![import numpy as np
import pandas as pd
import sklearn.linear_model # import sklearn.neighbors
# Load the data
oecd_bli = pd.read_csv(datapath + "oecd_bli_2015.csv", thousands=',')
gdp_per_capita = pd.read_csv(datapath + "gdp_per_capita.csv",thousands=',',delimiter='t', encoding='latin1',
na_values="n/a")
# Prepare the data
country_stats = prepare_country_stats(oecd_bli, gdp_per_capita)
X = np.c_[country_stats["GDP per capita"]]
y = np.c_[country_stats["Life satisfaction"]]
# Select a linear model
model = sklearn.linear_model.LinearRegression() # model = sklearn.neighbors.KNeighborsRegressor(n_neighbors=3)
# Train the model
model.fit(X, y)
# Make a prediction for Cyprus
X_new = [[22587]] # Cyprus' GDP per capita
print(model.predict(X_new)) # outputs [[ 5.96242338]]
Train a linear model using Scikit-Learn
43](https://image.slidesharecdn.com/handson-ml-ch1-210721104434/75/Hands-on-ML-CH1-43-2048.jpg)
![import numpy as np
import pandas as pd
import sklearn.linear_model # import sklearn.neighbors
# Load the data
oecd_bli = pd.read_csv(datapath + "oecd_bli_2015.csv", thousands=',')
gdp_per_capita = pd.read_csv(datapath + "gdp_per_capita.csv",thousands=',',delimiter='t', encoding='latin1',
na_values="n/a")
# Prepare the data
country_stats = prepare_country_stats(oecd_bli, gdp_per_capita)
X = np.c_[country_stats["GDP per capita"]]
y = np.c_[country_stats["Life satisfaction"]]
# Select a linear model
model = sklearn.linear_model.LinearRegression() # model = sklearn.neighbors.KNeighborsRegressor(n_neighbors=3)
# Train the model
model.fit(X, y)
# Make a prediction for Cyprus
X_new = [[22587]] # Cyprus' GDP per capita
print(model.predict(X_new)) # outputs [[ 5.96242338]]
Train a linear model using Scikit-Learn
43](https://crownmelresort.com/image.slidesharecdn.com/handson-ml-ch1-210721104434/75/Hands-on-ML-CH1-43-2048.jpg)
Uploaded byJamie (Taka) Wang
Hands-on ML - CH1
This document provides an overview of machine learning concepts including: - The differences between deep learning, neural networks, machine learning, and artificial intelligence. - Examples of machine learning applications such as image classification, text summarization, and fraud detection. - The main types of machine learning including supervised, unsupervised, semi-supervised, and reinforcement learning. - Common challenges in machine learning like bad data, overfitting, and underfitting models. - Methods for evaluating machine learning models like validation sets and cross-validation.
More Related Content
![[系列活動] 手把手的深度學實務](https://cdn.slidesharecdn.com/ss_thumbnails/slidedlfinal1216-171213041058-thumbnail.jpg?width=640&height=640&fit=bounds)
![[系列活動] 手把手的深度學習實務](https://cdn.slidesharecdn.com/ss_thumbnails/slidesharestepbystepdl-161213072731-thumbnail.jpg?width=640&height=640&fit=bounds)
![[DSC x TAAI 2016] 林守德 / 人工智慧與機器學習在推薦系統上的應用](https://cdn.slidesharecdn.com/ss_thumbnails/md-161124235136-thumbnail.jpg?width=640&height=640&fit=bounds)
Hands-on ML - CH1
- 1.
- 2.
