|
|
|
|
|
|
|
¸ñÂ÷ |
|
PART01 Machine Learning ÀÔ¹®
Chatper 01 ¸Ó½Å·¯´×ÀÇ ±âº» °³³ä
1.1 éÅÍ ¼³¸í ¹× éÅÍ È°¿ë¹ý
1.2 ¸Ó½Å·¯´× ±âÃÊ
1.2.1 ¼±Çü ȸ±Í(Linear Regression)
1.2.2 Â÷¿øÀÇ È®Àå(Multi variable linear regression)
1.2.3 ·ÎÁö½ºÆ½ ȸ±Í(Logistic Regression)
1.2.4 ¼ÒÇÁÆ®¸Æ½º ȸ±Í(Softmax Regression)
1.3 ±âŸ ¾Ë¾ÆµÎ¸é ÁÁÀº °³³ä ¹× ÆÁ
1.3.1 ÇнÀ·ü(learning rate)
1.3.2 ¹èÄ¡ Á¤±ÔÈ(batch normalization)
1.3.3 °úÀûÇÕ(overfitting)
1.3.4 µö·¯´×¿¡ ´ëÇؼ
Chatper 02 µö·¯´×À» ÀÌ¿ëÇÑ À̹ÌÁö ºÐ¼® ½Ç½À
2.1 éÅÍ ¼³¸í ¹× ½Ç½À overview
2.2 °³¹ß ȯ°æ ¼¼ÆÃ
2.2.1 ±¸±Û ÄÚ·¦(Google colaboratory) ¼³¸í
2.2.2 ÄÚ·¦ ¼³Ä¡
2.2.3 ÄÚ·¦ ȯ°æ¼³Á¤
2.2.4 ÆÄÀÌ½ã ¹× ÄÉ¶ó½º ¼³Ä¡
2.3 µ¥ÀÌÅͼ Áغñ ¹× CNN ¸ðµ¨ ±¸Ãà
2.3.1 ±¸±Û µå¶óÀÌºê ¸¶¿îÆ®
2.3.2 ÇнÀ µ¥ÀÌÅͼ Áغñ ¹× À̹ÌÁö Àüó¸®
2.3.3 CNN ¸ðµ¨ ±¸Ãà
2.3.4 µ¥ÀÌÅͼ ÇнÀ
2.4 ÀüÀÌÇнÀ(transfer learning)
2.4.1 ÀüÀÌÇнÀÀÇ °³³ä°ú ¸ðµ¨ Àû¿ë
2.4.2 ÀüÀÌÇнÀ ÄÚµå Àû¿ë
PART02 Äí¹ö³×Ƽ½ºÀÇ ¸Ó½Å·¯´× ÅøŶ! Kubeflow!
Chatper 01 kubeflow
1.1 ML ¿öÅ©Ç÷οì
1.1.1 ML ¿öÅ©Ç÷οì¶õ
1.1.2 ¸ðµ¨ ½ÇÇè ´Ü°è
1.1.3 ¸ðµ¨ »ý»ê ´Ü°è
1.1.4 ML ¿öÅ©Ç÷οì Åø
1.2 kubeflow
1.2.1 kubeflow
1.2.2 kubeflow components on ML workflow
1.2.3 Äíº£Ç÷οì À¯Àú ÀÎÅÍÆäÀ̽º(UI)
1.2.4 API ¿Í SDK
1.2.5 Äíº£Ç÷οì ÄÄÆ÷³ÍÆ®µé
1.2.6 Äíº£ÇÃ·Î¿ì ¹öÁ¯ Á¤Ã¥
1.3 kubernetes
1.3.0 ¼¹®
1.3.1 ÄÁÅ×ÀÌ³Ê °³¹ß ½Ã´ë
1.3.2 Äí¹ö³×Ƽ½º¶õ
1.3.3 Äí¹ö³×Ƽ½º ±¸Á¶
1.3.4 ¿ÀºêÁ§Æ®¿Í ÄÁÆ®·Ñ·¯
1.3.5 ¿ÀºêÁ§Æ® ÅÛÇø´
1.3.6 ·¹À̺í°ú ¼¿·ºÅÍ, ¾î³ëÅ×À̼Ç
1.3.7 Àα׷¹½º
1.3.8 ÄÁÇÇ±× ¸Ê
1.3.9 ½ÃÅ©¸´
1.3.10 ÀÎÁõ°ú ±ÇÇÑ
1.4 Äíº£ÇÃ·Î¿ì ¼³Ä¡
1.4.1 ¼³Ä¡ Á¶°Ç
