multi-label classification problem
Budget: $10 – $30 USD
Here are further details about the project:
Problem statement
————————————
I need to train a deep neural network on a multi-label classification problem. The DNN should output the probability score of an example belonging to each of 40 possible output labels. The network will be trained using PyTorch lightning framework with tensorboard integration to show train validation and test metrics.
Input
————
- CSV file that contains > 300K samples
- The file contains > 200 input features (numeric and categorical) and > 40 binary target variables.
- Some of the features for some samples include null values. This means that the value for that feature and sample combination wasn’t available in the historical data.
Expected deliverable
—————————————
- The notebook used to train the model
- The trained PyTorch model with tensorboard dashboard that shows train/Val metrics over training iterations, as well as performance on test dataset. Metrics should show that the model is not overfitting.
- Inference function: given a pandas dataframe as input, generate the predicted probability scores for each label as a pandas dataframe as well
- Explanation of DNN predictions: using captum library, write a function that takes as input the trained DNN and an example for inference. As output generate the attribution values for each feature using Integrated Gradients and DeepLiftSHAP.
Problem statement
————————————
I need to train a deep neural network on a multi-label classification problem. The DNN should output the probability score of an example belonging to each of 40 possible output labels. The network will be trained using PyTorch lightning framework with tensorboard integration to show train validation and test metrics.
Input
————
- CSV file that contains > 300K samples
- The file contains > 200 input features (numeric and categorical) and > 40 binary target variables.
- Some of the features for some samples include null values. This means that the value for that feature and sample combination wasn’t available in the historical data.
Expected deliverable
—————————————
- The notebook used to train the model
- The trained PyTorch model with tensorboard dashboard that shows train/Val metrics over training iterations, as well as performance on test dataset. Metrics should show that the model is not overfitting.
- Inference function: given a pandas dataframe as input, generate the predicted probability scores for each label as a pandas dataframe as well
- Explanation of DNN predictions: using captum library, write a function that takes as input the trained DNN and an example for inference. As output generate the attribution values for each feature using Integrated Gradients and DeepLiftSHAP.