# IIIEDU-BDSE11-ML-DL-with-Python
本頁連結: https://bit.ly/2o0RK8L


## 檔案下載
* [講義](https://www.dropbox.com/s/oikm7bgxuyhvcr4/BDSE11ML-and-DL.pdf?dl=0)
* [Notebooks & Datasets](https://www.dropbox.com/s/rlnepy45epue397/bdse11_20190928.tar.gz?dl=0)

## Spark example
https://www.dropbox.com/s/a62cnm7ove8rrdr/sparkMachineLearning.html?dl=0

## 環境
若為Windows環境 (無GPU)，請安裝Anaconda後，加裝如下套件:
* 基本套件

    pip install pandas matplotlib seaborn numpy scipy

* 機器學習套件

    pip install tensorflow sklearn xgboost==0.90

* 其它套件

    pip install pillow graphviz
    pip install mlxtend
    pip install numexpr

    conda install -y shapely
    pip install imgaug opencv-python==3.4.7.28
    
    pip install jupyterlab

## 上課範例
* 00-quick-review [[html]](http://211.75.15.16:60000/bdse11-00-quick-review.html) [[ipynb]](http://211.75.15.16:60000/bdse11-00-quick-review.ipynb)
* 01-TFGradientTapeAndLinearRegression [[html]](http://211.75.15.16:60000/bdse11-01-TFGradientTapeAndLinearRegression.html) [[ipynb]](http://211.75.15.16:60000/bdse11-01-TFGradientTapeAndLinearRegression.ipynb)
* 02-SklearnLinearRegression [[html]](http://211.75.15.16:60000/bdse11-02-SklearnLinearRegression.html) [[ipynb]](http://211.75.15.16:60000/bdse11-02-SklearnLinearRegression.ipynb)
* 03-SklearnQuickTour [[html]](http://211.75.15.16:60000/bdse11-03-SklearnQuickTour.html) [[ipynb]](http://211.75.15.16:60000/bdse11-03-SklearnQuickTour.ipynb)
* 04-TitanicSurvivalEDA [[html]](http://211.75.15.16:60000/bdse11-04-TitanicSurvivalEDA.html) [[ipynb]](http://211.75.15.16:60000/bdse11-04-TitanicSurvivalEDA.ipynb)
* 05-TitanicSurvivalAnalysis[[html]](http://211.75.15.16:60000/bdse11-05-TitanicSurvivalAnalysis.html) [[ipynb]](http://211.75.15.16:60000/bdse11-05-TitanicSurvivalAnalysis.ipynb)
* 06-XGBoostBlFriday [[html]](http://211.75.15.16:60000/bdse11-06-XGBoostBlFriday.html) [[ipynb]](http://211.75.15.16:60000/bdse11-06-XGBoostBlFriday.ipynb)
* 07-SklearnPCA [[html]](http://211.75.15.16:60000/bdse11-07-SklearnPCA.html) [[ipynb]](http://211.75.15.16:60000/bdse11-07-SklearnPCA.ipynb)
* 08-SklearnDBSCAN [[html]](http://211.75.15.16:60000/bdse11-08-SklearnDBSCAN.html) [[ipynb]](http://211.75.15.16:60000/bdse11-08-SklearnDBSCAN.ipynb)
* 09-KerasLayersIO [[html]](http://211.75.15.16:60000/bdse11-09-KerasLayersIO.html) [[ipynb]](http://211.75.15.16:60000/bdse11-09-KerasLayersIO.ipynb)
---
* 15-KerasSimpleRNNManyToOne [[html]](http://211.75.15.16:60000/bdse11-15-KerasSimpleRNNManyToOne.html) [[ipynb]](http://211.75.15.16:60000/bdse11-15-KerasSimpleRNNManyToOne.ipynb)
* 16-KerasSimpleRNNManyToMany [[html]](http://211.75.15.16:60000/bdse11-16-KerasSimpleRNNManyToMany.html) [[ipynb]](http://211.75.15.16:60000/bdse11-16-KerasSimpleRNNManyToMany.ipynb)
* 17-KerasLSTMSurvivalPredICU [[html]](http://211.75.15.16:60000/bdse11-17-KerasLSTMSurvivalPredICU.html) [[ipynb]](http://211.75.15.16:60000/bdse11-17-KerasLSTMSurvivalPredICU.ipynb)
---
* 09-KerasLayersIO [[html]](http://211.75.15.16:60000/bdse11-09-KerasLayersIO.html) [[ipynb]](http://211.75.15.16:60000/bdse11-09-KerasLayersIO.ipynb)
* 10-KerasMNIST[[html]](http://211.75.15.16:60000/bdse11-10-KerasMNIST.html) [[ipynb]](http://211.75.15.16:60000/bdse11-10-KerasMNIST.ipynb)
* 11-KerasInceptionAndVGG[[html]](http://211.75.15.16:60000/bdse11-11-KerasInceptionAndVGG.html) [[ipynb]](http://211.75.15.16:60000/bdse11-11-KerasInceptionAndVGG.ipynb)
* 12-KerasResNetAndDenseNet[[html]](http://211.75.15.16:60000/bdse11-12-KerasResNetAndDenseNet.html) [[ipynb]](http://211.75.15.16:60000/bdse11-12-KerasResNetAndDenseNet.ipynb)
---
## 補充連結
* [Python Machine Learning 課本的筆記本教材](https://github.com/rasbt/python-machine-learning-book-2nd-edition)
* [Pandas 指令](https://studylib.net/doc/25268801/pandas-cheat-sheet)
* [有大量dummy variables時，可改用H2O做隨機森林/決策樹](https://roamanalytics.com/2016/10/28/are-categorical-variables-getting-lost-in-your-random-forests/) (Scikit-learn內建的樹不適合用來處理大量dummy variables)
* [XGBoost parameters](https://xgboost.readthedocs.io/en/latest/parameter.html#learning-task-parameters)
## 作業

* **hw1**: Linear regression: 離群值會讓模型學得比較不好。試著去掉離群值，然後再訓練一次模型，看$R^2$有沒有變好。
* **hw2**: XGBoost: 控制L1/L2規範項的強度，看模型準確率有無變化
## 補充投影片
* 9/29
* 10/5
* 10/6
* 上述例子是將一個類別欄位變成了一百個dummy欄位。這一百個欄位任選一個來做決策，只能降低約1%的整體不純度。
* 所以這個類別欄位，不管重要或不重要，都容易被決策樹/森林忽略。
* 做決策樹/森林時，建議不要將類別欄位變成dummy欄位。可惜的是目前Scikit-learn還是要求類別型欄位必須做dummy。建議改用XGBoost或是H2O來實作樹類演算法。
* 10/16
* 10/28

## GPU環境
* [01](http://192.168.20.89:32601/?token=152218be33c6ca86813d11ccb55061579a2e74fa6971a014)
* [02](http://192.168.20.88:30252/?token=a210e456b53fdbea1e6680cc8e3e65cbaf87a21e3d4aae1a)
* [03](http://192.168.20.88:31595/?token=f9cf12f0b1d57820c664c1c46e59b84e98e9755dbbb099b9)
* [04](http://192.168.20.89:30821/?token=a778e986dd5043c53277e74457f64d13ec41f3e38e75bb61)
* [05](http://192.168.20.89:31706/?token=cfe294e8fd7f8a03f6096b6e618b200d79602a3650f6b80f)
---
翁啟閎
[e-mail]