Me pēhea te hanga i tētahi AI i runga i tō rorohiko

Me pēhea te hanga i tētahi AI i runga i tō Rorohiko. Aratohu Katoa.

Āe, kei te hiahia koe ki te hanga i tētahi "AI". Ehara i te mea i roto i te tikanga o Hollywood e whakaaro ana ki te oranga, engari ko te momo ka taea e koe te whakahaere i runga i tō pona e matapae ana, e whakarite ana i ngā mea, e kōrerorero ana rānei. Ko tēnei aratohu mō te hanga i tētahi AI i runga i tō Rorohiko he ngana ki te kukume i a koe mai i te kore ki tētahi mea e mahi ana i te rohe . Me tumanako ki ngā pokatata, ki ngā whakaaro tika, me te ara peka i ētahi wā nā te mea, kia pono tātou, kāore te mahi whakamāori i te mea ma.

Ngā tuhinga ka pai pea koe ki te pānui i muri i tēnei:

🔗 Me pēhea te hanga i tētahi tauira AI: ngā taahiraa katoa kua whakamāramahia
He whakamārama mārama mō te hanganga tauira AI mai i te tīmatanga ki te mutunga.

🔗 He aha te AI tohu: nga mea katoa e hiahia ana koe ki te mohio
Akohia ngā kaupapa matua o te AI tohu, te hītori, me ngā tono o ēnei rā.

🔗 Ngā whakaritenga rokiroki raraunga mō te AI: ngā mea e hiahiatia ana e koe
Kia mārama ki ngā hiahia rokiroki mō ngā pūnaha AI whai hua, tauine hoki.

He aha koe ka whakararuraru ai ināianei? 🧭

Nā te mea kua kore te wā o "ngā taiwhanga tauine-Google anake ka taea te mahi AI". I ēnei rā, me te pona noa, ētahi taputapu tuwhera-pūtake, me te uaua, ka taea e koe te hanga tauira iti hei whakarōpū i ngā īmēra, hei whakarāpopoto i ngā tuhinga, hei tāpiri rānei i ngā whakaahua. Kāore he pokapū raraunga e hiahiatia ana. Ko tāu e hiahia ana ko:

  • he mahere,

  • he whakaritenga ma,

  • me tētahi whāinga ka taea e koe te whakaoti me te kore e hiahia ki te maka i te mīhini ki waho o te matapihi.

He aha te mea e tika ai te whai i tēnei ✅

Ko te hunga e pātai ana "Me pēhea te hanga i tētahi AI i runga i tō Rorohiko" kāore e hiahia ana ki te whiwhi tohu PhD. E hiahia ana rātou ki tētahi mea ka taea e rātou te whakahaere. He mahere pai e whakaoti ana i ētahi mea:

  • Tīmata mā te iti : whakarōpū i ngā kare ā-roto, kaua e "whakaoti i te mātauranga".

  • Te āheinga ki te whakaputa anō : conda venv rānei kia taea ai e koe te hanga anō āpōpō me te kore e mataku.

  • Te pono o ngā taputapu : he pai ngā CPU mō te ako-scikit, he pai ngā GPU mō ngā kupenga hōhonu (mēnā he waimarie koe) [2][3].

  • Raraunga ma : kāore he paraurehe kua he te tapanga; wehea tonutia kia tereina/whai mana/whakamātautau.

  • Ngā inenga e whai tikanga ana : te tika, te pūmau, te maumahara, F1. Mō te kore taurite, ROC-AUC/PR-AUC [1].

  • He huarahi hei tiritiri : he API iti, he CLI, he taupānga whakamātautau rānei.

  • Haumaru : kāore he huinga raraunga huna, kāore he turuturu mōhiohio tūmataiti, kia mārama te aro ki ngā mōrearea [4].

Kia tika ēnā, ā, ka tūturu tō tauira "iti".

