Predict PM2.5
Problem description:
Used a dataset from Central Weather Bureau in Taiwan to predict the value of PM2.5 at a specific station
Model
Chose PM2.5, SO2 and PM10 as features and trained this model with Adagrad
Adagrad
- Divide the learning rate of each parameter by the root mean square of its previous derivatives
- $w^{t+1}\leftarrow w^t-\frac{\eta^t}{\sigma^t}g^t$ , $\sigma:$ a root mean square of the previous derivatives of parameter $w$
- $g^t=\frac{\partial L(\theta^t)}{\partial w}$
- $\eta^t=\frac{\eta}{\sqrt{t+1}}$
Result
- Achieved Top 210/445 ($48\%$) rank in the Kaggle competition