1.4.2 Äí¹ö³×Ƽ½º ¼³Ä¡
1.4.3 ÇÁ¶óÀ̺ø µµÄ¿ ·¹Áö½ºÆ®¸®
1.4.4 k9s
1.4.5 kfctl
1.4.6 ¹èÆ÷ Ç÷§Æû
1.4.7 ½ºÅÄ´Ùµå ÄíºêÇÃ·Î¿ì ¼³Ä¡
1.4.8 DEX¹öÀü ¼³Ä¡
1.4.9 ÇÁ·ÎÆÄÀÏ
1.4.10 »èÁ¦
Chatper 02 Kubeflow Components
2.0 ¼·Ð
2.1 Dashboard
2.1.1 °³¿ä
2.1.2 ·ÎÄÿ¡¼ ´ë½¬º¸µå Á¢¼ÓÇϱâ
2.2 Notebook servers
2.2.1 °³¿ä
2.2.2 ³ëÆ®ºÏ »ý¼ºÇϱâ
2.2.3 Äí¹ö³×Ƽ½º ¸®¼Ò½º È®ÀÎÇϱâ
2.2.4 Ä¿½ºÅÒ À̹ÌÁö »ý¼º
2.2.5 TroubleShooting
2.3 Fairing
2.3.1 ¼Ò°³
2.3.2 ¾ÆÅ°ÅØó
2.3.3 Æä¾î¸µ ¼³Ä¡
2.3.4 Æä¾î¸µ ¼³Á¤
2.3.5 fairing.config
2.3.6 Preprocessor
2.3.7 Builder
2.3.8 Deployer
2.3.9 Config.run
2.3.10 Config.fn
2.3.11 fairing.ml_tasks
2.4 Katib
2.4.1 ¼Ò°³
2.4.2 ÇÏÀÌÆÛÆĶó¹ÌÅÍ¿Í ÇÏÀÌÆÛ¶ó¹ÌÅÍ ÃÖÀûÈ
2.4.3 ´º·² ¾ÆÅ°ÅØó Ž»ö
2.4.4 ¾ÆÅ°ÅØó
2.4.5 Experiment
2.4.6 °Ë»ö ¾Ë°í¸®Áò
2.4.7 Metric collector
2.4.8 Component
2.4.9 īƼºê Web UI
2.4.10 Rest API
2.4.11 Command-line interfaces
2.4.12 īƼºê ´Üµ¶ ¼³Ä¡
2.5 Pipeline
2.5.1 ¼Ò°³
2.5.2 ÆÄÀÌÇÁ¶óÀÎ
2.5.3 ¾ÆÅ°ÅÃÃÄ
2.5.3 ÄÄÆ÷³ÍÆ®
2.5.4 ±×·¡ÇÁ(Graph)
2.5.5 ·±(Run), ¸®Ä¿¸µ ·±(Recurring Run)
2.5.6 ·± Æ®¸®°Å(Run Trigger)
2.5.7 ½ºÅÜ(Step)
2.5.8 Experiment
2.5.9 Output Artifact
2.5.10 ÆÄÀÌÇÁ¶óÀÎ ÀÎÅÍÆäÀ̽º
2.5.11 ÆÄÀÌÇÁ¶óÀÎ ´Üµ¶ ¼³Ä¡
2.5.12 ÆÄÀÌÇÁ¶óÀÎ SDK ¼³Ä¡
2.5.13 ÆÄÀÌÇÁ¶óÀÎSDK ÆÐÅ°Áö µÑ·¯º¸±â
2.5.14 SDK·Î ÆÄÀÌÇÁ¶óÀÎ ¸¸µé±â
2.5.15 °æ¶û ÆÄÀ̼± ÄÄÆ÷³ÍÆ®
2.5.16 ÆĶó¹ÌÅÍ(PipelineParam)
2.5.17 ¸ÞÆ®¸¯½º(Matrix)
2.5.18 Äí¹ö³×Ƽ½º ¸®¼Ò½º ÄÄÆ÷³ÍÆ®
2.6 Training of ML models
2.6.1 TFJob
2.6.2 PyTorchJob
2.6.3 MXJob(MXNet)
2.6.4 MPIJob
2.6.5 ChainerJob
2.7 Serving Models
2.7.1 °³¿ä