He mahere ara kāore e whakamataku te āhua 🗺️

  1. Kōwhiria he raruraru iti + kotahi te inenga.

  2. Tāutahia a Python me ētahi whare pukapuka matua.

  3. Waihangatia he taiao ma (ka mihi koe ki a koe anō ā muri ake nei).

  4. Utaina tō huinga raraunga, wehea kia tika.

  5. Whakangunguhia he turanga kuware engari pono.

  6. Whakamātauria te kupenga io mēnā he hua anō kei roto.

  7. Tākaihia he whakaaturanga.

  8. Puritia ētahi tuhipoka, ā muri ake nei - ka mihi atu koe ki a koe.

Kete iti rawa: kaua e whakararuraru rawa 🧰

  • Python : tiki mai i python.org.

  • Taiao : Conda, venv me te pip.

  • Pukatuhi : He Jupyter mō te tākaro.

  • Etita : VS Code, he hoahoa, he kaha hoki.

  • Ngā pukapuka matua

    • panda + NumPy (whakawhitiwhiti raraunga)

    • scikit-ako (ML matarohia)

    • PyTorch, TensorFlow rānei (te ako hōhonu, te hanga GPU i te mea nui) [2][3]

    • Ngā Whakawhiti Kanohi Awhi, spaCy, OpenCV (NLP + tirohanga)

  • Whakaterenga (whiringa)

    • NVIDIA → Ngā hanganga CUDA [2]

    • AMD → Hangaia e ROCm [2]

    • Apple → PyTorch me te tuara whakarewa (MPS) [2]

⚡ Tuhipoka taha: ka ngaro te nuinga o te "mamae tāutanga" mēnā ka tukuna noa e koe te tika mō tō tāutanga e ngā kaiwhakauru mana. Tāruatia, whakapiritia, kua oti [2][3].

Ture matua: ngoki mā te CPU i te tuatahi, kātahi ka oma tere me te GPU i muri mai.

Te whiriwhiri i tō puranga: ātete ki ngā mea kanapa 🧪

  • Raraunga ripanga → scikit-learn. Tauritenga arorau, ngahere matapōkere, whakanui rōnaki.

  • Kuputuhi, whakaahua rānei → PyTorch, TensorFlow rānei. Mō te tuhinga, he wikitoria nui te whakatikatika i tētahi Transformer iti.

  • He rite ki te karetao kōrerorero → llama.cpp te whakahaere i ngā LLM iti i runga i ngā pona. Kaua e tumanako ki te makutu, engari ka mahi mō ngā tuhipoka me ngā whakarāpopototanga [5].

Te whakarite i te taiao ma 🧼

# Conda way conda create -n localai python=3.11 conda activate localai # OR venv python -m venv .venv puna .venv/bin/activate # Windows: .venv\Scripts\activate

Kātahi ka tāuta i ngā mea nui:

pip tāuta numpy pandas scikit-ako jupyter pip tāuta rama torchvision torchaudio # rānei tensorflow pip tāuta transformers huinga raraunga

(Mō ngā hanganga GPU, me whakamahi noa i te kōwhiringa mana [2][3].)

Tauira mahi tuatahi: kia iti noa iho 🏁

Tūāpapa tuatahi. CSV → ngā āhuatanga + tapanga → te whakatauiratanga logistic.

mai i te sklearn.linear_model kawemai LogisticRegression ... tā ("Te Tika:", kaute_tika(y_test, preds)) tā(pūrongo_whakarōpū(y_test, preds))

Ki te eke tēnei ki te taumata matapōkere, ka whakanui koe. Kawhe, pihikete rānei, nāu te whakatau ☕.
Mō ngā akomanga kāore i te taurite, mātakitaki i ngā pihi tika/whakamahara + ROC/PR kaua ki te tika mata [1].

Ngā kupenga io (mēnā he āwhina anake) 🧠

He tuhinga kei a koe, ā, e hiahia ana koe ki te whakarōpū i ngā kare ā-roto? Whakatikatikahia he Transformer iti kua oti te whakarite. Tere, ma, kāore e para i tō mīhini.

mai i ngā transformers kawemai AutoModelForSequenceClassification ... trainer.train() print(trainer.evaluate())

Tohutohu ngaio: tīmata ki ngā tauira iti. Mā te patuiro hapa i runga i te 1% o ngā raraunga ka penapenahia he hāora.

Raraunga: ngā kaupapa matua e kore e taea te peke atu 📦

  • Ngā huinga raraunga tūmatanui: Kaggle, Hugging Face, repo mātauranga (tirohia ngā raihana).

