[B! nmf] teddy-gã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯

teddy-g id:teddy-g

nmfã«é–¢ã™ã‚‹teddy-gã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯ (8)

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

${{author_name}}$
{{author_name}}{{created}}
{{ #comment }}{{ comment }}{{ /comment }}
- {{ label }}

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

éžè² å€¤è¡Œåˆ—å› ååˆ†è§£ï¼ˆNMFï¼‰ã«ã‚ˆã‚‹ãƒ¬ã‚³ãƒ¡ãƒ³ãƒ‰ã®ã¡ã‚‡ã£ã¨ã—ãŸä¾‹ - About connecting the dots.
æœ€è¿‘ç·šå½¢ä»£æ•°ã«ã¤ã„ã¦ã„ã‚ã„ã‚èªã¿ãªãŠã—ãŸã‚Šã—ã¦ã‚‹ã®ã§ã™ãŒï¼ˆç·šå½¢ä»£æ•°ãƒãƒ¼ãƒˆã‚·ãƒ¼ãƒˆã‚’å‰ã®è¨˜äº‹ã§ã¾ã¨ã‚ã¦ã‚ã‚Šã¾ã™ï¼‰ï¼Œãã®ä¸€ç’°ã§ãƒ¬ã‚³ãƒ¡ãƒ³ãƒ‰ã‚¢ãƒ«ã‚´ãƒªã‚ºãƒ ã«ã¤ã„ã¦ã„ãã¤ã‹è©¦ã—ã¦ã¿ãŸã®ã§ï¼Œãã‚Œã‚’è§£èª¬ã—ã¾ã™ï¼Žé †åºã¨ã—ã¦ã¯ï¼ŒåŸºæœ¬ã®å”èª¿ãƒ•ã‚£ãƒ«ã‚¿ãƒªãƒ³ã‚°ï¼ˆãƒ¦ãƒ¼ã‚¶ãƒ™ãƒ¼ã‚¹ãƒ‰ï¼Œã‚¢ã‚¤ãƒ†ãƒ ãƒ™ãƒ¼ã‚¹ãƒ‰ï¼‰â†’ç‰¹ç•°å€¤åˆ†è§£ï¼ˆSVDï¼‰â†’éžè² å€¤è¡Œåˆ—å› ååˆ†è§£ï¼ˆNMFï¼‰ã«ãªã‚Šã¾ã™ï¼Ž åŸºæœ¬çš„ãªè€ƒãˆæ–¹ ã“ã“ã§å–ã‚Šæ‰±ã†ã®ã¯ï¼Œã™ã¹ã¦ä»¥ä¸‹ã®ã‚ˆã†ãªãƒ¦ãƒ¼ã‚¶Ã—å•†å“ã®ãƒžãƒˆãƒªãƒƒã‚¯ã‚¹ã‚’ãƒ™ãƒ¼ã‚¹ã¨ã—ãŸãƒ¬ã‚³ãƒ¡ãƒ³ãƒ‰ã«ãªã‚Šã¾ã™*1ï¼Žã“ã“ã§ã¯æ˜ ç”»ãƒ¬ãƒ³ã‚¿ãƒ«ã‚µãƒ¼ãƒ“ã‚¹ã‚’ä¾‹ã«ã—ã¦è€ƒãˆã¾ã™ï¼Ž6äººã®ãƒ¦ãƒ¼ã‚¶ãŒï¼Œ4ã¤ã®æ˜ ç”»*2ã®ã†ã¡ãƒ¬ãƒ³ã‚¿ãƒ«è¦–è´ã—ãŸã‚‚ã®ã«ã¤ã„ã¦ã¯ï¼Œ1-5ç‚¹ã®5æ®µéšŽè©•ä¾¡ã‚’è¡Œã„ã¾ã—ãŸï¼Ž0ã«ãªã£ã¦ã„ã‚‹ã‚‚ã®ã¯ã€Œã¿ã¦ã„ãªã„ã€ã¨ã„ã†ã“ã¨ã«ãªã‚Šã¾ã™ï¼Ž ã¾ãšã¯ã–ã£ã¨è©•ä¾¡ã®çŠ¶æ³ã‚’ã¿ã‚‹ã¨ï¼Œã€Œåƒã¨åƒå°‹ã®ç¥žéš ã—ã€ãŒæœ€ã‚‚ã‚ˆãè¦–è´ã•ã‚Œã¦ã„ã¦ï¼Œ6äººä¸4äººãŒã¿ã¦ã„ã¾ã™ï¼Žæ¬¡ã«ã¿ã‚‰ã‚Œã¦ã„ã‚‹ã®ã¯ã€Œã¨ãªã‚Šã®ãƒˆãƒˆãƒã€
teddy-g 2016/03/01
ã‚¢ã‚¤ãƒ†ãƒ ãƒ™ãƒ¼ã‚¹ã€ãƒ¦ãƒ¼ã‚¶ãƒ™ãƒ¼ã‚¹ã®ãƒ¬ã‚³ãƒ¡ãƒ³ãƒ‰ã®åŸºæœ¬æ‰‹æ³•ã‹ã‚‰ã€SVD&NMFä½¿ã†æ–¹æ³•ã¾ã§ç°¡æ½”ã«èª¬æ˜Žã€‚ã‚¢ã‚¤ãƒ†ãƒ ã®é¡žä¼¼æ€§Ã—ãƒ¦ãƒ¼ã‚¶ãƒ—ãƒãƒ•ã‚¡ã‚¤ãƒ«ã§ãƒ¬ã‚³ãƒ¡ãƒ³ãƒ‰ã™ã‚‹éš›ã«ã€SVDæˆ–ã„ã¯NMFã§ãƒ—ãƒãƒ•ã‚¡ã‚¤ãƒ«ã‚’å†æ§‹ç¯‰ã™ã‚‹æ„Ÿã˜ã€‚

