機器學習及深度學習的資料

tags: `MLDL`

Machine learning

(小推)

(如果要實做的話，小推)

還是去翻書或上課，因為只是使用sklearn去套的話，大家都會壓!

https://buzzorange.com/techorange/2018/08/06/github-100-days-learning/ 印度佬100天機器學習(en)
https://github.com/Avik-Jain/100-Days-Of-ML-Code
https://zhuanlan.zhihu.com/p/54229077 有人翻譯印度佬的教學
https://github.com/MLEveryday/100-Days-Of-ML-Code 簡中翻譯版

https://buzzorange.com/techorange/2017/08/18/learn-machine-learning-and-python-in-14-steps/ (推)

(不錯，推個)
https://medium.com/@yehjames

https://medium.com/jameslearningnote/資料分析-機器學習-第1-3講-kaggle介紹-f0fd99d30f92

https://medium.com/jameslearningnote/資料分析-機器學習-第1-4講-資料科學領域的大神-網站-ae9577df2c19

https://medium.com/jameslearningnote/資料分析-機器學習-第2-1講-如何獲取資料-sklearn內建資料集-baa8f027ed7b

https://medium.com/jameslearningnote/資料分析-機器學習-第2-2講-如何獲取資料-google-map-api-beb7c88dc4e3

https://medium.com/jameslearningnote/資料分析-機器學習-第2-3講-pandas-基本function介紹-series-dataframe-selection-grouping-447a3fa90b60

https://medium.com/jameslearningnote/資料分析-機器學習-第2-4講-資料前處理-missing-data-one-hot-encoding-feature-scaling-3b70a7839b4a

https://medium.com/jameslearningnote/資料分析-機器學習-第2-5講-資料視覺化-matplotlib-seaborn-plotly-75cd353d6d3f

https://medium.com/jameslearningnote/資料分析-機器學習-第3-1講-python-機器學習以及scikit-learn介紹-fdb052463911

https://medium.com/jameslearningnote/資料分析-機器學習-第3-2講-線性分類-感知器-perceptron-介紹-84d8b809f866

https://medium.com/jameslearningnote/資料分析-機器學習-第3-3講-線性分類-邏輯斯回歸-logistic-regression-介紹-a1a5f47017e5
https://medium.com/jameslearningnote/資料分析-機器學習-第3-4講-支援向量機-support-vector-machine-介紹-9c6c6925856b

https://medium.com/jameslearningnote/資料分析-機器學習-第5-4講-機器學習進階實用技巧-正規化-8dd14fcd3140

pca

https://leemeng.tw/essence-of-principal-component-analysis.html (推推推)

https://medium.com/d-d-mag/淺談兩種降維方法-pca-與-t-sne-d4254916925b

線性代數

https://www.youtube.com/playlist?list=PLZHQObOWTQDPD3MizzM2xVFitgF8hE_ab

(不推)
https://ithelp.ithome.com.tw/users/20107244/ironman/1726
https://ithelp.ithome.com.tw/articles/10204460
https://ithelp.ithome.com.tw/articles/10204845

資料標準化

加速訓練收斂使用

https://ithelp.ithome.com.tw/articles/10197357 (推)
https://aifreeblog.herokuapp.com/posts/54/data_science_203/ (推)
https://stats.stackexchange.com/questions/70801/how-to-normalize-data-to-0-1-range (推)

資料集

http://yhhuang1966.blogspot.com/2018/04/keras-cifar-10.html

sklearn

https://stackoverflow.com/questions/54112307/deprecation-warning-in-scikit-learn 改用SimpleImputer
https://stackoverflow.com/questions/30667525/importerror-no-module-named-sklearn-cross-validation 改用model_selection
https://blog.csdn.net/qq_40773512/article/details/82662191
https://www.itread01.com/content/1541881152.html fit transform fit_transform函數的差別
https://www.studyai.cn/modules/impute.html 改用SimpleImputer

https://zhuanlan.zhihu.com/p/33569866 OneHotEncoder和LabelEncoder的講述(比較好)

https://stackoverflow.com/questions/54345667/onehotencoder-categorical-features-depreciated-how-to-transform-specific-column (categorical_features 參數被廢棄!)

https://tree.rocks/python/sklearn-explain-onehotencoder-use/ OneHotEncoder
https://www.cnblogs.com/zhoukui/p/9159909.html OneHotEncoder
https://www.itread01.com/elllp.html OneHotEncoder

https://zhengheng.me/2017/08/01/house-pricing/

https://www.cnblogs.com/chaosimple/p/4153167.html 標準化(z分數)

結果將資料變成平均數0,標準差為1的資料!!!

