[AI 이노베이션 스퀘어] 기본 저녁반 11 기
- Jupyter Notebook & Markdown
- Python
- 자료형: atomic, container
- string, List, Tuple, Range, Set, Frozenset
- PEP: Naming, Trailing comma
- 할당6가지 : 기본, 복합,unpacking, starred, 증감, global & nonlocal
- 조건문 : if, else, elif, pass, none,
- 삼항 연산자, 논리 연사자(AND,OR), IN, NOT IN
- 반복문 : for, break, continue
- 예외처리문 :Try, except, as, else,finally, assert, raise from,
- Function : 함수 선언/정의, 디폴트값, Signature, docstring, 매개변수 할당하는 7자기
- 함수
- 함수 반환값 ,중첩,closure, higher order func, scope, overloading, 재귀, lambda, *, namespace, callable
- 함수형 프로그래밍 , iterable, enumerate , dis, comprehension, filtering, generator, itertools, map, filter , reduce, 함수: recursion, Higher Order Function
- callable : function ,calss ,class()
- 함수 중첩: closure, LEGB, Higher Order Function , map,filter,reduce Module
- Numpy
- array, ndarray, zeros, ones,full,empty, identity, linespace,logspace
- shape, ndim,size, dtype,reshape, concatnate
- arange ,repeat,reshape ,resize
- Image
- Indexing :Array Indexing / slicing , Boolean Indexing, Fancy Indexing
- 차원 증가 방법: shape, reshape, None, newaxis, expand_dims
- 배열 연산 : broadcasting
- Array 정보 : shape,ndim size ,dtype ,itemsize , data
- Array 생성 :_like( zeros,onex,empty,full_like) fromfunction
- Copy :pyton shallow /deep copy, Numpy shallow/ deep copy, new copy: reshape,resize
- Array 한줄로 변환 flatten, ravel
- Array 분할: split, hsplit,vsplit, unpacking응용
- Universal Function(ufunc)): np.abs, np.sum
- 행렬 연산: np.multiply *,np.dot @
- Array Iterating : flat, ndenmerate, strides
- Numpy Help : info, lookfor
- Scipy
- repr, str
- Array 분할: np.split, np.hsplit,np.vsplit , Array 결합:concatnate,append,hstack,column_stack, vstack,row_stack, stack, r_,c_
- 원소 배열 변환 ; np.swapaxes, T transpose
- Structured Arrays 생성,indexing
- Pandas 통한 분석:읽기:read_csv, 정보: info, 열, 행, 인덱싱,filtering: boolean indexing, Aggregation: groupby, 시각화: line,bar, pie
- Pandas: 1.데이터불러오기 2. 데이터 정보: info, head, tail,sample 3.EDA: Unique,nunique,describe,values_countes, 시각화:pie,bar,boxplot
- descibe의 다형성, 산점도 행렬, Aggregation:groupby 데이터 정렬: sort, sorted, sort_values, sort_index; index 변경: set_index, reset_index
- 데이터 불러오기: read, 데이저 정보: info, 데이터 결합:concat, append, columns :추가, drop, rename , Aggregation: groupby, pivot table, crosstab,
- Tidy Data 연습, 기계학습 예시(KNN)
- 추천 시스템
- README, 데이터 불러오기, 결합: rename, concat, append, merge, 전처리 :drop, indexing, select_dtypes,to_numeric, max_rows
- Reshaping: melt, stack, unstack, pivot ,결측치: fillna, 상관분석: 조건문,apply
- 기계학습(KNN) 맛보기, DESCR,data,feature_names, target
- 결측치 : isnull, isna, fillna,
- 데이터 타입 변환: map, category, astype(object)
- 시간날짜 :calendar, datetime,strptime, strftime; to_datetime, resample, OHLC
- 지도 folium : Map, Marker, CircleMarker,Circle + iterrows,itertuples
- 그래프 matplotlib.pyplot: State Machine, figure,axes,subplots, title,label, legend, grid,ticks, lim, style.use, style.context
- sklearn : 학습: knn.fit, 예측: knn.perdict, 데이터 분할: train_test_split , 검정(정확도): knn.score
- 과정: 데이터 준비, EDA,데이터 분할, 모델 학습, 모델 성능 평가, 모델 비교, 전처리에 따른 성능변화 : scaling ( MinMaxScaler,StandardScaler,RobustScaler , MaxAbsScaler)최적의 성능 : Overfiting, Underfiting, cross-validation(cross_val_score)
