Prediction of prices of apartments in Prague

Instructions for SU2

Working example url is here https://8862-95-103-189-152.eu.ngrok.io/ . Web will run on my machine and will be forwarded to internet by https://ngrok.com/. Therefore you do not need to install dependencies and create environment. Please do not use web simultaneously on more computers/ browser tabs as it can result in unexpected behaviour. Also please be patient as it can take few secs to load all elements.

Web interface looks like this

You can use

• Predikce pomocí URL option

Here please provide url of apartment from Prague from (https://www.sreality.cz/hledani/prodej/byty/praha?strana=1 or bezrealitky.cz) or try whatever link you want it should handle it...

• Predikce pomocí ručně zadaných příznaků

Here expand please Zadej příznaky block and provide some attributes of apartment. Usable area and location are required.

Your result should be similar to this one

Requirements

---

Make sure you first create (e.g. conda) environmnent with either of commands

• conda env create re && conda activate re && pip install -r requirements.txt
• conda env create -f environment.yml

We use python 3.10.6

Run the code

---

There are two options:

• use web interface which can be launched locally using following command
  • streamlit run web.py
• use CLI (for details use python main.py --help) e.g.
  • python main.py --train --tune -> performs hyperparameter search (data loaded from ../data/dataset.csv)
  • python main.py -> runs prediction on data from ../data/dataset.csv and saves result in ../data/result.csv
  • python main.py --train --scrape -> scrapes new data and performs training

There is no jupyter notebook demo as we (or just Many98) do not like Jupyter notebooks 😎

Processing logic is as follows

---

Processing runs in two phases:

• train -> all advertisements are obtained by crawlers
• inference -> advertisement url/data is/are provided by user via web app

Implementation is also implemented in 2 main classes:

• ETL class handles data obtaining and preprocessings
• Model class operates on preprocessed data by ETL and the model itself is based on XGBoost

Graphical proposal of processing logic