官方文件

https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.OneHotEncoder.html
https://scikit-learn.org/stable/modules/generated/sklearn.compose.ColumnTransformer.html

Deep learning

https://jerrynest.io/deep-learning-resource/
https://zhuanlan.zhihu.com/p/25794795 (不推)
https://medium.com/雞雞與兔兔的工程世界/機器學習-ml-note-overfitting-過度學習-6196902481bb (推)
https://medium.com/雞雞與兔兔的工程世界/機器學習-ml-note-convolution-neural-network-卷積神經網路-bfa8566744e9 (中大推)
https://medium.com/雞雞與兔兔的工程世界/機器學習-ml-note-cnn演化史-alexnet-vgg-inception-resnet-keras-coding-668f74879306 (中大推)
https://www.zhihu.com/question/36686900 (推)
http://scs.ryerson.ca/~aharley/vis/conv/?source=post_page–––––––––––––- 模擬lenet的運算 (推)

https://zh.wikipedia.org/wiki/過適

https://docs.aws.amazon.com/zh_tw/machine-learning/latest/dg/model-fit-underfitting-vs-overfitting.html 過適&乏適

(推推)
https://medium.com/@syshen/入門深度學習-1-ed873b65bc5
https://medium.com/@syshen/入門深度學習-2-d694cad7d1e5 寫得好推
https://medium.com/cubo-ai/物體偵測-object-detection-740096ec4540 (物體偵測推推!)

https://brohrer.mcknote.com/zh-Hant/
https://brohrer.mcknote.com/zh-Hant/how_machine_learning_works/how_neural_networks_work.html
https://brohrer.mcknote.com/zh-Hant/how_machine_learning_works/how_convolutional_neural_networks_work.html (推)
https://brohrer.mcknote.com/zh-Hant/how_machine_learning_works/how_rnns_lstm_work.html (推)

(可)

https://ithelp.ithome.com.tw/users/20001976/ironman/1395
https://ithelp.ithome.com.tw/articles/10191404
https://ithelp.ithome.com.tw/articles/10191528
https://ithelp.ithome.com.tw/articles/10191627
https://ithelp.ithome.com.tw/articles/10191725
https://ithelp.ithome.com.tw/articles/10192950

(推)
https://medium.com/jameslearningnote/資料分析-機器學習-第5-1講-卷積神經網絡介紹-convolutional-neural-network-4f8249d65d4f

(推)
https://tomkuo139.blogspot.com/2018/03/ai.html
https://tomkuo139.blogspot.com/2018/05/aimlp-cifar10.html
https://tomkuo139.blogspot.com/2018/05/aicnn-cifar10.html

(推)
https://medium.com/@baubibi
https://medium.com/@baubibi/速記ai課程-深度學習入門-一-68e27912ce30
https://medium.com/@baubibi/速記ai課程-convolutional-neural-networks-for-computer-vision-applications-一-e7059b3e0071

https://mropengate.blogspot.com/2017/02/deep-learning-role-of-activation.html 使用激勵函數的原因
https://kknews.cc/zh-tw/tech/y848kvn.html 權重初始化(另一篇)
https://zhuanlan.zhihu.com/p/25110150 權重初始化

sigmoid, tanh => Xavier (別用，反向傳播之下有梯度消失的問題)
relu => He
減少對初始化的依賴 Batch Norm.

改善cnn辨識率

內心os:實際訓練會常常發生overfit的狀況!XD

https://medium.com/@syshen/改善-cnn-辨識率-dac9fce59b63 (推)

