Riba:
- Ikon kafa aikin bincike mai maimaitawa wanda ke ɗaukar nauyi da tsaftace telemetry, gwaji da bayanan samarwa tare da pandas.
- Ikon fahimta da tabbatar da layin fitarwa ta layi yayin karɓar lamba, halaye da taimakon gani daga hankali na wucin gadi
- Ikon tantance sakamakon bincike don daidaiton raka'a, zubewar bayanai da sakewa
A cikin raka'o'in da suka gabata, mun yi amfani da hankali na wucin gadi kamar "mai ba da shawara". A cikin wannan rukunin, za mu ci gaba mataki ɗaya kuma mu kafa tsarin aiki wanda ke nazarin bayanan mota da hannunmu, ta amfani da Python. Manufar ba shine a sanya ku mai haɓaka software ba; Shi ne don yin injiniya wanda zai iya fahimta da kuma tabbatar da layin lambar ta layi yayin samun taimakon lambar daga AI. Domin lambar da AI ta samar na iya zama kuskure, kamar rubutun da AI ta samar; na iya rikitar da ƙara, ɗigo bayanai, ginshiƙin kuskure na tsakiya. Gudun lambar ba tare da fahimta ba kamar buga rahoton da ba a sa hannu ba ne.
Python me yasa? Domin shi ne ma'auni na de facto a cikin nazarin bayanai: zaka iya sauƙaƙe sarrafa na'urorin telemetry, gwaji da samarwa tare da pandas (tabular data), ƙididdiga (aiki na lambobi), matplotlib (makirci) da scikit-learn (injin koyo) dakunan karatu.
Matakai shida don sake yin nazari
Tsayayyen bincike koyaushe yana bin tsari iri ɗaya:
- Load: Karanta bayanai daga fayil ɗin.
- Dubawa: Girma, ginshiƙai, nau'ikan bayanai, ƙimar da suka ɓace.
- Tsaftace: Bace/fita, naúrar, gyaran tambarin lokaci.
- Cire fasalin: Samo masu canji masu ma'ana.
- Analysis/samfuri: ƙididdiga, gani, ko samfuri.
- Tabbatar da bayar da rahoto: Samar da sakamako, duba ƙara/yaɗi, rikodi.
Tukwici: Lokacin tambayar AI don lambar, ce "yi bayanin abin da kuke yi a kowane mataki kuma ku rubuta tunanin ku." Don haka zaku iya tabbatar da lambar yayin da kuke karanta ta.
Misali na mataki-mataki: nazarin telemetry
Lambar mai zuwa tana ɗaukar rikodin CAN/ telemetry kuma tana yin bincike na asali. (Lura: tabbatar da fahimtar lambobin kafin gudanar da su; Sunayen shafi sun bambanta dangane da bayanan ku.)
shigo da pandas a matsayin pd# 1) Load: CSV mai dige-dige, ginshiƙi na farko timestampdf = pd.read_csv("telemetry.csv", parse_dates=["lokaci"])# 2) Checkprint(df.shape) # layuka nawa, ginshiƙan nawa (df.dtypes) # nau'in kowane ginshiƙi (df.dtypes) # adadin ko kowane ginshiƙi. adadin ƙimar da aka ɓace a kowane shafi
Fitowar siffa () ita ce damarku ta farko don tabbatarwa: idan ƙimar max ɗin ingin ya kasance 45,000 rpm (nau'in fasinja na yau da kullun ~ 7,000 rpm), akwai naúrar ko kuskuren firikwensin.
Abubuwan da aka fi amfani da pandas akai-akai suna umarni a cikin matakin duba da abin da suke yi:
umarni
Me yake aikatawa
Me ya tabbatar?
df.siffa
Adadin layuka/ginshiƙai
Shin girman da ake sa ran?
df.dtypes
Nau'in ginshiƙi
Rukunin lamba ya zama rubutu?
df.isna().sum()
Ƙimar ƙima ta ɓace
Nawa sarari yake?
df.bayyana()
Min/max/matsakaici
Shin yana da hankali a zahiri?
df.kwafi().sum()
kwafi jere
Akwai rajista biyu?
# 3) A bayyane: yi alama akan ƙima na zahiri da ba zai yiwu ba
Tsanaki: Kar a share bayanan nan da nan; Da farko ka fahimci dalilin da ya sa yake da ban mamaki. Shin wani lamari ne na gaske (duba kwatsam) ko gazawar firikwensin? Share a makance na iya ɓoye ainihin laifin.
