- Regression vs Classification (Identifying the problem type)
- Regression - when theres an range of labels with infinite values
- Classification - when the labels are discrete and limited
- Features vs Labels
- Test vs Training sets of data
- Feature normalization
- Feature selection
- Understanding datastructures
- Using Lodash is extremely slow and not Numbers focused. Also, some things are harder to do in Lodash.
- Tensorflow JS has an API similar to Lodash.
- It is extremely fast for numeric calculations.
- Learn some fundamentals around Tensorflow.
- Go through some exercises on Tensorflow.
- Implement already implemented algorithms in Tensorflow for practice.
- Try to build other algorithms with Tensorflow.
Some common terms:
-
Tensors - They are JS objects that wraps an array of numbers.
-
Dimensions - How many directions a tensor extends.
-
Shape(property) - An array of the length of each Dimension of a Tensor (Visualize .length property from top level)
-
Shapes are always interpreted as [Rows, Columns]
- We try to figure out a relationship between independent and dependent variables.
- It's faster than K-Nearest Neighbor, as you train it once and then can use to to predict values.
- It uses methods which are important in much more complicated Machine Learning practices.
- Gradient descent is a way to implement Linear Regression
- MSE of (Mean Squarred Error) should be as low as possible.