statistics

math

nmf

R

recommendation

machinelearning

bigdata
ãƒªãƒ³ã‚¯
NMF
NMF# class sklearn.decomposition.NMF(n_components='warn', *, init=None, solver='cd', beta_loss='frobenius', tol=0.0001, max_iter=200, random_state=None, alpha_W=0.0, alpha_H='same', l1_ratio=0.0, verbose=0, shuffle=False)[source]# Non-Negative Matrix Factorization (NMF). Find two non-negative matrices, i.e. matrices with all non-negative elements, (W, H) whose product approximates the non-negative
teddy-g 2016/02/27
Scikit-Learnã®NMFã®ä½¿ã„æ–¹èª¬æ˜Žå…¬å¼ãƒšãƒ¼ã‚¸ã€‚Initialå‡¦ç†ã«ãƒ©ãƒ³ãƒ€ãƒ ä»¥å¤–ã«NNDSVDã‚‚æŒ‡å®šã§ãã‚‹ã€‚fit_transformã§æ¬¡å…ƒå‰Šæ¸›ã—ãŸè¡Œåˆ—ãŒå¾—ã‚‰ã‚Œã¦ã€components_ã«å‰Šæ¸›ã—ãŸFeatureãŒæ ¼ç´ã•ã‚Œã‚‹ã€‚ä¾¿åˆ©ã€‚

python

statistics

bigdata

machinelearning

scikit-learn

nmf
ãƒªãƒ³ã‚¯
NMF for Topic Modeling | Derek Greene's Home
Topic modeling is a key tool for the discovery of hidden structure in large collections of documents. Probabilistic methods, such as Latent Dirichlet allocationTopic modeling is a key tool for the discovery of hidden structure in large collections of documents. Probabilistic methods, such as Latent Dirichlet allocation (LDA), are often employed by using tools such as the Java MALLET library. Howev
teddy-g 2016/02/25
Scikit Learnã§BBCã®ãƒ‹ãƒ¥ãƒ¼ã‚¹ï¼ˆãƒ†ã‚ã‚¹ãƒˆï¼‰ã‚’NMFã§åˆ†æžã™ã‚‹è©±ã€ã®è§£èª¬è¨˜äº‹ã®ã»ã†ã€‚ãµã‚€ãµã‚€ãªã‚‹ã»ã©ã€‚

