qucy/udacitymachinelearning

Udacity Machine Learning

Udacity Machine Learning(Advanced)

This repository is regarding machine learning course i took at Udacity China.

Projects

P0 - Titanic survival exploration

• ENV : Anaconda + Jupyter notebook + python 2.7
• Try to write a decision tree by ourself and make predict accuracy greater than 80%

P1 - Boston housing

• ENV : Anaconda + Jupyter notebook + python 2.7
• Regression problem regarding supervised learning
• Use numpy API to calculate max,min,median and std(do not use pandas API std is different with numpy's std).
• Understanding the features
• Split data into train and test
• R2 score
• Understand learning curve when will overfit
• Understand K-fold cross-validation
• Understand and use scikit-learning API GridSearchCV to find best parameters
• Understand and use quartile calculator to analyze data
• Train your first model
• Predit test data and using R2 score to validate your prediction
• Check whether your model is robust

P2 - Finding donors

• ENV : Anaconda + Jupyter notebook + python 2.7
• Binary classification problem reagrding supervised learning
• Glance your data like how many data in total, how many categories in total ect.
• Using pandas API to extract your features and labels from pandas dataframe
• Explore your data and visulize some features which have very big and very small numbers
• Using log function to handle skewed data which find by last step
• Using sickit-learn MinMaxScaler to normalize your data
• Using pandas get_dummies to do one-hot encode
• Understand recall, precision, F-score and use F-score to evaluate your model
• Understand basic classification models like, decision tree, Navie Bayes, Bagging, Boosting, SVM and NN.
• Understand the pros and cons regarding above models.
• Choose 3 models and train them, find best models according to their F-score and accuracy
• Using GridSearhCV to find best parameters for your best model
• Using model.feature_importances_ to find most 5 important features
• Train and validate model only using most 5 important features to train your model
• Evaluate model via test data

P3 - Creating customer segments

• ENV : Anaconda + Jupyter notebook + python 2.7
• Clustering problem regarding unsupervised learning
• Using pandas API dataframe.describe to glance data
• Understand KDE and find correlations between features
• Understand correlation
• Using tukey method to find outliers and remove outliers
• Understand PCA and analyze PCA
• Use PCA to transform data
• Understand and explain bio-plot
• Unserstand and compare K-means with Gaussion Mixture
• Choose one as your model train and predict
• Use cluster result as one feature may help improve model performance(Unspuervised learning can help supervised learning)

P4 - Qlearning robot new

• ENV : Anaconda + Jupyter notebook + python 3.5
• Reinforcement learning problem
• Understand Markov decision process
• Understand environment,state,action and reward
• Understand Q-learning
• Implement Q-learning in python
• Understand parameter like epsilon, gamma, alpha and epoch
• Train models with different parameters
• Understand how parameter will affect each other

P5 - Dog project

• ENV : Anaconda + Jupyter notebook + python 3.5
• Understand Tensorflow and Keras
• Understand converlutional layer, max pooling layer, averaging layer and dense layer
• Understand transfer learning
• Construct your CNN via Keras API
• Using transfer learning train your model
• Evaluate your model use other images

P6 - Capstone project dog_vs_cat

• ENV : Anaconda + Jupyter notebook + python 3.5
• Write proposal
• Write your own code
• Write your report

Nano Degree

• Machine Learning Advanced Nano Degree