Today I've worked on 2 main sides:
Theory:
- Took notes of Classification in general
- Understanding of why in Logistic Regression there is no possibility to find easily the weights as with Normal Equation in Linear Regression. Instead, it has to be done by using Partial Derivatives of the Cost function
- By using gradient descent with the cost function, then the global minima can be found.
Credit Score Project:
- Data cleaning and understanding of the data and how features are distributed. I found a long-tailed distribution of a feature that was mainly due to an error on the data itself.
- Cleaned data by removing correlated features and also worked on missing values
- It took me a long time, but finally got a grasp on tools, and what to look for when performing EDA.
Theory:
- Took notes of Classification in general
- Understanding of why in Logistic Regression there is no possibility to find easily the weights as with Normal Equation in Linear Regression. Instead, it has to be done by using Partial Derivatives of the Cost function
- By using gradient descent with the cost function, then the global minima can be found.
Credit Score Project:
- Data cleaning and understanding of the data and how features are distributed. I found a long-tailed distribution of a feature that was mainly due to an error on the data itself.
- Cleaned data by removing correlated features and also worked on missing values
- It took me a long time, but finally got a grasp on tools, and what to look for when performing EDA.