Please enable JavaScript.

Coggle requires JavaScript to display documents.

machine learning, Data analytics, AI (types of AI, search algorithms,…

- - - - .
        
        ~we usually see regression and classification problems.
        ~for classification we usually use RandomForestclasiffier, because it giver better success in test data.
        ~for regression we usually use RandomForestRegressor .
        
        example: search the text in the link
        
        2. Choosing the right estimator/algorithm for your problem
    - - Splitting the data into features (usually X) and labels (usually y)
        
        Filling (also called imputing) or disregarding missing values
        
        Converting non-numerical values to numerical values (also call feature encoding)
        
        3-->
        ~from sklearn.preprocessing import OneHotEncoder |
        ~from sklearn.compose import ColumnTransformer |
        ~transformer = ColumnTransformer([("one_hot",one_hot, categorical_features)],remainder="passthrough")
        ~transformed_X = transformer.fit_transform(car_sales_missing)
        
        2-->
        ~drop label column Nan rows
        ~from sklearn.impute import SimpleImputer
        ~from sklearn.compose import ColumnTransformer
        ~cat_imputer = SimpleImputer(strategy="constant", fill_value="missing")
        num_imputer = SimpleImputer(strategy="mean")
        //Fill categorical values with 'missing' & numerical with mean
        ~categorical_features = ["Make", "Colour"]
        numerical_feature = ["Odometer (KM)"]
        //Define different column features
        ~imputer = ColumnTransformer([("cat_imputer", cat_imputer, categorical_features),
        ("num_imputer", num_imputer, numerical_feature)])
        //Fill train and test values separately
        ~filled_X_train = imputer.fit_transform(X_train)
        ~filled_X_test = imputer.transform(X_test)
        
        1-->
        //Split into X & y and train/test
        ~X = car_sales.drop("Price", axis=1)
        ~y = car_sales["Price"]
        ~X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)
    - - ~ X_train, X_test are feature data train 80% and test 20%
        ~ y_train, y_test are label data train 80% and test 20%
        ~ clf.fit(X_train, y_train) //training our model //Calling the fit() method will cause the machine learning algorithm to attempt to find patterns between X and y. Or if there's no y, it'll only find the patterns within X.
      - prediction can be done using two functions
        1.model.predict(test_data) # o/p--> [0,1,0,1,1....]
        2.model.predict_proba(test_data) #o/p-->[[0.83,0.0.17],[0.15,0.85].....] this give probablity
    - - there 3 types for evaluation
        1.Estimator score method #model.score(test_data,test_label)
        2.The scoring parameter
        3.Problem-specific metric functions.
        
        steps
        ~from sklearn.model_selection import cross_val_score
        ~cross_val_score(clf, X, y, cv=5)
        
        why we are using cross validator.
  - - - other names:
        X:Data, Feature variables , Features
        y :Labels, Target variable
- - - - without pipeline
        
        Splitting Data
        
        Proper data
        
        Improper data
        
        REMOVE MISSING VALUES
        
        IMPORT LIBRARY
        a. from sklearn.compose import ColumnTransformer
        b. from sklearn.impute import SimpleImputer // filling missing values
        
        instantiating imputer variable
        a. cat_imputer = SimpleImputer(strategy="constant", fill_value="missing")
        strategy ~ constant , median, mean, frequent
        
        using column transformer transforming data
        a. imputer=ColumnTransformer([('name',cat_imputer,['make','color']),() ]
        (name of imputer, object, column to apply)
        b. filled_X_train = imputer.fit_transform(X_train)
        filled_X_test = imputer.transform(X_test)
        
        CONVERTING COLUMN TO NUMERICAL
        
        IMPORT LIBRARY
        a. from sklearn.preprocessing import OneHotEncoder
        b. from sklearn.compose import ColumnTransformer
        
        COVERTINNG TO NUMERIC
        a. onehotencoder= OneHotEncoder()
        
        COLUMN TRANSFORMER
        a. trans=ColumnTransformer(['name',one_hot_obj,['make , 'color'])],remainder="passthrough")
        b. trans.fit_trandform(X_train)
        trans.tranform(X_test)
        
        1 more item...
      - with pipeline
        
        improper data
        
        PIPELINING MISSING AND ENCODING DATA AT ONE USING PIPELINE
        
        STARTING WITH PIPELINE OBJ TO CONVERT IMPROPER DATA
        a. converter_pipe=pipeline(steps=list of tupple)
        [("imp",SimpleImputer(strategy='constant', fill_value=4) ),
        ("enco",Encoder() )
        ]
        (name,object )
        
        NOW PIPELINEING COLUMN TRANSFORMER
        a. prepro=CoulmnTranformer(transformers=list of tupples)
        [ ( 'cat_trans_name', converter_pipe, ['make',color'] )
        ]
        (name,object,columns to apply)
        
        now data is ready for modeling
        
        USING RidgeRegression MODEL
        
        APPLYING IN MODEL
        a. ridge_model=pipeline(steps=list of tupple)
        [ ( "preprocessors", prepro )
        ("redge_model", Ridge() )
        ]
        
        EVALUATION OF MODEL
        
        DIFFERENT METRICS
        a. R^2 (pronounced r-squared)
        b. Mean absolute error (MAE)
        c. Mean squared error (MSE)
        
        1 more item...
        
        proper data
- - - - LinearSVC and SVC from svm
        
        RandomForestClassifier from ensemble
        
        KNeighborsClassifier from neighbors
        
        LogisticRegression from linear_model
    - - linear regression
  - - - K-means
        
        hierarchal
    - - apriori
        
        association rules
- - - - PROBLEMS
        
        Plateau
        
        Local Maximum
        
        Ridges