(推)
https://medium.com/@CinnamonAITaiwan

https://medium.com/@CinnamonAITaiwan/cnn模型-損失函數-loss-function-647e13956c50

https://medium.com/datadriveninvestor/l1-l2-regularization-7f1b4fe948f2

https://medium.com/@CinnamonAITaiwan/cnn入門-overfitting-d10acd15ec21 overfit討厭!
https://medium.com/@CinnamonAITaiwan/深度學習-cnn原理-keras實現-432fd9ea4935
https://medium.com/@CinnamonAITaiwan/cnn入門-圖像增強-fa654d36dafc
https://medium.com/@CinnamonAITaiwan/cnn模型-resnet-mobilenet-densenet-shufflenet-efficientnet-5eba5c8df7e4

https://medium.com/@chih.sheng.huang821/卷積神經網路-convolutional-neural-network-cnn-cnn運算流程-ecaec240a631

http://elmer-storage.blogspot.com/2018/07/cnn-convolutional-neural-network-cnn.html

https://medium.com/bryanyang0528/deep-learning-keras-手寫辨識-mnist-b41757567684

(基本上我覺得還不錯!!!XD)

http://hemingwang.blogspot.com/2016/12/ai_20.html (推)

http://hemingwang.blogspot.com/2018/09/ai_10.html (可)

http://hemingwang.blogspot.com/2018/09/ai_50.html (可)

http://hemingwang.blogspot.com/2017/06/aiweight-decay.html

http://hemingwang.blogspot.com/2019/05/trilogy.html

圖解dl

https://leemeng.tw/deep-learning-for-everyone-understand-neural-net-and-linear-algebra.html

https://leemeng.tw/deep-learning-resources.html

李弘毅

https://violin-tao.blogspot.com/2017/07/ml-introduction-of-deep-learning.html
https://violin-tao.blogspot.com/2017/07/ml-backpropagation.html
https://violin-tao.blogspot.com/2017/07/ml-tips-for-training-dnn.html
https://violin-tao.blogspot.com/2017/07/ml-convolutional-neural-network-cnn.html
https://violin-tao.blogspot.com/2017/07/ml-why-deep.html
https://violin-tao.blogspot.com/2018/02/ml-batch-normalization.html

keras (推)

https://machinelearningmastery.com/how-to-reduce-overfitting-in-deep-learning-with-weight-regularization/
https://stackoverflow.com/questions/41260042/global-weight-decay-in-keras
https://ithelp.ithome.com.tw/articles/10191725
https://github.com/keras-team/keras/issues/2717
https://ithelp.ithome.com.tw/articles/10191627
https://chtseng.wordpress.com/2017/09/23/學習使用keras建立卷積神經網路/

官方文件

https://keras.io/zh/losses/
https://keras.io/zh/metrics/
https://keras.io/zh/optimizers/
https://keras.io/zh/activations/
https://keras.io/zh/datasets/
https://keras.io/zh/initializers/
https://keras.io/zh/regularizers/ L2正規化去懲罰權重
https://keras.io/zh/visualization/
https://keras.io/zh/preprocessing/image/

循序模型
https://keras.io/zh/getting-started/sequential-model-guide/
https://keras.io/zh/models/sequential/

函數api(平行模型)
https://keras.io/zh/getting-started/functional-api-guide/
https://keras.io/zh/models/model/

Q&A
https://keras.io/zh/getting-started/faq

核心網路層
https://keras.io/zh/layers/core/

卷積層(Conv2D)
https://keras.io/zh/layers/convolutional/

池化層(MaxPooling2D)
https://keras.io/zh/layers/pooling/

Batch Norm.層
https://keras.io/zh/layers/normalization/

資料增強(data augmentation)
https://chtseng.wordpress.com/2017/11/11/data-augmentation-資料增強/
https://github.com/ch-tseng/data-augmentation-Keras/blob/master/DataAugmentation-1.ipynb 範例筆記本

cnn padding same stride > 1
https://www.chzzz.club/post/192.html
https://blog.csdn.net/wuzqChom/article/details/74785643

tensorflow

一些colab的線上教學

https://colab.research.google.com/notebooks/mlcc/hello_world.ipynb
https://colab.research.google.com/notebooks/mlcc/tensorflow_programming_concepts.ipynb
https://colab.research.google.com/notebooks/mlcc/creating_and_manipulating_tensors.ipynb

colab (推)

電腦太老的好東西

記得開gpu訓練否則你要等很久…

https://medium.com/pyradise/上傳檔案到google-colab-dd5369a0bbfd

https://medium.com/@white1033/利用google-colaboratory-使用免費gpu-b98352b9575d 免費的tpu及gpu

https://technews.tw/2017/04/07/first-in-depth-look-at-googles-tpu-architecture/ tpu的好處

https://mattwang44.github.io/en/articles/colab/ 基本操作

https://mc.ai/第一次用-google-colab-就上手/

特殊技巧

Pytorch

https://medium.com/pyladies-taiwan/深度學習新手村-pytorch入門-511df3c1c025
https://zhuanlan.zhihu.com/p/54350088
https://fgc.stpi.narl.org.tw/activity/videoDetail/4b1141305d9cd231015d9d0992ef0030