- 과정 : 결측치 처리 :fillna, SimpleImputer, KNNImputer, Encoding : Label Encoding (map, LabelEncoder), One-Hot Encoding(get_dummies, OneHotEncoder)
- EDA: 상관관계 :pairplot, corr, heatmap, 모델분류 : Linear Regression, KNeighborsRegressor, DecisionTreeRegressor, 평가척도 : confusion_matrix
- ML 복습, 성능조적 1.Model Selection ,2.Preprocessing : scaler
- Classification Metric : confusion_matrix,
- classification_report: Accuracy,Precision, Recall, F-1
- Titanic : Encodng, 범주화 : cut, qcut, Aggregation
- ML 복습, 이미지 : 이미지 특성, 채널, 이미지 데이터, 전처리,데이터 학습
- 파이프라인: Pipeline, make_pipeline, ColumnTransformer, overloading
- 검정: Holdout( 2-way,3-way), cross_val_score, learning_curve
- 자동화 : Pipeline,GridSearchCV
- feature seleciton: filter, wrapper ,embeded
- 데이터형 변환 : pandas 수치형 변환 :to_numeric, astype; Label Encoding :category, LabelEncoder, OrdinalEncoder, ; One-Hot Encoding :get_dummies,OneHotEncoder
- Dimentional Reduction :PCA
- 분류 예측의 불확실성 추정 : predict_proba, decision funciton,
- Ensemble: Boosting, Bagging, Random Forest, Voting, Stacking
- 비지도학습 : clustering : KMeans, AgglomerativeClustering /w dendrogram,DBSCAN
- Deep Learning: Perceptron : OR, AND , XOR, MLP/ANN , 책추천
- 퍼셉트론 한계점:XOR
- Activation Function: Step,Sigmoid,Tanh,ReLu,
- Graphical representation, 원리:feed forward, back propagation, gradient descent, 문제점:overfitting, Gradient Vanishing
- Tensor Flow: 설치, Numpy vs TF, 연산:add, reduce_sum,Ufunc, broadcasting
- DL 과정: 데이터 , holdout, normalization, 모델 구조, 모덱 학습전략 compile, 학습fit,검정 evaluate
- 모델 생성 :Sequential, Dense, Flaten , 구성요소: (kernel_initializer, bias_initializer, input_shape, activation ), output layer: binary/multi- classification, Regresson
- 학습전략 compile(loss, optimizer, metrics), 학습: fit(epochs, batch_size), 예측: predict
- tf.keras.utils: plot_model. to_categorical
- Callback: overriding,
- Tensorboard: model, 이미지 데이터 , magic commands
- wrappers: KerasClassifier(build_fn),KerasRegressor(build_fn)
- 성능 : Underfitting(high bias), Overfitting(High variance), 방지 전략: 모델 볼륨(노드,레이어 ) 줄이기, EarlyStopping ,Dropout by trainable=False , regulation L1,L2 ,Data Augmentattion,Normalizatoin
- Kears의 Functional Paradigm :callable , Sequential 방식, Model 방식: input(shape), Model(inputs, outputs), Concatnate()
- layers(weights, variables), __init__, __call__, build(),call()
- 학습 방법 4가지 : 1) train, 2) fit( batch,epochs, verbose, callbacks, validation_split, class_weight, initial_epoch, workers, trainable) , 3)train_on_batch , 4) GradientTape
- Gradient Descent : Algorithms (step 방향, step size), GradientTape(watch_accessed_variables, persistent), None Gradient
- 시각 CNN : CNN의 재부활, DNN과 비교 , 합성곱,padding, Feature Map: stride, kernel, pooling, CNN 실습:Conv2D, MaxPool2D
- 언어 : Text Classification : unique,join, get, sequence 처리 :pad_sequences , Embedding, GlobalAveragePooling1D, Imbalanced Data
- tensorflow_datasets :load(split, shuffle_files, as_supervised. with_info), batch, Numpy dataset vs TF dataset : iterable, 성능 최적화: Cache, Prefetch(AUTOTUNE)
- Transfer Learning by tenseorflow_hub, keras_layer(dtype, trainable, inputshape)
