机器学习工程师纳米学位

Machine Learning Engineer Nanodegree

学位概述

该学位旨在让学员成长为一名机器学习工程师，在各领域如金融，健康，教育等行业中，应用机器学习的各种算法来建模。学位完成所需的时长约为400小时。

项目集目录

1. 预测世界料理

   所需技能：python, jupyter notebook, numpy, sklearn, 自然语言处理，特征提取，网格搜索，逻辑回归

2. 预测波士顿房价

   所需技能：python, jupyter notebook, sklearn，统计，特征选择，网格搜索，交叉验证，训练模型，分析模型

3. 监督学习：为慈善机构寻找捐助者

   所需技能：python, jupyter notebook, sklearn，特征工程，数据预处理，探索性数据分析及可视化，比较各监督学习模型，训练模型并优化模型，参数调优，网格搜索

4. 非监督学习：创建客户细分

   所需技能：python, jupyter notebook, sklearn，特征工程，特征缩放，异常值检测, PCA，比较各非监督学习模型，创建聚类及可视化，模型回归与分类

5. 深度学习： CNN算法进行狗的品种分类

   所需技能：卷积神经网络，keras， 迁移学习，CNN模型训练及参数优化，人脸检测，深度学习，python

6. 强化学习：训练机器人走迷宫

   所需技能：强化学习，Q-learning算法，python, 线性代数

7. 毕业项目：预测欧洲连锁药妆店Rossmann的营业额

   所需技能：数据挖掘，数据预处理，特征工程与选择，算法比较、选择与参数优化，模型回归，数据可视化，商业分析

Project Portfolio

Predict Your Cuisine

• Load the data set (39774 rows for train set and 9944 rows for test set) from Kaggle competition; perform statistical analysis
• Apply NLP techniques to pre-process the cuisine and ingredients
• Perform Logistic Regression and grid search to train and optimize the model and then make prediction for the test set

Predict Boston’s Housing Price

• Perform statistical analysis and feature engineering; split the data set and train the model with decision tree algorithm
• Perform grid search and cross validation to optimize the model and make price prediction
• Evaluate the coefficient of determination, the robustness and the applicability of the model

Supervised Learning – Finding Donors for CharityML

• Perform EDA and data wrangling on the data set
• Select 3 algorithms (Decision Tree, SVM and Adaboost) from 7 supervised learning techniques to train the model, compare and evaluate the models using different metrics and select the optimal model
• Optimize the model and make the prediction

Unsupervised Learning – Customer Segments

• Carry out feature engineering and scaling, outliers detection and removal
• Perform PCA, compare unsupervised models and create K-means clusters
• Conduct clusters and data distribution visualization, and A/B test discussion

Deep Learning – Dog Breed Classifier

• Use OpenCV's classifier to detect human faces
• Pre-process the data for CNN architecture
• Build a CNN from scratch to predict the dogs' breeds
• Apply a CNN model for transfer learning
• Write and test my algorithm based on the optimum CNN model selected to classify dog breeds

Reinforcement Learning – Robot Maze

• Create the maze; apply Q-learning algorithm to guide the robot’s movement in the maze
• Use the most updated reinforcement technique to train a robot in a maze and avoid traps

Capstone Project – Predict Rossmann Store Sales

• Download data sets from Kaggle and perform data wrangling preprocessing, EDA and data visualization
• Perform feature engineering; conduct comparison, selection and optimize of different algorithms
• Train the model and fine-tune the parameters to make prediction and visualization