# 4) Cire fasalin: ƙimar hawan zafin jiki (wanda aka samo asali) df = df.sort_values ("lokaci") df ["sic_increase_speed"] = df ["motar_sic"] .diff () / df ["lokaci"].diff () .dt.total_seconds () # 5) Sauƙaƙe na gani. pltplt.plot(df ["lokaci"], df ["motar_sic") plt.xlabel("Lokaci"); plt.ylabel("Injin zafin jiki (C)")
Hana zubewar bayanai a cikin Python
Kuskure mafi haɗari lokacin gina samfurin tsinkaya shine zubewar bayanai (raka'a 5): lokacin da ƙirar ta ga bayanan da ba za a iya sani ba a lokacin tsinkaya. Dokar zinariya don kauce wa wannan a cikin jerin lokaci: jirgin kasa tare da baya, gwada tare da gaba - babu shuffing bazuwar.
daga sklearn.model_selection shigo da jirgin kasa_test_split# FALSE: bazuwar raba jerin lokaci yana leaks gaba # df["lokaci"]> bakin kofa] # 20% na ƙarshe
Tsanaki: train_test_split yana jujjuya bayanan ta tsohuwa (shuffle=Gaskiya). A cikin jerin lokaci, wannan yana rikitar da makomar gaba tare da ilimi kuma yana haifar da babban maki na ƙarya. Idan AI ya yi wannan a cikin lambar da yake samarwa, tabbatar da gyara shi.
Daidaiton raka'a: shiru kisa
Mafi kuskuren kuskure a cikin mota shine kurakurai naúrar: km/h zuwa m/s, Nm zuwa lb-ft, mashaya zuwa kPa, °C zuwa K. Tsayawa ƙamus na abin da sashin kowane shafi yake cikin bincike da rubuta jujjuyawar a sarari yana ceton rayuka.
# Rubuce raka'o'in a bayyane = {"gudun": "km/h", "motor_sic": "C", "matsi": "bar"}# Fassarar juyawa idan ya cancanta (km/h -> m/s) df ["speed_ms"] = df ["gudu"] / 3.6
Karamin karatu
Case 1 - Kuskuren kama tare da siffantawa(). Wani manazarci yana gudanar da lambar daga AI kuma yana yin kiyasin kewayon; Sakamakon yana da girma. Lokacin da muka kalli fitowar () da aka kwatanta, zamu ga cewa an shigar da ƙarfin baturi a cikin Wh a wasu layi kuma a cikin kWh a wasu (banbancin sau 1000). Lokacin da aka kawo naúrar zuwa nau'in uniform, sakamakon yana inganta. Ƙarshe: Ƙididdiga mai sauƙi ta hana mummunan tsinkaya.
Hali na 2 - tarkon rudani. Samfurin lalacewa na birki ya kasance daidai 98% akan saitin gwajin. Lokacin da aka bincika lambar, za a iya ganin cewa train_test_split(shuffle=Gaskiya) da jerin lokaci sun haɗu, ma'ana cewa maki na gaba suna shiga cikin horon. Lokacin da aka raba ta lokaci, daidaito yana raguwa zuwa 80%, amma yanzu ya zama gaskiya. Kammalawa: Kawai saboda lambar AI ta yi aiki, ba daidai ba ne; Mutum ya kama ruwan.
Hali na 3 - Fahimtar abin da ya wuce. A cikin rikodin dorewa, lambar AI ta atomatik tana share ƙimar ƙima. Injiniyan yana kallon wuraren da aka goge; Waɗannan su ne ainihin lokacin ƙaddamar da gwajin da aka yi sha'awar. An cire sharewa kuma ana ɗaukar cin zarafi a cikin bita daban-daban. Sakamako: A ƙarƙashin "Cleaning" ana iya share siginar na ainihi; tambaya kowane mataki.
m samfuri
Samfura 1 - Lambar nema (tare da bayani):
Role: Kai mai ba da shawara ce mai nazarin bayanan Python.Task: Rubuta lambar da ke lodawa da bincika na'urar CSV.Tsarin: Rukunin: lokaci, saurin (km/h), injin_sic (C), juyin juya hali(rpm) .Taƙaitawa: Yi sharhi kowane jere; Bayyana wane cak aka yi da dalilin da ya sa; ƙara duban ƙimar da ba zai yiwu ba; shiru babu abin da yake sharewa.Fitowa: Lambar da aka yi sharhi + wanne fitarwa yakamata in duba kuma me yasa.
Samfurin 2 - Duban zube:
Matsayi: Kai mai duba lambar koyan inji ne. Aiki: Bincika lambar horo / gwaji mai zuwa don leaks bayanai. Mahimmanci: Wannan jerin lokaci ne (motar tambarin abin hawa). Ƙuntatawa: Gargaɗi idan akwai bazuwar shuffing; Ba da shawara kuma ku ba da hujjar rarrabuwa ta lokaci. Fitowa: Bincike + lambar da aka gyara + bayani.
