Automatic Flavour profile display

[ljouhault]

Checklist

• I have checked that this feature is not already requested
• I have checked that this feature does not already exist in the project

Is your feature request related to a problem? Please describe.
The server already support ingredients profile ( % acidity, %sweetness, ABV ) but does nothing with it.

Describe the solution you'd like
Under each recipe, add a flavor profile graph of the cocktail (either 2-3 bar range visualizer, or a single 2d flavor profile map).

Additional context
This is not as easy as displaying the total % acidity, sweetness and alcohol, as the different component balance each other (perceived acidity depends on real acidity, but also on sweetness).
One would need to calculate:

1. the sweet to sour balance : $$\frac{Sweet}{Acid + \epsilon}$$
2. the Acid/Sugar Masking Index : $$\ln(1 + Acid) \cdot \left( \frac{1}{1 + 0.04 \cdot Sweet} \right)$$
3. "Perceived Booziness" : $$ABV \cdot \exp\left( - \big( 0.025 \cdot Sweet + 0.05 \cdot Acid \big) \right)$$

and maybe other stuff too (even the formulas are guidelines and not ground trough, i'm far from an expert).

[ljouhault]

Another approach might be to only calculate interactive feature and use a simple regression model to discover the perfect formula itself, although this is a lot more empirical as it would require rating manually the perceived sweetness, sourness and booziness of enough recipes to fit the model on.

[ljouhault]

Another approach might be to only calculate interactive feature and use a simple regression model to discover the perfect formula itself, although this is a lot more empirical as it would require rating manually the perceived sweetness, sourness and booziness of enough recipes to fit the model on.

I just made a quick python poc using sklearn and some of my own recipe and the regressive approach is surprisingly accurate. The one miss is for cocktails that are very bitter and/or have a lot of fat (from coconut cream for example). However, adding these two new column (fat content and bitterness) allows a regressive model to predict the perceive flavor profil of a new recipe with a very high accuracy.