python

scikit-learn

statistics

bigdata

machinelearning

numpy

nmf

ipython
ãƒªãƒ³ã‚¯
Jupyter Notebook Viewer
teddy-g 2016/02/25
Scikit Learnã§BBCã®ãƒ‹ãƒ¥ãƒ¼ã‚¹ï¼ˆãƒ†ã‚ã‚¹ãƒˆï¼‰ã‚’NMFã§åˆ†æžã™ã‚‹è©±ã€‚ã“ã®é€šã‚Šã«ipythonã§èµ°ã‚‰ã›ãŸã‚‰ã¡ã‚ƒã‚“ã¨çµæžœãŒå‡ºã¦ãã‚‹ã€‚ãªã‚‹ã»ã©ã€‚

python

scikit-learn

statistics

bigdata

machinelearning

numpy

nmf

ipython
ãƒªãƒ³ã‚¯
éžè² å€¤è¡Œåˆ—å› ååˆ†è§£ - sonoshouã®ã¾ã˜ã‚ãªãƒ–ãƒã‚°
ã“ã®ã‚¨ãƒ³ãƒˆãƒªãƒ¼ã¯ã€é›†åˆçŸ¥ãƒ—ãƒã‚°ãƒ©ãƒŸãƒ³ã‚°ç¬¬ï¼˜ç« ã‚’å‚ç…§ã«ã—ã¦ã„ã¾ã™ã€‚ éžè² å€¤è¡Œåˆ—å› ååˆ†è§£ã¯ã€ãƒ‡ãƒ¼ã‚¿ãƒžã‚¤ãƒ‹ãƒ³ã‚°ã®æ‰‹æ³•ã®ä¸€ã¤ã§ã‚ã‚‹ã€‚ ãƒ‡ãƒ¼ã‚¿ã®é‡è¦ãªç‰¹å¾´ã‚’æŠ½å‡ºã™ã‚‹ãŸã‚ã«ç”¨ã„ã‚‰ã‚Œã‚‹ã€‚ éžè² å€¤è¡Œåˆ—å› ååˆ†è§£ã¯ã€non-negative matrix factorizationã®æ—¥æœ¬èªžè¨³ã§ã‚ã‚Šã€ ã‚ˆãNMFã¨çœç•¥ã•ã‚Œã‚‹ã®ã§ã€ã“ã¡ã‚‰ã®çœç•¥å½¢ã‚‚è¦šãˆã¦ãŠããŸã„ã€‚ éžè² å€¤è¡Œåˆ—å› ååˆ†è§£ã®åŸºæœ¬çš„ãªã‚¢ã‚¤ãƒ‡ã‚£ã‚¢éžè² å€¤è¡Œåˆ—å› ååˆ†è§£ã¯ã€ãã®åã®é€šã‚Šã€è¡Œåˆ—ã‚’æ£ã®æ•°ï¼ˆéžè² å€¤ï¼‰ã§å› ååˆ†è§£ã™ã‚‹ã“ã¨ã§ã€ ç‰¹å¾´ã®æŠ½å‡ºã‚’è¡Œã†ã€‚ å› ååˆ†è§£ã¨ã¯ã€æŽ›ã‘åˆã‚ã›ã‚‹ã“ã¨ã§å†ã³åˆ†è§£å‰ã®è¡Œåˆ—ã‚’æ§‹ç¯‰ã§ãã‚‹ã‚ˆã†ãª ï¼’ã¤ã®å°ã•ãªè¡Œåˆ—ã‚’æŽ¢ã—å‡ºã™ã¨ã„ã†ã“ã¨ã§ã‚ã‚‹ã€‚ éžè² å€¤è¡Œåˆ—å› ååˆ†è§£ã®ä¾‹ä»¥ä¸‹ã€å…·ä½“ä¾‹ã‚’äº¤ãˆãªãŒã‚‰è§£èª¬ã™ã‚‹ã€‚ æ–‡æ›¸ã®è¨˜äº‹ã¨è¨˜äº‹å†…ã«å˜åœ¨ã™ã‚‹å˜èªžã¨ã®å¯¾å¿œä»˜ã‘ãŒã‚ã‚‹ã¨ãã€ ã“ã‚Œã‚‰ã«å¯¾ã—ã¦NMFã§ç‰¹å¾´ã‚’æŠ½å‡ºã™ã‚‹ã€‚ å¯¾å¿œä»˜ã‘ã®è¡¨ã¯ä»¥ä¸‹ã®ã‚ˆã†ã«ãªã£ã¦ã„ã‚‹ã¨ã™ã‚‹ã€‚
teddy-g 2016/02/25
Collective Intelligenceã®NMFã®èª¬æ˜Žãã®ã¾ã¾ã¨ã„ã£ã¦ã‚ˆã„ã€‚å‚™å¿˜ã€‚

