom teoreticheskix znaniy izlojennyx v punkte «Dokumenty v
veroyatnost takix shagov maloveroyatna.
parameter «semeynoe polojenie».
uvelichivaetsya, pri razbivke po uchastkam chislo oshibok neznachitelno
izbiratelnym kampaniyam. On osnovanii prodelannoy raboty mojno
under number 8.
vzyaty vse dannÿe vyborov 14.01.01, a v kachestve testovoy vyborki dannÿe po
The results of the test are given in Table 1.
On the first chasti issledovaniya prognozirovanie porazheniya
deyatelnost zakanchivaetsya 22.07.02 mojno dat rekomendatsii kandidatam.
146
Machine Translated by Google

---

7. Training and optimization. Teaching methods: gradient, sluchaynÿy, partan
and dr. kvazinyutonovskiy and sopryajennyx gradients; odnomernaya
3. Biological neuron
13. Otsenka raboty seti
6. Input and output signals, functionalization, training, testing, evaluation
12. Tasks for neural networks: mathematical and practical tasks
1. Basic architecture and types of neural networks: sloistye, polnosvyaznÿe,
sigmoidnye, monotonnye; neyroseti s uchitelem i bez
8. Obuchaemye neuroset. Training for prime ministers, pages, for everyone
neuroimitator
optimization