- 3. 3 Difference between deeplearning and usual ML
- 4. Why use machinelearning Traditional Approach Machine Learning Approach Learning from data 4 Email Spam
- 5. Automatically adapting tochange It's easier to automate than traditional methods 5
- 6. Machine Learning canhelp humans learn 6
- 7. When to usemachine learning? ● Exists some underlying pattern to be learned ● Performance measure can be improved ● But no programmable definition ● There is data about the pattern ● ... 7 Ref: NTU 林軒田教授 機器學習基石
- 8. Example of Applications ★分析產線線的產品圖片，自動分類。 ○ CNNs (Classification) ★ 在腦部腫瘤照片內找出腫瘤 ○ CNNs (Segmentation) ★ 自動分類新聞文章/論壇評論 ○ NLP (Text Classification) ○ RNNs, CNNs, Transformers ★ 自動摘要長篇文章 ○ NLP (Text Summarization) ○ RNNs, CNNs, Transformers ★ 建立自動聊天機器人或個人助理 ○ NLP, NLU ★ 讓 APP 回應語音指示 ○ Speech recognition ○ RNNs, CNNs, Transformer ★ 根據許多績效評量預估公司明年營收 ○ Linear/Polynomial Regression ○ SVMs, Random Forests, ANNs ○ RNNs, CNNs, Transformers (過往) ★ 偵測信用卡詐欺 ○ 異常偵測 (Anomaly Detection) ★ 根據顧客購買情況進行劃分，為不同市場類 別設計不同的行銷策略 ○ 分群法 (Clustering) ★ 用清晰且富洞察力的圖表表示複雜的高維度 資料 ○ Data visualization，降維 ★ 根據過往的購買習慣推薦顧客可能感興趣的 產品 ○ 推薦系統 ★ 建立遊戲中的智慧機器人 ○ 強化式學習 (Reinforcement Learning) 8
- 9. Types of MachineLearning Systems ★ levels of human supervision ○ supervised ○ unsupervised ○ semi-supervised ○ reinforcement learning ★ learn on-the-fly ○ online learning ○ batch learning ★ predict via known instances or learned pattern ○ instance-based ○ model-based 以上為本書提到的 criteria，還有其他分類方式。 這些 criteria 並沒有互斥，很多系統都是採用混合式的做法。 9
- 10.
- 11. Unsupervised Learning (非監督式學習) Clustering Unlabeledtraining set 11 Customer
- 12. Unsupervised Learning (非監督式學習) AnomalyDetection (Visualization) Semantic Clusters 12 Credit card fraud
- 13. Unsupervised Learning (非監督式學習) 13 AssociationRule Learning
- 14. Semi-supervised Learning (半監督式學習) 14 有三角形 與 正方形 兩種類別 圓形 代表無標籤的樣本 Google Phtos
- 15.
- 16.
- 17. Batch vs OnlineLearning ● 批次學習也稱為離線學習 ● 舊資料+新資料 → 重新訓練 ● 一段時間重新訓練 17 learning on-the-fly 批次學習 mini-batch Learning Rate (遺忘速率)
- 18. Instance-based vs Model-basedLearning 18 model-based learning Instance-based learning memory-based
- 19. Model-based Learning Example 19 CountryGDP ($) Life Satisfaction Hungary 12,240 4.9 Korea 27,195 5.8 France 37,675 6.5 Australia 50,962 7.3 United States 55,805 7.2 Vague Correlation
- 20. Model-based Learning SimpleLinear Model 20 life_satification = θ0 + θ1 x GDP Model* Selection Parameter Measurement cost function, utility function Training
- 21. 資料 vs 演算法的重要性資料的影響力出人意料 21 應該把錢花在演算法還是語料庫？
- 22.
- 23. Main Challenges ofMachine Learning Traditional Approach Machine Learning Approach 23
- 24. Bad Data 24 1. InsufficientQualitity of Training Data 2. Nonrepresentative Training Data a. too small → sampling noise b. very large → sampling bias 3. Poor-Quality Data a. Missing value b. Outliers 4. Irrelevant Features a. Feature Extraction b. Feature Selection Model Bias
- 25. 1936 Literary DigestPoll 文學文摘 (The Literary Digest) 於事前進行的民 調。文學文摘郵寄1,000萬份問卷予其讀者，回收 230萬份。 文學文摘準確預測此前 5次總統選舉結果，並於 1936年10月31日宣布，他們預測共和黨候選人阿 爾夫·蘭登會在531張選舉人票中獲得370張選票 並勝出 (實際上他只獲得8張選舉人票)
- 26.