2.7.2 KFServing
2.7.3 InferenceService
2.7.4 Seldon Serving
2.8 Metadata
2.8.1 °³¿ä
2.8.2 ¼³Ä¡
2.8.3 SDK
2.8.4 Metadata Web UI
2.8.5 Watcher
Chatper 03 ÇÚÁî¿Â Äíº£Ç÷οì
3.1 Traning Mnist with Fairing
3.1.1 Notebook provisioning
3.1.2 fashion mnist ½ÇÇà
3.1.3 fashion Mnist¸¦ Fairing jobÀ¸·Î ¹Ù²Ù±â
3.1.4 Job ½ÇÇàÇغ¸±â
3.1.5 ÀÌÁ¦ ÀâÀº ±×¸¸ ´øÁ®µµ µÉ²¨ °°Àºµ¥
3.2 īƼºê·Î ÇÏÀÌÆÛÆĶó¹ÌÅÍ ÃÖÀûÈÇϱâ
3.2.1 fashion Mnist¸¦ katib jobÀ¸·Î ´øÁú ¼ö ÀÖ°Ô º¯ÇüÇϱâ
3.2.2 īƼºê experiment CRD »ý¼ºÇϱâ
3.2.3 jupyter notebook¿¡¼ katib job ½ÇÇàÇϱâ
3.2.4 īƼºê Trial ±×·¡ÇÁ ºÐ¼®Çϱâ
3.3 Ãß·Ð ¸ðµ¨ ¼¹ö ¸¸µé¾î º¸±â
3.3.1 ¸ðµ¨ ÁغñÇϱâ
3.3.2 KFServingÀ» ÀÌ¿ëÇÑ Ãß·Ð ¸ðµ¨ ¼¹ö ±¸¼º
3.3.3 Ãß·Ð ¸ðµ¨ Å×½ºÆ®
3.4 ÆÄÀÌÇÁ¶óÀÎÀ¸·Î ML¿öÅ©ÇÃ·Î¿ì ¸¸µé±â
3.4.1 ÆÄÀÌÇÁ¶óÀο¡ ºÒ·ý ºÙ¿©º¸±â
3.4.2 ¸®Ä¿¸µ ·±(Recurring Run)À¸·Î ½ºÅ丮Áö¿¡ °è¼Ó µ¥ÀÌÅ͸¦ ½×¾Æº¸±â
3.4.3 ÇнÀºÎÅÍ ¼ºù±îÁö ÆÄÀÌÇÁ¶óÀÎÀ¸·Î
3.5 Caltech101 ÃÖÀûÈ
3.5.0 °³¿ä
3.5.1 ÀÏ´Ü Æä¾î¸µ
3.5.2 īƼºê¸¦ À§ÇÑ ¸ÞÆ®¸¯¼³Á¤
3.5.3 īƼºê Submit!
3.5.4 Trial ±×·¡ÇÁ ºÐ¼®Çϱâ
3.5.5 ³ëÆ®ºÏ¿¡¼ īƼºê Experiment ½ÇÇàÇϱâ
3.5.6 Experiment ½ÇÇàÀ» Æä¾î¸µÀ¸·Î °¨½Î±â
3.5.7 ÆÄÀÌÇÁ¶óÀο¡¼ Experiment ½ÇÇàÇغ¸±â
3.5.8 īƼºê °á°ú Á¶È¸Çϱâ |
|
|
|
ÀúÀÚ
|
|
À̸íȯ
µÎ´ÙÁö ¿£Áö´Ï¾î, kb@dudaji.com, J2ME/WIPI¸¦ ½ÃÀÛÀ¸·Î °³¹ßÀ» ½ÃÀÛÇß°í, ÀÚ¹Ù·Î ÇÒ ¼ö ÀÖ´Â ´ëºÎºÐÀÇ ÇÁ·ÎÁ§Æ®¸¦ ´Ù¼ö ¼öÇàÇß´Ù. ÃÖ±Ù µÎ´ÙÁö¿¡¼ ÇൿÀÎ½Ä °ü·Ã ÇÁ·ÎÁ§Æ® µîÀ» ÁøÇàÇÏ¸é¼ Äí¹ö³×Ƽ½º¿Í ÄíºêÇ÷ο쿡 ÀÔ¹®ÇÏ¿´°í ¿©±âÀú±â ÇÚÁî¿ÂÀ» ÅëÇØ ÄíºêÇ÷ο츦 È®»ê½ÃÅ°·Á ³ë·ÂÇÏ°í ÀÖ´Ù. Àΰ£°ú µ¿¹°µéÀÌ Áñ°Ì°Ô °ø»ýÇÒ ¼ö ÀÖ´Â ±â¼ú¿¡ ¾ÆÁÖ °ü½ÉÀÌ ¸¹´Ù.