python

machinelearning

bigdata

statistics

nmf
ãƒªãƒ³ã‚¯
Pythonã®nimfaã§NMFã‚’è©¦ã™ - About connecting the dots.
Pythonã§NMFã‚„ã‚‹ã«ã¯ï¼Œnimfaã¨ã„ã†ãƒ‘ãƒƒã‚±ãƒ¼ã‚¸ã‚’ä½¿ãˆã°ã‚ˆã„ã‚‰ã—ã„ã§ã™ï¼Žã¨ã‚Šã‚ãˆãšä½¿ã†ã ã‘ãªã‚‰ï¼Œé©å½“ãªnumpyè¡Œåˆ—vecã‚’ç”¨æ„ã—ã¦ï¼Œä»¥ä¸‹ã®ã‚ˆã†ã«é–¢æ•°ã«æŠ•ã’ã¦ã‚ã’ã¾ã™ï¼Ž factor = nimfa.mf(vec, seed='random_vcol', method='nmf', rank='5', max_iter=10) res = nimfa.mf_run(factor).basis() ã¨ã‚Šã‚ãˆãšã‚·ãƒ¼ãƒ‰ã¯ãƒ©ãƒ³ãƒ€ãƒ ã§ï¼Œæ‰‹æ³•ã¯ãƒ™ãƒ¼ã‚·ãƒƒã‚¯ãªnmfï¼Žä½•æ¬¡å…ƒã«å‰Šæ¸›ã™ã‚‹ã‹ã‚’rankã§æŒ‡å®šã—ã¦ï¼Œã‚¤ãƒ†ãƒ¬ãƒ¼ã‚·ãƒ§ãƒ³å›žæ•°ã‚’æ±ºã‚ã‚Œã°OKã§ã™ï¼Ž nmfã¯é–¢é€£æ‰‹æ³•ãŒå±±ã»ã©ã‚ã£ã¦ï¼Œã–ã£ã¨ä»¥ä¸‹ã®ã‚ˆã†ã«ãªã‚Šã¾ã™ï¼Žèª¬æ˜Žæ–‡ã¯åŸºæœ¬çš„ã«æ„è¨³ã§ã™ï¼Žæ£ç›´æ„è¨³ãŒã‚ã£ã¦ã‚‹ã‹ã‚‚è‡ªä¿¡ã¯ãªã„ã®ã§ï¼Œã“ã¡ã‚‰ã‹ã‚‰å…ƒè«–æ–‡ã‚’èªã¿ã¾ã—ã‚‡ã†*1ï¼Ž æ‰‹æ³• æ¦‚è¦ BD ã‚®ãƒ–ã‚¹ã‚µãƒ³ãƒ—ãƒ©ãƒ¼ã‚’ä½¿ã£ãŸãƒ™ã‚¤ã‚¸ã‚¢ãƒ³NMF BM ãƒã‚¤ãƒŠãƒªã®MF ICM It
teddy-g 2016/02/25
nimfaã£ã¦ã„ã†ãƒ‘ãƒƒã‚±ãƒ¼ã‚¸ã§NMFã§ãã‚‹ã‚‰ã—ã„ã€‚ãŒã€Scikit-Learnã«ã‚‚ã‚ã‚‹ã—ä½¿ã†ã‹ã©ã†ã‹ã¯å¾®å¦™ã€‚

