Data Workshop with Vladimir (#6)
Let me invite you to Data Workshop #6.
Note: if it's your first time... you're always welcome :).
You can check how it was on previous workshops:
Motivation
https://www.kaggle.com/c/allstate-claims-severity
Allstate is currently developing automated methods for predicting the cost and hence the severity of claims. The goal is predict for each ID in test set the loss of value.
To solve this challenge we will use neural network and XGBoost. For this session will focus more on neural network (and if the time lets us, we'll show how to improve the result using both).
What will you learn?
- Introduction to Neural Network (by solving MNIST)
- Build your (possibly first) network, with fully connected layers
- Some more advanced techniques, e.g. Dropout (help to avoid overfitting) or PReLU (other activation function, which get better score than ReLU)
- Tuning hyperparameters for Neural Networks
- Ensembling with XGBoost
About speaker
Vladimir likes traveling... also in the IT world. He worked in different areas in IT (with different technologies). Last 3-4 years he spends his time on learning and getting insights from the data. He was involved in building infrastructure for Big Data, he prepared an ETL (based on Hadoop stuff) and he made data prediction (sales forecasting) and many others. He learns from different MOOCs (Coursera, Udacity, edX and so on), books and he regularly participated in Kaggle's competitions. He loves (data) challenges.
Prerequisites
[1] Auto-magic via docker - https://github.com/dataworkshop/environment
[2] Half-manual. The package manager “conda" - install through the Miniconda installer (http://conda.pydata.org/miniconda.html) or the Anaconda installer (https://www.continuum.io/downloads) + seaborn, keras, theano, xgboost
[3] Totally manually. Install manually those packages: ipython, scikit-learn, pandas, matplotlib, seaborn, keras, theano, xgboost
Use this script to verify your environment: https://github.com/dataworkshop/prerequisite
Please bring your laptop with you.
Please come an hour before if you need help with setting up the environment!
HOW TO GET TO THE MEETING?
By public transportation
You can reach our office by tram 4, 5, 9, 10, 52 or 72. The nearest stop is 'AWF'
and have a walk along Politechnika Krakowska buildings (Życzkowskiego street). Avia building is located at the end of this small street. Remember that total travel time from the city center may take around 30 minutes.