- 27. Bad Algorithms -Overfitting the Training Data 27 ★ Good training error ★ Bad error generalization error ★ 字母w開頭的國家滿意度都很高?! ➢ Simplify model ➢ Regularization ➢ Gather more data ➢ Reduce the noise 可能解法
- 28. Bad Algorithms -Overfitting the Training Data 28
- 29. Bad Algorithms -Underfitting the Training Data 29 ★ Powerful model with more parameters ★ Better features (Feature engineering) ★ Reducing the constraints (regularization hyperparameter)
- 30. Stepping Back ★ 機器學習是讓電腦從資料中學習而變得更好，不是明確寫規則。 ★機器學習有很多種類，監督與否，批次或線上，基於實例或模型。 ★ 演算法是基於模型(參數)或者實例的，透過訓練集微調。 ★ 訓練集太小，雜訊，沒有代表性。 ★ 模型太簡單，模型太複雜。 30
- 31.
- 32. 思路 1. 如何評估模型效果？直接上線測試？ 2. 如何選擇模型？ 3.選定模型後，如何調整超參數？ 32
- 33.
- 34. Traing Set andTest Set 34
- 35. Model Selection 35 比較模型的 TestingError (Generalization error)
- 36. Hyperparameter Tuning 實驗各種參數組合，得到最好的 GeneralizationError 但實測還是不佳→可能已經對測試集優化了 36
- 37. Training Set TestSet Validation Set Holdout Validation 37 aka development set or dev set
- 38. Validation Set 遭遇的問題 驗證集切太大訓練集剩太少 無法代表驗證集與 訓練集合起來的效果 驗證集切太小 評估很不準 38
- 39.
- 40.
- 41. Common Supervised LearningAlgorithms 41 ★ K-Nearest Neighbors ★ Linear regression ★ Logistic regression ★ Support Vector Machines (SVMs) ★ Decision Tree ★ Random Forest ★ Neural Networks*
- 42. Common Unsupervised LearningAlgorithms 42 ★ Clustering ○ K-Means ○ DBSCAN ○ Hierarchical Cluster Analysis (HCA) ★ Anomaly detection and novelty detection ○ One-class SVM ○ Isolation Forest ★ Visualization and dimensionality reduction ○ Principal Component Analysis (PCA) ○ Kernel PCA ○ Locally-Linear Embedding (LLE) ○ t-distributed Stochastic Neightboar Embedding (t-SNE) ★ Association rule learning ○ Apriori ○ Eclat
- 43. import numpy asnp import pandas as pd import sklearn.linear_model # import sklearn.neighbors # Load the data oecd_bli = pd.read_csv(datapath + "oecd_bli_2015.csv", thousands=',') gdp_per_capita = pd.read_csv(datapath + "gdp_per_capita.csv",thousands=',',delimiter='t', encoding='latin1', na_values="n/a") # Prepare the data country_stats = prepare_country_stats(oecd_bli, gdp_per_capita) X = np.c_[country_stats["GDP per capita"]] y = np.c_[country_stats["Life satisfaction"]] # Select a linear model model = sklearn.linear_model.LinearRegression() # model = sklearn.neighbors.KNeighborsRegressor(n_neighbors=3) # Train the model model.fit(X, y) # Make a prediction for Cyprus X_new = [[22587]] # Cyprus' GDP per capita print(model.predict(X_new)) # outputs [[ 5.96242338]] Train a linear model using Scikit-Learn 43
- 44.
- 45.
- 46. No Free LunchTheorem 46
- 47.