I just moved to Paris from Germany, and I find the flats-renting market quite different and more complicate. Thus, in this series of notebooks I want to get a quantitative visualization of the situation.
The idea is to scrap some data from websites to measure the obvious: eg. price vs district; and the less obvious: e.g. a non-linear size vs price/sqm relation, or lack of relations with respect to cases in which one would expect a correlation.
The following data is for furnished, 2 rooms flats, with minimum size of 25m2, within Paris only.
Figure 1: Histogram of 2 rooms flats per price and per arrondissement (left) and per binned size and per arrondissement (right).
*Figure 2: Two visualizations of the price per square meter (€/m2) versus flat sizes, averaged over arrondissement. Blue: unfournished, Green: fournished.
Figure 3: Visualization of the price per square meter versus arrondissement and flat size. Red dots: unfurnished. Filled dots: furnished.
Figure 4: Energy efficiency of Paris flats. Same color-coding as previous plot. There is no need to comment.
Figure 6: PCA to identify some important features. There is no attempt to do a more robust assessment with a kernel PCA (I don't need it now). Note, however, that the 3 modes shown represent about 97% of the data, so they are enough (in this case)
So, Price correlates with size, which in turns weakly correlates with the arrondissement. One would expect also that the energy level correlates with the price as a proxy for the flat quality but it's not the case. I think that for the choosen features, the data is too noisy; but the current perspective is enough for the moment.
How much does a couple has to earn to afford to live in Paris? Using the rule that a flat can't be more expensive than 1/3 of the net total income of the inhabitants, the average needed income per person per month is about 1900 €/monnth, assuming that the flat is inhabited by 2 persons with equal income.
Figure 7: (Left) Distribution of average income for a couple (green), and per person (assuming 2 persons in the flat, and equal income) (blue). -- (Right) Price distribution of all flats of 2 rooms (blue), and of flats with size between 35m2 and 50m2 (green) -- White lines correspond to the average of the distribution. Binning: 100€.
[Analysis notebook on nbviewer](https://nbviewer.jupyter.org/github/astyonax/machine-learning-paris-flat/blob/master/v2/Visualize on-sight insights.ipynb)
the jupyter notebook as web-page.
TL;DR: the jupyter notebook as web-page.
The second (and current) version is more straightforward than the 1st one thanks to the use of a better data source (Le Bon Coin).
The flats price versus flats features are quite self-explaining by the plots. A heatmap of price per arrondissement is done with the Leaflet ipython extension.
In the first version I took the data from a complicate website, thus we use some creative way to fill in the empty or wrong records.s
JupyterBeautifulsoupPandasScikit-learn- fuzzy string matching
Seaborn