python

nmf

machinelearning

bigdata
ãƒªãƒ³ã‚¯
non-Negative Matrix Factorization (NMF) - naoyaã®ã¯ã¦ãªãƒ€ã‚¤ã‚¢ãƒªãƒ¼
ä»¥å‰ã« Latent Semantic Indexing (LSI) ã‚„ HITS çµ¡ã¿ã§ SVD ã‚„ä¸»æˆåˆ†åˆ†æžã«ã¤ã„ã¦å°‘ã—æ›¸ãã¾ã—ãŸã€‚ http://d.hatena.ne.jp/naoya/20090212/latent_semantic_indexing http://d.hatena.ne.jp/naoya/20090301/hits LSI ã§ã¯ SVD ã‚’ä½¿ã£ã¦å˜èªžæ–‡æ›¸è¡Œåˆ—ã‚’åˆ†è§£ã—ã€ä½ŽéšŽæ•°è¿‘ä¼¼ã‚’è¡Œã„ã¾ã™ã€‚ã“ã‚Œã«ã‚ˆã‚Šã€ä¼¼ãŸã‚ˆã†ãªæ¬¡å…ƒã‚’ã¾ã¨ã‚ãŸã‚Šã¨ã„ã£ãŸåŠ¹æžœãŒå¾—ã‚‰ã‚Œã‚‹ã®ã§ã—ãŸã€‚è‡ªåˆ†ã®è€ƒå¯Ÿã§ã¯ HITS ã‚‚åŒæ§˜ã®ã“ã¨ã‚’è¡Œã£ã¦ã„ã‚‹ã¨ã„ã†èªè˜ã§ã—ãŸã€‚ ã•ã¦ã€é›†åˆçŸ¥ãƒ—ãƒã‚°ãƒ©ãƒŸãƒ³ã‚°ã‚’èªã‚“ã§ã„ãŸã‚‰ã€ç¬¬10ç« ã§ "non-Negative Matrix Factorization" (éžè² å€¤è¡Œåˆ—å› ååˆ†è§£, ä»¥ä¸‹NMF) ã¨ã„ã†æ‰‹æ³•ãŒå‡ºã¦ãã¾ã—ãŸã€‚NMF ã‚‚ SVD ã‚„ä¸»æˆåˆ†åˆ†æžã«åŒã˜ãè¡Œåˆ—ã‚’åˆ†è§£
teddy-g 2016/02/24
NMFã®ã‚¢ãƒ«ã‚´ãƒªã‚ºãƒ ã®ç°¡æ˜“ãªèª¬æ˜Žã€‚ã¨ã„ã†ã‹Collective Intelligenceã«æ›¸ã„ã¦ã‚ã‚‹å†…å®¹ã®æ¦‚ç•¥ã€‚

