ml
Notes from Coursera course
introduction
- Supervised Learning: we have a data set and already know what our correct output should look like
- Regression: Predict value on a curve
- Classification: Predict Discrete value output, usually 1 or 0
- Unsupervised Learning: Expected output unknown.
- No feedback based on the prediction results
- Derive relation/structure by grouping the data into clusters
- e.g.
- Market segmentation
- Social network analysis.
- Non Clustering : Find relation within chaotic environment
- Cocktail Party problem
- Separate out two voices, one closer to mic other away from mic.
- Cocktail Party problem
Model representation
- $X^{(i)}$ represents input data $Y^{(i)}$ represents output value
- These are not exponentiation
- A pair $(x^{(i)}, y^{(i)})$ is called a training example, and the dataset — a list of $m$ training examples $(x^{(i)},y^{(i)});i=1,...,m$ — is called a training set.
- We need to find function
h
so thath(x)
returns good predictedy
h
is calledhypothesis
Cost Function
- Squared Error Function Or Mean Squared Error
$J(θ0,θ1)=2m1i=1∑m(y^i−yi)2=2m1i=1∑m(hθ(xi)−yi)2$