# Industry-Location Relationship Analysis
Now we want to analyze how port locations correlate with dominant industries using categorical-categorical visualization, while practicing log-scale transformations for skewed distributions.
We will create a heatmap to show the relationships between continents and top industries. Here's some reading to get you familiar with the concept:
* [Heatmaps and correlation plots](https://medium.com/@rgr5882/100-days-of-data-science-day-53-heatmaps-and-correlation-plots-33f7fea60bc0), Ricardo García Ramírez, Medium.com
* [Demystifying heapmaps: A comprehensive beginner's guide](https://datasciencedojo.com/blog/heatmaps/)
***
### **🔧 Your tasks:**
* [ ] First calculate the contingency table.
* [ ] Plot the contingency table as a heatmap. It will likely look like this:
* There is a problem here: The "Mineral Products" industry absolutely dominates all other industries, this prevents us from seeing the relationships of other columns. We can use a log scale to prevent this.
- [ ] Transform the contingency table values to log scale. Now you should have something like this:
* Optionally, to reduce clutter and focus on the most important industries, you can filter the dataset first to only include top 5 industries and then calculate the contingency table based on that. That would result in this:
***
**Interpretation tasks:**
* [ ] 🤔 Do some online research to find an explanation for such dominance of the "Mineral Products" industry in maritime trade.
* [ ] 🤔 Do some online research to find out what real-world factors explain Asia's high "Animal & Animal Products" presence?
>
>
> You can use pandas' `.crosstab()` for computing a contingency table between two variables.
>
> For log scale transformation, you can use numpy's `log1p()`. This calculates the natural logarithm of the value plus 1. (🤔Why plus one?)
>
> You can use `value_counts()` with `nlargest(x)` to filter the top `x` values.
>
> You can use pandas' `.crosstab()` for computing a contingency table between two variables.
>
> For log scale transformation, you can use numpy's `log1p()`. This calculates the natural logarithm of the value plus 1. (🤔Why plus one?)
>
> You can use `value_counts()` with `nlargest(x)` to filter the top `x` values.