دانلود کتاب یادگیری ماشین عملی با PyTorch و Scikit-Learn

In 2006, Ge of frey Hin ton et al. pub lished a pa per 1 show ing how to train a deep neu ral net work ca pa ble of rec og niz ing hand writ ten dig its with state-of-the-art pre ci sion (>98%). They branded this tech nique “deep learn ing”. A deep neu ral net work is a (very) sim pli fied model of our cere bral cor tex, com posed of a stack of lay ers of ar ti fi cial neu rons. Train ing a deep neu ral net was widely con sid ered im pos si ble at the time, 2 and most re searchers had aban doned the idea in the late 1990s. This pa per re vived the in ter est of the sci en tific com mu nity, and be fore long many new pa pers demon strated that deep learn ing was not only pos si ble, but ca ‐ pa ble of mind-blow ing achieve ments that no other ma chine learn ing (ML) tech nique could hope to match (with the help of tremen dous com put ing power and great amounts of data). This en thu si asm soon ex tended to many other ar eas of ma chine learn ing. A decade later, ma chine learn ing had al ready con quered many in dus tries, rank ing web re sults, rec om mend ‐ ing videos to watch and prod ucts to buy, sort ing items on pro duc tion lines, some times even driv ing cars. Ma ‐ chine learn ing of ten made the head lines, for ex am ple when Deep Mind’s Al phaFold ma chine learn ing sys tem solved a long-stand ing pro tein-fold ing prob lem that had stomped re searchers for decades. But most of the time, ma chine learn ing was just work ing dis cretely in the back ground. How ever, an other decade later came the rise of AI as sis tants: from Chat GPT in 2022, Gem ini, Claude, and Grok in 2023, and many oth ers since then. AI has now truly taken off and it is rapidly trans form ing ev ery sin gle in dus try: what used to be sci-fi is now very real. This book as sumes that you know close to noth ing about ma chine learn ing. Its goal is to give you the con ‐ cepts, tools, and in tu ition you need to im ple ment pro grams ca pa ble of learn ing from data. We will cover a large num ber of tech niques, from the sim plest and most com monly used (such as lin ear re ‐ gres sion) to some of the deep learn ing tech niques that reg u larly win com pe ti tions. For this, we will be us ing Python—the lead ing lan guage for data sci ence and ma chine learn ing—as well as open source and pro duc ‐ tion-ready Python frame works: – Scikit-Learn is very easy to use, yet it im ple ments many ma chine learn ing al go rithms ef fi ciently, so it makes for a great en try point to learn ing ma chine learn ing. It was cre ated by David Cour na peau in 2007, then led by a team of re searchers at the French In sti tute for Re search in Com puter Sci ence and Au to ma tion (In ria), and re cently Prob abl.ai. – Py Torch is a pow er ful and flex i ble li brary for deep learn ing. It makes it pos si ble to train and run all sorts of neu ral net works ef fi ciently, and it can dis trib ute the com pu ta tions across mul ti ple GPUs (graph ics pro cess ing units). Py Torch (PT) was de vel oped by Face book’s AI Re search lab (FAIR) and first re leased in 2016. It evolved from Torch, an older frame work coded in Lua. In 2022, Py ‐ Torch was tran si tioned to the Py Torch Foun da tion, un der the Linux Foun da tion, to pro mote com mu ‐ nity-driven de vel op ment.