|
|
|
¹®±Ù¹Î
µÎ´ÙÁö COO, moon@dudaji.com, KAIST »ê¾÷µðÀÚÀÎÇаú Á¹¾÷, µå·Ð â¾÷¿¡ ÀÌ¾î¼ AI ½ºÅ¸Æ®¾÷ (ÁÖ)µÎ´ÙÁö¿¡ Á¶ÀÎÇÏ¿© µö·¯´× ¹× Ŭ¶ó¿ìµå »ýÅ°踦 ÇнÀÁßÀÌ´Ù. ÀΰøÁö´ÉÀ» ÅëÇÑ ÀÚµ¿È¿Í °ü·Ã »ç¾÷ °³¹ß¿¡ °ü½ÉÀÌ ¸¹À¸¸ç, ½ÇÁ¦·Î °³¹ßÀ» ÇÒ ¼ö ÀÖ´Â Àü·« ¹× ±âȹÀÚ°¡ µÇ±â À§ÇØ ³ë·ÂÁßÀÌ´Ù
|
|
|
È«¼®È¯
µÎ´ÙÁö ´ëÇ¥, shhong@dudaji.com KAIST Àü»êÇÐ Á¹¾÷, ³Ø½¼ °ÔÀÓ °³¹ßÀÚ, KBÅõÀÚÁõ±Ç Áõ±Ç µ¥ÀÌÅÍ ºÐ¼®°¡, (Çö) µÎ´ÙÁö ´ëÇ¥ Kubernetes Forum 2019 Seoul ±âÁ¶¿¬¼³ 2020³â ÀÌÈ¿©ÀÚ´ëÇб³ °âÀÓ±³¼ö, 99ȸ ÄÄÇ»ÅͽýºÅÛÀÀ¿ë±â¼ú»ç µÎ ¹ø »ç¾÷À» ¸»¾Æ¸Ô°í ¼¼ ¹ø° »ç¾÷ ÁßÀÌ´Ù.
|
|
|
±èÈ¿µ¿
µÎ´ÙÁö CTO, button@dudaji.com, °æÈñ´ë ÀüÀÚ°ú Á¹¾÷ 2009³â â¾÷À» ÇÏ¿© IOT Ç÷§Æû °³¹ßÀ» ÇÏ¸é¼ Ç÷§ÆûÀÌ »ê¾÷¿¡ ¹ÌÄ¡´Â ¿µÇâ°ú ±× Á߿伺¿¡ ´ëÇØ ¾Ë°Ô µÇ¾ú°í, 4Â÷ »ê¾÷ÀÇ ÇÙ½ÉÀÎ ÀΰøÁö´É°ú °ü·ÃÇÏ¿© °³¹ßÀÚµéÀÌ Á» ´õ ½±°Ô Á¢±Ù ÇÒ ¼ö ÀÖ´Â AI Ç÷§ÆûÀ» °³¹ß ÁßÀÌ´Ù. Kubernetes & Kubeflow µîÀÇ open source ±â¹Ý AI ÀÚµ¿È¿¡ °ü½ÉÀÌ ¸¹À¸¸ç, °ü·Ã ³»¿ëÀ» ÁÖº¯ »ç¶÷µé¿¡°Ô ÃÖ´ëÇÑ °øÀ¯ÇÏ°í ¹ßÀü½ÃÅ°·Á°í ³ë·ÂÇÏ°í ÀÖ´Ù.
