Please enable JavaScript.

Coggle requires JavaScript to display documents.

DATA SCIENCE (Machine Learning (DEEP LEARNING (Neural Networks (Loss…

- - - - Binary Class Classification
        
        Confusion Matrix used (TP,FP, FN, FP)
        
        Evaluation Technique - Precision, Recall, F1Score
        
        Targets are in Binary Form
      - Multi-class Classification
        
        Targets are Multi-class 2. Evaluation Used same as Binary 3. Confusion matrix used is with multiple targets
      - Metrics
        
        Precision, Recall, F1 and Fbeta
      - Algortihms
        
        Naive Bayes
        
        SVM
        
        Decision Tree
    - - Metrics
        
        Mean Absolute Error
        
        Mean Squared Error
        
        R2 Score
        
        SSE (Sum of squared error)
        
        Algos in SSE
        
        Ordinary Least Square(inbuilt in sklearn LinearRegression())
        
        GRADIENT DECENT
    - - Bagging - Build in one-go
        
        Random Forest
        
        Decision tree leads to overfitting problem therefore majority of the time Random Forest is only used
        
        Random forest is the group of multiple Decision Trees to which data is given in random fashion
        
        Random forest is the group of multiple Decision Trees to which data is given in random fashion
        
        And incase of regression it is decide on the basis of average
      - Boosting - creates trees in sequences - generate 1 tree then passes the results it to 2nd Tree
        
        Adaboost
        
        Gradient Boost
      - Stacking
  - - - KMeans. -
        
        In the K-means algorithm 'k' represents the number of clusters you have in your dataset. In this video, you saw that a k value of two makes a lot of sense. There is one cluster of points with shorter distances for when I travel to work. A second cluster is created when I travel to my parents' house.
        
        Visually inspecting your data easily shows these two clusters. On the next page, you will have an opportunity to make sure you have this technique for finding clusters mastered.
        
        So far you have identified k when you can visually inspect your data to identify the number of clusters. However, in practice, you often have tons of data with many features. This can make visualizing your clusters impossible.
        
        In that case we use ELBOW METHOD
      - Hierarchical Clustering
        
        Single Link Clustering
        
        Not a part of Scikit Learn therefore we can't use this
        
        Complete Link Clustering
        
        Is available in scikitlean.AggloramativeClustering package
        
        uses 2 methods for
        distance mesuring (farthest point / ward method)
        
        Same as Single Link but has difference distance measuring
      - DBSCAN (Density Based Clustering)
        
        Takes arguments (Epsilon point and Mini
      - Gaussian Mixture Model (GMM) Clustering
      - Clustering Analysis
        
        Clustering Validation
        
        External Indices
        
        ADJUSTED RANDOM INDEX
        
        Internal Validation Indices
        
        SILHOUETTE CO-EFF
        
        FOR DBSCAN Don't use this (READ BELOW PAPER) http://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=83C3BD5E078B1444CB26E243975507E1?doi=10.1.1.707.9034&rep=rep1&type=pdf
    - - PCA.(Principle Component Analysis)
        
        Takes full dataset and reduce it to the parts that only hold most info
        
        Reduces Features while keeping the output same
        
        In general PCA is used to reduce the dimensionality of your data
      - RANDOM PROJECTION
      - Independent Component Analysis (ICA) - ICA assumes that features are mixtures of independent sources and therefore isolates these independent sources and therefore isolates these independent sources completely
  - - - Activation Functions
        Used when Feed Forwading
        
        Hidden Layers.
        
        Relu
        
        PreRelu
        
        Output Layers
        
        Sigmoid (for Binary Problems)
        
        tanh
        
        Softmax (Used for Multiclass Problems)
      - Loss Functions
        Used while Backpropogation
        
        REGRESSION PROBLEM
        
        MSE
        
        MAE
        
        RMSE
        
        CLASSIFICATION PROBLEM
        
        Categorical cross entropy
        
        Binary cross entropy
        
        Sparse cross entropy
        
        Reducing Loss functions [OPTIMIZERS]
        
        Gradient Descent
        
        SGD
        
        Mini Batch SGD
        
        SGD with Momentum
        
        Adagrad
        
        adadelta and RMSprop
        
        Adam [BEST]
        ( made of RMSprop + Momentum )
      - Weight Inititializing Techniques
        
        Uniform weights
        
        Xavier Glorat (Use Sigmoid / tanh)
        
        He init (Use ReLu)
- - - - Central Tendency(Mean,Median and Mode
        )
      - Spread
        a . Find the 5 Number Spread policy (Range, IQR, Min, Max)
        b . If you don't want to show the spread with 5 Number policy just find the STANDARD DEVIATION which will show the spread
      - SHAPE
        There can be 3 Types of Distribution.
        
        A. Left Skiwed
        B. Right Skewed
        C. Symmetric (Normal Distribution)
      - OUTLIERS
        Below are my guidelines for working with any column (random variable) in your dataset.
        
        1. Plot your data to identify if you have outliers.
        2. Handle outliers accordingly via the methods above.
        3. If no outliers and your data follow a normal distribution - use the mean and standard deviation to describe your dataset, and report that the data are normally distributed.
        
        4. If you have skewed data or outliers, use the five number summary to summarize your data and report the outliers.