What is Unsupervised Learning?
It is a learning model where it is trained ONLY on the inputs.
REMEMBER: Supervised learning uses input-output pairs to parameterize the model for improvement.
- Ex of Unsupervised Learning: Music recommendations by using
clusters
of users that have similar music taste.
What is Clustering?
No classification exists to determine data clusters. Therefore, it is a challenge to separate data in their respective classes/groups.
Differences between CLustering and Classification:
Clustering | Classification |
---|---|
No Labels (Unsupervised) | Labels/Classes (Supervised) |
Determine # of Clusters is part of task | Training data to determine # of classes. |
Dimensionality Reduction w/ Principal Compinents Analysis (PCA):
Linearly transforms input features by removing the correlated features w/ the objection of maximizing the variance
OG Dataset ->PCA-> Dataset represented w/ components -> Choose the PCs that explain a certain variability in the data -> Apply the ML Algo
- PCA is a “feature extraction” method, because the original input features are transformed and not included.
- A “feature selection” method selects a subset of input features that explain a certain percentage of variability.
- The input features that are not selected are eliminated (feature elimination).
REMEMBER: A lot of input features does not necessarily mean they are useful!
Input/Output Types
- Binary (Categorical w/ two possible values)
- Categorical (W/ more than 2 possible values)
- Real-valued (continuous)
Examples of each type:
- Real-valued: 7, 9, -3.3, 10001, …
- Binary: {0,1}, {yes,no}, {pass,fail}, …
- Categorical: {‘Panda’, ‘Cat’, ‘Lion’}, {‘Sedan’, ‘Sport’, ‘Truck’}, …
We cannot use single integers to represent classes because the numbers have a natural order.
Two Interpretations
- Frequentist Probabilities represent long run frequencies of events.
- Ex: If we flip a coin many times, we expect half of those flips to land on heads.
- Bayesian Probability is used to quantify our uncertainty about something.
- Ex: Probability a coin lands on heads is 0.5. We can also believe that it will equally likely to land on heads or tails on the next toss.