|
|
|
|
|
|
|
|
Ãâ°í¾È³» |
|
|
Ãâ°í¶õ ÀÎÅÍÆÄÅ© ¹°·ùâ°í¿¡¼ µµ¼°¡ Æ÷ÀåµÇ¾î ³ª°¡´Â ½ÃÁ¡À» ¸»Çϸç, ½ÇÁ¦ °í°´´Ô²²¼ ¼ö·ÉÇϽô ½Ã°£Àº »óÇ°Áغñ¿Ï·áÇØ Ãâ°íÇÑ ³¯Â¥ + Åùè»ç ¹è¼ÛÀÏÀÔ´Ï´Ù. |
|
ÀÎÅÍÆÄÅ© µµ¼´Â ¸ðµç »óÇ°ÀÇ Àç°í°¡ ÃæÁ·ÇÒ ½Ã¿¡ ÀÏ°ý Ãâ°í¸¦ ÇÕ´Ï´Ù. |
|
ÀϺΠÀç°í¿¡ ´ëÇÑ Ãâ°í°¡ ÇÊ¿äÇÒ ½Ã¿¡´Â ´ã´çÀÚ¿¡°Ô Á÷Á¢ ¿¬¶ôÇϽðųª, °í°´¼¾ÅÍ(°í°´¼¾ÅÍ(1577-2555)·Î ¿¬¶ôÁֽñ⠹ٶø´Ï´Ù. |
|
¹è¼Ûºñ ¾È³» |
|
|
ÀÎÅÍÆÄÅ© µµ¼ ´ë·®±¸¸Å´Â ¹è¼Û·á°¡ ¹«·áÀÔ´Ï´Ù. |
|
´Ü, 1°³ÀÇ »óÇ°À» ´Ù¼öÀÇ ¹è¼ÛÁö·Î ÀÏ°ý ¹ß¼Û½Ã¿¡´Â 1°³ÀÇ ¹è¼ÛÁö´ç 2,000¿øÀÇ ¹è¼Ûºñ°¡ ºÎ°úµË´Ï´Ù. |
¾Ë¾ÆµÎ¼¼¿ä! |
|
|
°í°´´Ô²²¼ ÁÖ¹®ÇϽŠµµ¼¶óµµ µµ¸Å»ó ¹× ÃâÆÇ»ç »çÁ¤¿¡ µû¶ó Ç°Àý/ÀýÆÇ µîÀÇ »çÀ¯·Î Ãë¼ÒµÉ ¼ö ÀÖ½À´Ï´Ù. |
|
Åùè»ç ¹è¼ÛÀÏÀÎ ¼¿ï ¹× ¼öµµ±ÇÀº 1~2ÀÏ, Áö¹æÀº 2~3ÀÏ, µµ¼, »ê°£, ±ººÎ´ë´Â 3ÀÏ ÀÌ»óÀÇ ½Ã°£ÀÌ ¼Ò¿äµË´Ï´Ù.
(´Ü, Åä/ÀÏ¿äÀÏ Á¦¿Ü) |
|
|
|
|
ÀÎÅÍÆÄÅ©µµ¼´Â °í°´´ÔÀÇ ´Ü¼ø º¯½É¿¡ ÀÇÇÑ ±³È¯°ú ¹ÝÇ°¿¡ µå´Â ºñ¿ëÀº °í°´´ÔÀÌ ÁöºÒÄÉ µË´Ï´Ù.
´Ü, »óÇ°À̳ª ¼ºñ½º ÀÚüÀÇ ÇÏÀÚ·Î ÀÎÇÑ ±³È¯ ¹× ¹ÝÇ°Àº ¹«·á·Î ¹ÝÇ° µË´Ï´Ù.
±³È¯/¹ÝÇ°/º¸ÁõÁ¶°Ç ¹× Ç°Áúº¸Áõ ±âÁØÀº ¼ÒºñÀڱ⺻¹ý¿¡ µû¸¥ ¼ÒºñÀÚ ºÐÀï ÇØ°á ±âÁØ¿¡ µû¶ó ÇÇÇظ¦ º¸»ó ¹ÞÀ» ¼ö ÀÖ½À´Ï´Ù.
Á¤È®ÇÑ È¯ºÒ ¹æ¹ý ¹× ȯºÒÀÌ Áö¿¬µÉ °æ¿ì 1:1¹®ÀÇ °Ô½ÃÆÇ ¶Ç´Â °í°´¼¾ÅÍ(1577-2555)·Î ¿¬¶ô Áֽñ⠹ٶø´Ï´Ù.