  • Ngā Tikanga Matatika: te horoi i ngā mōhiohio whaiaro, te whakaute i ngā mana.

  • Wehewehenga: whakangungu, whakamana, whakamātautau. Kaua rawa e titiro whakamuri.

  • Ngā Tapanga: he mea nui ake te ōritetanga i ngā tauira papai.

Te pōma pono: 60% o ngā hua i ahu mai i ngā tapanga ma, ehara i te makutu hoahoanga.

Ngā inenga hei pupuri i a koe kia pono 🎯

  • Whakarōpūtanga → tika, pūmau, maumahara, F1.

  • Ngā huinga taurite kore → He mea nui ake a ROC-AUC, PR-AUC.

  • Whakamuri → MAE, RMSE, R².

  • Tirotiro i te pono → tirohia ētahi putanga; tera pea ka teka ngā tau.

Tohutoro whai hua: aratohu inenga scikit-learn [1].

Ngā tohutohu whakatere 🚀

  • NVIDIA → Hanganga CUDA PyTorch [2]

  • AMD → ROCm [2]

  • Apple → Papamuri MPS [2]

  • TensorFlow → whai i te tāutanga GPU mana + manatoko [3]

Kaua e arotau i mua i te oma o tō turanga. He rite tēnā ki te whakakanapa i ngā awhi i mua i te whai wira o te motuka.

Ngā tauira whakaputa uri ā-rohe: ngā tarakona pēpi 🐉

  • Reo → ngā LLM kua inehia mā te llama.cpp [5]. He pai mō ngā tuhipoka, ngā tohu waehere rānei, ehara i te kōrero hōhonu.

  • Ngā Whakaahua → Kei reira ngā momo horapa pumau; pānuihia ngā raihana me te tupato.

I ētahi wā ka pai ake te mahi a te Transformer kua whakatikatikahia kia rite ki tētahi LLM pupuhi i runga i ngā taputapu iti.

Ngā whakaaturanga mōkihi: tukua te iwi kia pāwhiri 🖥️

  • Gradio → te atanga kaiwhakamahi māmā rawa atu.

  • FastAPI → API ma.

  • Ipu → ngā tuhinga tere.

kawemai gradio hei gr clf = pipeline("tātari-hinengaro") ... demo.launch()

He rite ki te makutu ina whakaatuhia e tō pūtirotiro.

Ngā tikanga hei whakaora i te hinengaro pai 🧠

  • Git mō te whakahaere putanga.

  • MLflow, ngā pukatuhi rānei mō te aroturuki i ngā whakamātautau.

  • Te whakaputanga raraunga mā te DVC, mā te whakamahi rānei i ngā hash.

  • Docker mēnā me whakahaere e ētahi atu āu mea.

  • Ngā whakawhirinakitanga pine ( requirements.txt ).

Whakapono mai, ā muri ake nei - ka nui tōu whakawhetai.

Te rapurongoā: ngā wā "āe" noa 🧯

  • Ngā hapa tāuta? Mukua noa te taiao ka hanga anō.

  • Kāore i kitea te GPU? Kāore i te ōrite te taraiwa, tirohia ngā putanga [2][3].

  • Kāore te tauira e ako? Whakaitihia te tere ako, whakangawaritia, horoia rānei ngā tapanga.

  • He nui rawa te whakauru? Whakahāngaitia, whakataka atu rānei, tāpirihia noa iho rānei ngā raraunga.

  • He pai rawa ngā inenga? I turuturu mai te huinga whakamātautau (he maha atu ngā wā ka tupu tēnei i tō whakaaro).

Haumarutanga + kawenga 🛡️

  • Rīpene PII.

  • Whakautehia ngā raihana.

  • Ā-rohe-tuatahi = tūmataitinga + mana whakahaere, engari me ngā herenga rorohiko.

  • Tuhia ngā mōrearea (te tika, te haumaru, te manawanui, me ētahi atu) [4].