Yolo v2 v3

https://zhuanlan.zhihu.com/p/58776542

https://www.zhihu.com/appview/p/45845454 (國人佛心來著!)

https://github.com/pjreddie/darknet (原作者佛心來著!)

https://github.com/AlexeyAB/darknet (windows下安裝推!)

http://yy-programer.blogspot.com/2019/01/yolo-darknet.html (大推安裝好文!)

總之! 不要不信邪! 他要什麼版本的軟體你就照著他去弄! 不要像我一樣! 像個87試了老半天才成功!(淚灑!)

https://medium.com/@chih.sheng.huang821/深度學習-物件偵測yolov1-yolov2和yolov3-cfg-檔解讀-75793cd61a01
https://medium.com/@chih.sheng.huang821/深度學習-物件偵測yolov1-yolov2和yolov3-cfg-檔解讀-二-f5c2347bea68

https://medium.com/@wayne18308/yolo實作教學-train出屬於自己的model使用darkflow-windows-9b633b7350d8
http://dreamisadream97.pixnet.net/blog/post/172166721-利用-yoiov3-訓練自己的數據-（詳細教學）

https://medium.com/雞雞與兔兔的工程世界/機器學習-ml-note-yolo-利用影像辨識做物件偵測-object-detection-的技術-3ad34a4cac70 ** yolov2 **

https://saferauto.home.blog/2019/04/06/4o-how-to-install-yolo-darknet-with-cuda-and-opencv-in-ubuntu/

https://zhuanlan.zhihu.com/p/41230124

某一個大佬的v3教學

https://github.com/inhail/darkflow

https://medium.com/@yanweiliu/python影像辨識筆記-八-分別在windows和ubuntu-18-04上安裝並執行yolov3-使用gpu-d2b77347fde

https://medium.com/@yanweiliu/python影像辨識筆記-九之二-關於yolov3的一些心得-754b3a7e70e1

https://medium.com/@yanweiliu/python影像辨識筆記-九之三-yolov3結合深度學習的object-detector-65440b54422b

https://medium.com/@yanweiliu/nvidia-jetson-tx2學習筆記-三-安裝opencv-c62e2435ad57

labelimg

https://tzutalin.github.io/labelImg/

darkflow(python版)

https://github.com/thtrieu/darkflow

樹梅派也要秀yolov3

ninja 潮他媽的！

https://medium.com/@yanweiliu/raspberry-pi學習筆記-二十七-在pi上執行yolov3-9cf124d5d582

Anchor

https://github.com/pjreddie/darknet/issues/568
https://stackoverflow.com/questions/54068396/should-i-change-the-value-of-anchors-in-yolo-obj-cfg
https://stackoverflow.com/questions/56442413/generating-anchor-boxes-using-k-means-clustering-yolo
https://www.angtk.com/article/56
https://github.com/pjreddie/darknet/issues/911

Map

https://www.zhihu.com/question/53405779
https://blog.csdn.net/hysteric314/article/details/54093734

技巧

https://ithelp.ithome.com.tw/articles/10190971 遷移學習

作者大神論文

https://hackmd.io/@allen108108/r1-wSTAjS 好心人

學習地圖

技能樹:

微積分的微分
統計
線性代數(矩陣、向量的操作)
Python(Numpy、matplotlib.pyplot、pandas、opencv、tensorflow、keras、sklearn)

機器學習(ML)範圍:

感知器(perceptron) => NN網路(affine, 全連接)

深度學習(DL)範圍:

影像方面:

接續 => CNN(小難) => 深度學習

文字方面:

接續 => RNN => LSTM => 深度學習

python

pickle 序列化反序列化

把物件存成檔案!XD

https://morvanzhou.github.io/tutorials/python-basic/basic/13-08-pickle/

PIL的image與numpy矩陣互相轉換

https://www.jianshu.com/p/18dabefa6778

numpy (必學！矩陣運算包) (推)

https://www.twblogs.net/a/5b8fccbb2b71776722159c6f

https://blog.csdn.net/baoqian1993/article/details/52116164

numpy axis概念整理

http://changtw-blog.logdown.com/posts/895468-python-numpy-axis-concept-organize-notes

pandas (似excel) (推)

https://medium.com/@weilihmen/python-pandas-基本操作教學-成績表-f6d0ec4f89
https://colab.research.google.com/notebooks/mlcc/intro_to_pandas.ipynb colab有關pandas的入門教學
https://blog.csdn.net/qq1483661204/article/details/77587881 iloc用法

Matplotlib (推)

https://www.itread01.com/content/1542579970.html

numpy pandas matplotlib (畫圖包)

https://codertw.com/程式語言/565247/

opencv (影像處理包) (小推，做cv時，用得到)

https://blog.gtwang.org/programming/opencv-basic-image-read-and-write-tutorial/ 基本讀取操作
http://jennaweng0621.pixnet.net/blog/post/403254017-opencv-彩色轉灰階(rgb-to-gray)
https://blog.csdn.net/on2way/article/details/46812121 反相黑白
https://www.jianshu.com/p/3977d674da85
http://monkeycoding.com/?page_id=12

軟體

anaconda 相關 (python資料科學懶人包) (推)

https://medium.com/python4u/用conda建立及管理python虛擬環境-b61fd2a76566
https://tomkuo139.blogspot.com/2018/03/anaconda-tensorflow-keras.html 安裝tensorflow keras

jupyter notebook (推)

http://opus.konghy.cn/ipynb/jupyter-notebook-keyboard-shortcut.html

(可能)參考書籍

https://www.books.com.tw/products/0010822932 (o) 推推推
https://www.books.com.tw/products/0010761759 (o) (大大大大大推)
https://www.books.com.tw/products/0010817138 (x)
https://www.books.com.tw/products/0010754327?sloc=reprod_i_10 (x)
http://tensorflowkeras.blogspot.com/2017/08/keras.html (x)
https://www.books.com.tw/products/0010822845?sloc=reprod_i_11 (o) 不太推
https://www.books.com.tw/products/0010811811 (x, 評價…恩)
https://www.books.com.tw/products/0010797283?loc=P_asb_002 (o)
…(待增加)

課程

https://zh-tw.coursera.org/learn/machine-learning 史丹佛大學吳恩達教授
https://www.youtube.com/channel/UC2ggjtuuWvxrHHHiaDH1dlQ 台大李宏毅教授

似然函數及最大似然估計法(經二項分布) https://pansci.asia/archives/124043

由結果去求原始的機率
e.g. 取10顆球有1顆紅球的機率為0.1
其實就是代表「取10顆球有1顆紅球的機率「最有」可能是10%，可能還是有誤差!!!

https://kknews.cc/zh-tw/game/j6vgbnl.html (最大似然估計法)

而極大似然法就是令樣本出現的機率最大，進而估計整體的模型參數。
(最符合這個(觀察出來的)機率分佈的模型)

http://ccckmit.wikidot.com/st:maximumlikelihood (最大似然估計法)

因此，設定 p(x)=p′(x) 的想法，其背後的目標乃是要最大化機率原模型 p 產生 p' 現象的可能性，這個最大化的目標就稱為最大似然法則。

機器學習及深度學習的資料

tags: MLDL

Machine learning

pca

線性代數

資料標準化

資料集

sklearn

官方文件

Deep learning

圖解dl

李弘毅

keras (推)

官方文件

tensorflow

colab (推)

特殊技巧

鐵人賽文章

在上面跑yolo(先跪再說)

開啟camera

裝cuda cudnn

Pytorch

Yolo v2 v3

某一個大佬的v3教學

labelimg

darkflow(python版)

樹梅派也要秀yolov3

Anchor

Map

技巧

作者大神論文

學習地圖

python

pickle 序列化 反序列化

PIL的image與numpy矩陣互相轉換

numpy (必學！矩陣運算包) (推)

numpy axis概念整理

pandas (似excel) (推)

Matplotlib (推)

numpy pandas matplotlib (畫圖包)

opencv (影像處理包) (小推，做cv時，用得到)

軟體

anaconda 相關 (python資料科學懶人包) (推)

jupyter notebook (推)

(可能)參考書籍

課程

統計相關

tags: `MLDL`

pickle 序列化反序列化