¼ÒºñÀÚ ÇÇÇغ¸»óÀÇ ºÐÀïó¸® µî¿¡ °üÇÑ »çÇ×Àº ¼ÒºñÀÚºÐÀïÇØ°á±âÁØ(°øÁ¤°Å·¡À§¿øȸ °í½Ã)¿¡ µû¶ó ºñÇØ º¸»ó ¹ÞÀ» ¼ö ÀÖ½À´Ï´Ù.
|
±³È¯ ¹× ¹ÝÇ°ÀÌ °¡´ÉÇÑ °æ¿ì |
|
|
»óÇ°À» °ø±Þ ¹ÞÀ¸½Å ³¯·ÎºÎÅÍ 7ÀÏÀ̳» °¡´ÉÇÕ´Ï´Ù. |
|
°ø±Þ¹ÞÀ¸½Å »óÇ°ÀÇ ³»¿ëÀÌ Ç¥½Ã, ±¤°í ³»¿ë°ú ´Ù¸£°Å³ª ´Ù¸£°Ô ÀÌÇàµÈ °æ¿ì¿¡´Â °ø±Þ¹ÞÀº ³¯·ÎºÎÅÍ 3°³¿ùÀ̳», ±×»ç½ÇÀ» ¾Ë°Ô µÈ ³¯ ¶Ç´Â ¾Ë ¼ö ÀÖ¾ú´ø ³¯·ÎºÎÅÍ 30ÀÏÀ̳» °¡´ÉÇÕ´Ï´Ù. |
|
»óÇ°¿¡ ¾Æ¹«·± ÇÏÀÚ°¡ ¾ø´Â °æ¿ì ¼ÒºñÀÚÀÇ °í°´º¯½É¿¡ ÀÇÇÑ ±³È¯Àº »óÇ°ÀÇ Æ÷Àå»óÅ µîÀÌ ÀüÇô ¼Õ»óµÇÁö ¾ÊÀº °æ¿ì¿¡ ÇÑÇÏ¿© °¡´ÉÇÕ´Ï´Ù.
|
|
|
|
±³È¯ ¹× ¹ÝÇ°ÀÌ ºÒ°¡´ÉÇÑ °æ¿ì |
|
|
|
°í°´´ÔÀÇ Ã¥ÀÓ ÀÖ´Â »çÀ¯·Î »óÇ° µîÀÌ ¸ê½Ç ¶Ç´Â ÈÑ¼ÕµÈ °æ¿ì´Â ºÒ°¡´ÉÇÕ´Ï´Ù. (´Ü, »óÇ°ÀÇ ³»¿ëÀ» È®ÀÎÇϱâ À§ÇÏ¿© Æ÷Àå µîÀ» ÈѼÕÇÑ °æ¿ì´Â Á¦¿Ü) |
|
½Ã°£ÀÌ Áö³²¿¡ µû¶ó ÀçÆǸŰ¡ °ï¶õÇÒ Á¤µµ·Î ¹°Ç°ÀÇ °¡Ä¡°¡ ¶³¾îÁø °æ¿ì´Â ºÒ°¡´ÉÇÕ´Ï´Ù. |
|
Æ÷Àå °³ºÀµÇ¾î »óÇ° °¡Ä¡°¡ ÈÑ¼ÕµÈ °æ¿ì´Â ºÒ°¡´ÉÇÕ´Ï´Ù. |
|
|
´Ù¹è¼ÛÁöÀÇ °æ¿ì ¹ÝÇ° ȯºÒ |
|
|
|
´Ù¹è¼ÛÁöÀÇ °æ¿ì ´Ù¸¥ Áö¿ªÀÇ ¹ÝÇ°À» µ¿½Ã¿¡ ÁøÇàÇÒ ¼ö ¾ø½À´Ï´Ù. |
|
1°³ Áö¿ªÀÇ ¹ÝÇ°ÀÌ ¿Ï·áµÈ ÈÄ ´Ù¸¥ Áö¿ª ¹ÝÇ°À» ÁøÇàÇÒ ¼ö ÀÖÀ¸¹Ç·Î, ÀÌÁ¡ ¾çÇØÇØ Áֽñ⠹ٶø´Ï´Ù. |
|
|
|
|
|
|