He ripanga whakatairite whai hua 📊

Utauta Pai rawa atu mō He aha i whakamahia ai?
ako-scikit Raraunga ripanga Wikitōria tere, API ma 🙂
PyTorch Ngā kupenga hōhonu ritenga Hapori nui, ngāwari
TensorFlow Ngā paipa whakaputa Pūnaha rauropi + ngā kōwhiringa ratonga
Ngā Whakawhiti Ngā mahi tuhinga Ka tiakina te rorohiko e ngā tauira kua whakangungua i mua
SpaCy Ngā paipa NLP Kaha ā-ahumahi, whai hua
Gradio Ngā Whakaaturanga/Ngā UI 1 kōnae → UI
FastAPI Ngā API Ngā tuhinga tere + aunoa
Wā Whakahaere ONNX Te whakamahinga whakawhiti-anga Kawe + whai hua
rama.cpp Ngā LLM iti o te rohe Te ine inenga e pai ana ki te CPU [5]
Docker Te tiritiri i ngā taiao "E mahi ana i ngā wāhi katoa"

E toru ngā ruku hohonu (ka whakamahia e koe) 🏊

  1. Te hangarau āhuatanga mō ngā ripanga → te whakataurite, te whakamahana-kotahi, te whakamātau i ngā tauira rākau, te whakamana-whakawhiti [1].

  2. Whakawhiti ako mō te tuhinga → whakatikatika i ngā Transformers iti, kia iti te roa o te raupapa, F1 mō ngā akomanga onge [1].

  3. Te arotautanga mō te whakatau ā-rohe → te ine i te rahi, te kaweake i te ONNX, ngā tohu keteroki.

Ngā mahanga tawhito 🪤

  • He nui rawa te hanga, he wawe rawa.

  • Te kore aro ki te kounga o ngā raraunga.

  • Te peke i te wehenga whakamātautau.

  • Tārua-whakapiri waehere matapo.

  • Kāore e tuhi ana i tetahi mea.

Ahakoa he README ka penapena i ngā haora i muri mai.

Ngā rauemi ako e tika ana te wā 📚

  • Ngā tuhinga whaimana (PyTorch, TensorFlow, scikit-learn, Transformers).

  • Akoranga Whakateretere mō te Google ML, DeepLearning.AI.

  • Ngā tuhinga OpenCV mō ngā kaupapa matua o te tirohanga.

  • He aratohu whakamahi spaCy mō ngā paipa NLP.

He mahi iti noa iho: ko ngā kaiwhakauru mana e whakaputa ana i tō whakahau tāuta GPU he mea whakaora i ngā oranga [2][3].

Te whakakotahi i ngā mea katoa 🧩

  1. Whāinga → whakarōpūtia ngā tīkiti tautoko kia toru ngā momo.

  2. Raraunga → Kaweake CSV, hunaia, wehea.

  3. Pūtake → scikit-ako TF-IDF + tauhohenga logistic.

  4. Whakapai ake → Whakatikatika i te mīhini whakawhiti mēnā ka tū te mahi taketake.

  5. Whakaaturanga → Taupānga pouaka tuhinga Gradio.

  6. Tuku → Docker + README.

  7. Whakahou → whakatika hapa, tapanga anō, tāruarua.

  8. Tiaki → mōrearea tuhinga [4].

He tino whai hua.

TL;DR 🎂

Te ako i te hanga i tētahi AI i runga i tō Rorohiko = whiriwhiria tētahi raruraru iti, hangaia he pūtake, kia piki ake te mahi ina āwhina, ā, kia taea tonu te tārua i tō tatūnga. Me mahi kia rua, ā, ka mōhio koe he pūkenga koe. Me mahi kia rima ngā wā, ā, ka tīmata te tangata ki te tono āwhina mai i a koe, koinei te wāhanga ngahau.

Āe, i ētahi wā he rite ki te whakaako i te tōhi ki te tito rotarota. Kāore he aha. Me mahi tonu. 🔌📝

Ngā Tohutoro

[1] scikit-learn — Ngā Inenga me te Aromatawai Tauira: hononga
[2] PyTorch — Kōwhiringa Tāuta Ā-Rohe (CUDA/ROCm/Mac MPS): hononga
[3] TensorFlow — Tāuta + Manatoko GPU: hononga
[4] NIST — Anga Whakahaere Mōrearea AI: hononga
[5] llama.cpp — Repo LLM Ā-Rohe: hononga

Kimihia te AI Hou i te Toa Kaiāwhina AI Whaimana

Mō Mātou

Hoki ki te rangitaki