machinelearning

bigdata

statistics

nmf
ãƒªãƒ³ã‚¯
ã™ã¹ã¦ãŒMFã«ãªã‚‹ - Fire and Motion
ã™ã¹ã¦ãŒFã«ãªã‚‹ï¼Œæ˜ åƒåŒ–ã™ã‚‹ã¿ãŸã„ã§ã™ãï¼ŽçŠ€å·å‰µå¹³ã‚‚è¥¿ä¹‹åœ’èŒçµµã‚‚é…å½¹ãŒã‚¤ãƒ¡ãƒ¼ã‚¸ã¨é•ã£ã¦ä¸€éƒ¨ã§è©±é¡Œã«ãªã£ã¦ã„ã¾ã—ãŸï¼Žã•ã¦ï¼Œæœ€è¿‘ãƒ†ãƒ³ã‚½ãƒ«åˆ†è§£ã‚’ä½¿ã£ãŸè«–æ–‡ã‚’ã‚ˆãè¦‹ã‹ã‘ã‚‹ã®ã§ã™ãŒï¼Œã„ã¾ã ã«ãã¡ã‚“ã¨æ•´ç†ã§ããšã«ã„ã¾ã™ï¼Žãƒ†ãƒ³ã‚½ãƒ«ã‹ã‚ã„ã„ã‚ˆï¼Œãƒ†ãƒ³ã‚½ãƒ«ï¼Ž ãã“ã§ï¼Œã¾ãšã¯è¡Œåˆ—åˆ†è§£(matrix factorization, matrix decomposition)ã‚’æ•´ç†ã—ã¦ã¿ã‚ˆã†ã¨æ€ã„ã¾ã™ï¼Žè¡Œåˆ—ã®åˆ†è§£æ‰‹æ³•ã¨ã„ã†ã¨ç·šå½¢ä»£æ•°çš„ãªè¦³ç‚¹ã‹ã‚‰ã¯ç°¡å˜ã«æ€ã„ã¤ãã ã‘ã§ã‚‚ å›ºæœ‰å€¤åˆ†è§£ LUåˆ†è§£ ã‚³ãƒ¬ã‚¹ã‚ãƒ¼åˆ†è§£ ãªã©ãŒã‚ã‚Šã¾ã™ãŒï¼Œã“ã‚Œã‚‰ã¯åˆ†è§£å‰ã®è¡Œåˆ—ã¨åˆ†è§£å¾Œã®è¡Œåˆ—ãŒä¸€è‡´ã™ã‚‹(ãŸã¨ãˆã°A=LU)æ–¹æ³•ã§ã™ï¼Žä¸€æ–¹ã§ï¼Œæ©Ÿæ¢°å¦ç¿’ã‚„ãƒ‡ãƒ¼ã‚¿ãƒžã‚¤ãƒ‹ãƒ³ã‚°ç•Œéšˆï¼ˆç‰¹ã«ãƒ¬ã‚³ãƒ¡ãƒ³ãƒ‡ãƒ¼ã‚·ãƒ§ãƒ³ç‰ï¼‰ã§å‡ºã¦ãã‚‹è¡Œåˆ—åˆ†è§£ã¨ã„ã†ã®ã¯ï¼Œå¤§è¦æ¨¡ãƒ‡ãƒ¼ã‚¿ã‚„é–¢ä¿‚æ€§ãƒ‡ãƒ¼ã‚¿ã®ä¸ã‹ã‚‰ä½Žãƒ©ãƒ³ã‚¯ã®æ§‹é€ ã‚’æŠ½å‡ºã™ã‚‹ã“ã¨ã‚„æ¬¡å…ƒåœ§ç¸®ã‚’ç›®çš„ã¨ã—ã¦ã„ã¾ã™ï¼Žãªã®ã§ï¼Œæ£ç¢ºã«è¨€ã†ãªã‚‰ã°ï¼Œè¡Œåˆ—åˆ†è§£ã¨
teddy-g 2016/02/24
SVDã€PCAã€CURåˆ†è§£ã€NMFã®é•ã„ã¨é¡žä¼¼æ€§ã‚’å›³ã‚’ä½¿ã£ã¦èª¬æ˜Žã—ã¦ã‚‹ã€‚ç›´æ„Ÿçš„ã«ç†è§£ã™ã‚‹ã®ã«ã‚ˆã„ã€‚

statistics

machinelearning

bigdata

pca

svd

nmf

cur
ãƒªãƒ³ã‚¯
1