Optimal Stopping: Lingerie Selection Strategy

The Fitting Room Dilemma: A Problem of When to Stop

Imagine the scenario: you have 15 lingerie sets lined up for a model to try. She tries them on one by one. After each tryout, you must make a decision: "This is the one," ending the process, or "Let's see the next one," knowing you cannot choose the rejected piece later. This situation has several key characteristics.

As the video explains, our fitting room problem aligns perfectly with the conditions of an optimal stopping problem:

• You have a known number of options, $n$ (e.g., 15 lingerie pieces).
• The pieces are tried in a random sequence.
• You can rank the pieces you've seen so far, but you don't know the absolute quality of the entire set in advance.
• Your decision to select or reject is immediate and irrevocable.
• Your goal is to maximize the probability of selecting the single best piece.

Simply picking a piece at random gives you a meager $1/n$ chance of success. Fortunately, a much better strategy exists.

The Optimal Strategy: Explore, then Select

The solution to the optimal stopping problem is elegant and involves a two-phase approach. Given your background in operations research, you'll recognize this as a type of sequential decision strategy under uncertainty.

The strategy is as follows:

1. The "Explore" Phase: First, you observe and evaluate a set number of initial candidates, say $k$, without making a selection. You reject all of them, even if one seems perfect. The purpose of this phase is purely informational: to establish a benchmark for what a "good" lingerie piece looks like in this specific collection.

2. The "Select" Phase: After rejecting the first $k$ pieces, you continue to the next ones. You then select the very first piece that is better than the best one you saw in the "Explore" phase. If no such piece appears by the end, you are obligated to choose the last one ($n$).

This process is visualized perfectly in the image below, showing the "Explore" and "Select" zones.

The crucial question is, how do you choose the optimal number of pieces to scout, $k$? The mathematical answer converges on a famous constant. For a large number of candidates $n$, the optimal number to scout is approximately $n/e$, where $e\approx 2.71828$. This means you should explore roughly the first 37% of your options.

Amazingly, by following this rule, the maximum probability of successfully picking the single best candidate also converges to $1/e$, or about 37%.

The Mathematical Intuition: Why $e$?

Given your statistical background, you might find the appearance of $e$ intriguing. It's not just a coincidence; it arises from the fundamental trade-off at the heart of the problem.

• If you explore too little (small $k$), your benchmark will be low, and you risk committing too early to a mediocre piece that just happens to be the best of a poor initial sample.
• If you explore too much (large $k$), you increase the risk that the absolute best piece was in your exploration sample, which you were forced to reject.

The $1/e$ rule perfectly balances these two risks.

The plot below shows this optimization visually. The probability of success, $P$, is plotted against the number of candidates, $n$. The blue crosses show the ratio $k/n$ for the optimal scouting number $k$. You can see that this ratio quickly approaches the dashed line at $1/e$, and the probability of success (red dots) hovers around 37%.

For a small number of candidates, like in our fitting room scenario, we don't have to rely on the approximation. An exact optimal value for $k$ can be calculated, though the $n/e$ heuristic remains remarkably accurate. For example, for $n=9$, the exact optimal number to scout is $k=3$. The heuristic gives $9/e\approx 3.3$, which rounds to 3.

Numerical Exercise: The Final Fitting

Now, let's apply this to your project. You have successfully curated a final selection of 15 lingerie pieces for the photoshoot. The model will try them on one by one in a random sequence. You must make an irrevocable select/reject decision after each tryout.

Your Task:
Formulate this as an optimal stopping problem and determine your precise strategy.

1. How many lingerie pieces should you have the model try on for the "explore" phase to establish your benchmark?
2. Clearly state your decision rule for the "select" phase.
3. What is your approximate probability of selecting the single best piece for the photoshoot if you follow this strategy?

Take a moment to calculate the value and formulate the strategy before proceeding to the solution.

Solution

1. Determine the "Explore" Phase Size (k):

• Total number of pieces, $n=15$.
• The optimal number of pieces to scout is $k\approx n/e$.
• $k\approx 15/2.71828\approx 5.518$.
• We round this to the nearest integer, giving $k=6$.

So, your "explore" phase consists of the first 6 lingerie pieces.

2. State the Decision Rule:
Your strategy should be as follows:

• Step 1 (Explore): Have the model try on the first 6 pieces. Reject all of them, regardless of how good they seem. Mentally note the appeal of the single best piece from this group of 6. This becomes your benchmark.
• Step 2 (Select): Starting with the 7th piece, continue the tryouts. Select the very first piece that you rank as being better than your benchmark.
• Step 3 (Contingency): If you get to the 15th and final piece, and it is not better than your benchmark, you are obligated to select it.

3. Calculate the Probability of Success:
By following this optimal strategy, your probability of successfully selecting the single best piece from the 15 is approximately $1/e\approx 36.8\%$. While this may not seem like a guarantee, it is mathematically proven to be the highest possible success rate given the constraints of the problem—far superior to the $1/15\approx 6.7\%$ chance from a random guess.

Conclusion

In this final lesson of our advanced optimization module, you've learned to apply the principles of optimal stopping to the critical decision of finalizing a look for a photoshoot. This powerful yet simple rule transforms a situation of high uncertainty into a structured process with a maximized probability of success.

Key Takeaways:

• Sequential, irrevocable decision-making under uncertainty can be modeled as an Optimal Stopping Problem.
• The optimal strategy involves two phases: exploring a subset of options to set a benchmark, then selecting the first option that surpasses it.
• The optimal number of options to explore is approximately $n/e$, or the first 37% of the sequence.
• This strategy maximizes your chance of picking the single best item, with a success probability of about 37%.

You have now completed the key topics for your lingerie design project, moving from data cleaning and trend analysis to predictive modeling and, finally, to advanced optimization for curation and selection. You are now equipped with a robust statistical toolkit to navigate the entire process, from Trend Analysis (correlation, regression), through Style Optimization (clustering, collaborative filtering), to Fitting Room Experimentation (ANOVA, optimal stopping).