Samfurin 3 - Tabbatar da raka'a:
Matsayi: Kai mai duba ingancin bayanai ne.Aiki: Ba da shawarar cak ɗin da ke neman rashin daidaituwar naúra a cikin DataFrame.Tsarin: Ƙarfin yana iya zama kWh a wasu layuka da Wh a wasu.Fitowa: Duba lamba + yadda ake nemo ƙirar da ake tuhuma.
Samfura 4 - Kallon gani + sharhi:
Matsayi: Kai kwararre ne na ganin bayanai. Aiki: Rubuta lambar da ke tsara yanayin zafin injin da ƙimar haɓaka. Ƙuntatawa: Lakabi ga gatari tare da naúrar; gani alamar anomaly; kar a yi da'awar takamaiman kuskure lokacin fassara jadawali. Fitowa: Code + abin da ake nema a cikin jadawali.
Rauni mai ƙarfi / Ƙarfi mai ƙarfi
Rawanin faɗakarwa:
Gina samfuri tare da wannan bayanan.
Ba a bayyana wace manufa ba, wane sashi, wane tabbaci; AI na iya fitar da tarkace, leaky, lambar makafi.
Ƙarfi mai ƙarfi:
Matsayi: Kai mai ba da jagoranci ne na Python ML. Aiki: Rubuta aikin farko don tsinkayar lalacewar kushin birki da kuma nuna ramummukan tabbatarwa. Maudu'i: Tsarin lokaci na telemetry; manufa: ragowar kushin kauri.Tuni: Rarraba jerin lokaci ta lokaci (ba shuffling); Siffofin tuta a cikin haɗarin zubewar bayanai; daftarin aiki raka'a;fassara kowane mataki; Rubuta abubuwan da ake buƙata kafin sakamakon ya fita cikin filin. Fitowa: lambar da aka yi sharhi + jerin tantancewa.
Kuskuren gama gari
- Gudun code ba tare da fahimtar shi ba. Lambar AI kuma na iya zama kuskure; Karanta layi ta layi.
- Mixing jerin lokaci. shuffle=Gaskiya tana fitar da gaba kuma tana samar da maki na karya.
- A makance share abin waje. Ana iya share ainihin siginar (fararen kuskure).
- Ba daftarin aikin naúrar ba. Kurakurai kamar km/h vs m/s, Wh vs kWh girma shiru.
- Tsallake bayanin ()/duba mataki. Yawancin kurakurai suna bayyana a ƙididdiga ta farko.
A takaice
- Python (pandas, numpy, matplotlib, scikit-learn) shine ainihin ma'aunin binciken bayanan mota.
- Binciken mai maimaitawa: kaya, dubawa, tsabta, fasali, nazari, ingantawa da rahoto.
- Yana da mahimmanci don fahimta da tabbatar da lambar da AI ke samar da layi ta layi; Domin yana aiki ba yana nufin yana da kyau ba.
- Kula da bambance-bambancen da suka gabata / na gaba a cikin jerin lokaci; Juyawa bazuwar yana haifar da ɗigon bayanai.
- Daidaiton raka'a da fassarar waje shiru amma mahimman wuraren tabbatarwa.
Aikin aikace-aikace
Ɗauki ƙaramin saitin bayanai (na gaske ko na zamani na zamani). (1) Bi tsarin matakai shida, samar da lambar lodi da sarrafawa tare da Samfura 1 kuma tabbatar da cewa kun fahimci kowane layi. (2) Nemo aƙalla ƙima mai tuhuma ɗaya a cikin bayanin () fitarwa kuma bincika dalilin. (3) A cikin aikin tsinkaya, lokacin da horo/gwajin ya rabu kuma duba haɗarin ɗigowa tare da Samfura 2. (4) Rubuta sashin kowane shafi da kuke amfani da shi a cikin ƙamus.
jerin abubuwan dubawa
- [ ] Na kafa bincike tare da tsarin matakai shida.
- [ ] Na karanta kuma na fahimci lambar da AI line ta samar.
- [ ] Na nemi abubuwan da ake tuhuma tare da siffanta()/audit.
- [ ] Na raba jerin lokaci da lokaci (babu shuffling).
- [ ] Na yiwa alama alamomin da ke cikin haɗarin zubewar bayanai.
- [ ] Na rubuta sashin kowane shafi kuma na rubuta jujjuyawar a bayyane.