This project is based on the Wind-Generated-Power Prediction Hackathon by HackerEarth in 2021. The project has been made using the Pytorch library. The data which was not available were assumed to be having the values of their corresponding column. I have achieved a score of 93.5 using this model uptill now.

The project can be divided into following parts:

1. Preparing the data
2. Model
3. Test Set Predictions

For preparing the data, the Pandas Library has been used extensively, since it does the job of preparing the data in the easiest manner. One hot encoding has been used to replace the text data in the dataset. The characters and their onehotencods can be found here. As mentioned earlier all the missing data has been replaced by the median values of their corresponding column.

For preparing the dataset, I have used the Dataset class imported from torch.utils.data class provided in Pytorch. Since the data is raw so, I have done the custom implementation of the same which can be found here.

Further DataLoader provided in torch.utils.data have been used while training the model.

The model has been made by subclassing the nn.Module class provided in Pytorch. The model consists of 6 layers with 5 ReLU activation layers and 1 Softplus layer. I have trained the model for around 1500 epochs in total and to evaluate the model is not overfitting a dev set has been developed from the train dataset. The graph of loss vs iterations can also be found in above.

The test set predictions can then be made and saved in a csv file.

Everyone is free to use it if right credits are given