Интересное что-то – Telegram
Интересное что-то
517 subscribers
2.72K photos
253 videos
138 files
4.51K links
Материалы и мысли, понадерганные отовсюду
Блог: https://news.1rj.ru/str/asisakov_channel
Чат: https://news.1rj.ru/str/youknowds_chat
Download Telegram
Forwarded from Anton Eryomin
You should really learn to DEPLOY your Machine Learning models! Focus less on the algorithms and understand the products you are building. A junior data scientist tends to think about algorithms and model performance. A senior data scientists will think more efficient deployment pipelines, products, user experience and business metrics.

There are many ways to deploy ML models, but the most common one is as a ML microservice: a backend application would communicate to your ML service through http API calls (usually REST or RPC). There are many online tutorials to get you started. The first entries when searching on google:
- https://lnkd.in/gtUjMqaK
- https://lnkd.in/grPjpqz7
- https://lnkd.in/g4eGEyFr
- …

For example, you will be able to get a better feel on the latency of your model inference, you will be able to participate to system design conversations and you will be able to think about edge cases like “what happens when the data payload is empty?”, “what happens to the overall system when the ML server crashes?”. When trying to convince stakeholders, I always like to present a “product” instead of a “model”. A “model” or performance metrics are not really convincing to non-technical people, but show them in action what ML can do and they get impressed. I often like to use Dash (https://plotly.com/dash/) to prototype a quick UI to show the power of ML.

After learning to deploy API endpoints, learn about dockerizing your applications, about orchestrating your applications and push them to the cloud! After that learning this, you will be a long way in being a more senior data scientist.


Нашел пост чувака с ФБ в продолжении того, о чем писалось выше. Что мол МЛ это уже далеко не только какие-то метрики и модельки, это в первую очередь задача бизнеса и то, как ты свои модельки доведёшь до продакшена.
Forwarded from DL in NLP (Vlad Lialin)
🤗 запускает курс по RL
Syllabus: https://github.com/huggingface/deep-rl-class
Регистрация: тык

Обещают научить работать со stable baselines, RLlib, RL Baselines3 Zoo. Также будут не только заезженные Space Invaders но и новые environments, включая работающие на Unity.

Кроме классических топиков (Q learning, policy gradients, PPO) будут также offline RL и decision transformers.

Заучит классно, мне давно пора подтянуть свои RL скилы.
Forwarded from Борис опять
P.S. Яндекс рисерч не так давно выпустил статью про свойства метрик классификации.

Tldr: лучше использовать symmetric balanced accuracy чем f1 score. Но еще лучше почитать статью.

https://arxiv.org/abs/2201.09044
#python
Интересные туториалы по питону и библиотекам: pythonguides.com
Forwarded from DL in NLP (Vlad Lialin)
Стенфорд выложил все видосы cs224n Winter 2021 Natural Language Processing. 🔥🔥
Это один из лучших курсов по nlp в мире, и теперь доступна его более свежая версия. Есть нормальная лекция по трансформерам, T5, low resource MT.
Всем смотреть.

https://youtube.com/playlist?list=PLoROMvodv4rOSH4v6133s9LFPRHjEmbmJ
#статьи
Подборка самых популярных статей по банковской тематике за последние 7 лет (указано число ссылок в гугл-академии).

545 ссылок, 2021 год
Dwivedi Y. K. et al. Artificial Intelligence (AI): Multidisciplinary perspectives on emerging challenges, opportunities, and agenda for research, practice and policy //International Journal of Information Management. – 2021. – Т. 57. – С. 101994.

434 ссылки, 2017 год
Xia Y. et al. A boosted decision tree approach using Bayesian hyper-parameter optimization for credit scoring //Expert Systems with Applications. – 2017. – Т. 78. – С. 225-241.

235 ссылок, 2018 год
Alessi L., Detken C. Identifying excessive credit growth and leverage // Journal of Financial Stability. – 2018. – Т. 35. – С. 215-225.

206 ссылок, 2015 год
Iturriaga F. J. L., Sanz I. P. Bankruptcy visualization and prediction using neural networks: A study of US commercial banks // Expert Systems with applications. – 2015. – Т. 42. – №. 6. – С. 2857-2869.

183 ссылки, 2017 год
Abellán J., Castellano J. G. A comparative study on base classifiers in ensemble methods for credit scoring // Expert systems with applications. – 2017. – Т. 73. – С. 1-10.

153 ссылки, 2019 год
Kou G. et al. Machine learning methods for systemic risk analysis in financial sectors // Technological and Economic Development of Economy. – 2019. – Т. 25. – №. 5. – С. 716-742.

136 ссылок, 2017
Chakraborty C., Joseph A. Machine learning at central banks. – 2017.
#ml #systemdesign #interview
Список тем для подготовки к ML System Design