An optimizer is a powerful tool to harness in DFS play, capable of bulk operations in an instant that would otherwise take hours of manipulation to perform manually. Like any other tool, one must first learn how to wield an optimizer before its true power can be realized. That is exactly what we will look to sort through in this weekly series.
We’ll focus on the Contrarian Edge Optimizer to use for Monday Night Football this season to explore the tool itself, maximize expected value through optimal utilization and provide a sneak peek into the newest, and most dynamic optimizer in the industry.
Get Projection CSV
Before manipulating any of the settings in the optimizer, I first like to export the projections via a comma-separated values document, which provides all the raw projections used by the optimizer for each player on the slate. Before continuing, it is important to understand what these values represent.
Median Projections
By definition, a median represents a projection whose final outcome would land above the projection and below the projection an equal 50 percent of the time. As such, the modeling in these algorithms can be back tested to find reliability indexes and tweak the algorithm to provide further accuracy.
The idea of median is difficult for the human brain to comprehend. We like things simple, direct and to the point – which is what median projections aim to provide. We must realize that these top-level values are a numerical representation of a range of outcomes for each player on a given slate.
This range of outcomes will be different shapes, sizes, and magnitudes for every player and becomes one of the better inputs to manipulate to alter the output from the optimizer. Give it a try! Run the optimizer without manipulating any of the median projections and see what it provides.
Test-Drive Our Optimizer for 7 Days, $1
Then, manipulate just one player’s median to a 60 percent outcome (multiply the raw projection by six and divide by five) and run the optimizer again to see how that changes the output in roster form. Higher on a player on a given slate than the median projections are accounting for? Bump their value in the CSV within their range of outcomes and see how the optimizer responds! And best of all, the directions to complete this step in the process are readily available in the top-level of the optimizer design.
Predictive Analytics Modeling
Predictive analytics utilizes statistical modeling methods to predict future outcomes. In other words, predictive analytics attempts to utilize machine learning algorithms to create predictive models. With the optimizer, the behind-the-scenes work has been done for us, but we can manipulate the outputs by changing inputs as previously discussed.
Variance in Median Projections
The best way to visualize a range of outcomes projection, assuming we are provided with a median projection, is to utilize a bell curve. This bell curve will be situated about the median with an array of potential outcomes. Most bell curves are symmetrical about the median, but some players carry an asymmetrical distribution of values within their broader range. These players, largely considered “low floor-high ceiling” plays, are some of the most difficult to predict and project due to their lopsided array.
Median projections must also account for ambiguity in certain situations. For example, there is significant ambiguity associated with the Buffalo Bills and the expected snap rates for players from the slot. Deonte Harty, Khalil Shakir, Trent Sherfield, and rookie tight end Dalton Kincaid could all see slot usage, but the optimizer must account for these wide ranges of potential outcomes and display it through median projections in numerical form. This introduces significant variance in those projections, something we can look to manipulate to harness in our favor.
In numerical models, these statistical anomalies are best represented through standard deviation – but we can do things to manipulate these players manually in the optimizer.
Auto Ownership Bonus
First, select “Set Pro Options” on the top left of your screen in the optimizer. Next, toggle “Auto Ownership Bonus” in the dropdown menu. This function is used to set an ownership threshold and bonus to encourage the inclusion of players that are less owned, leveraging the variance associated with median projections and ownership values.
This functionality will also help to harness the second major statistical input to the modeling – expected ownership. Since the game of football includes high rates of variance, ownership projections are a valuable input to leverage in the process. These ownership bonus thresholds can be manipulated to increase or decrease exposure to variance.
Week 1 MNF Example
Let’s put these practices in action for Week 1 using the Contrarian Edge Optimizer. We’ll focus on the ambiguity with the expected slot snap rates from the Buffalo Bills, alter inputs, and see the outputs from those deviations. We won’t be able to see the full roster outputs for obvious reasons (the optimizer is a paid tool), but we should be able to conceptualize the effects of these manipulations.
I changed the projection of Deonte Harty to an 80 percent outcome, accounting for the potential for him to see a slot snap rate that is higher than his expectation, imported the new data into the optimizer and ran the simulation without adjusting any other values or manipulating any of the Pro Options.
Deonte Harty returned as the optimal Captain in that run, at 0.6 percent expected ownership. Doing the same for Dalton Kincaid returns him as the optimal Captain. The same can be done in the other direction to account for outcomes below median projection, which is useful for variant acts like injuries and matchup induced outliers.
The first installment in this series was broader and more conceptual than it will be in the future, but hopefully it helped to establish a foundational and working knowledge of the power of the optimizer and how to manipulate top-level statistical inputs. From here on out, we’ll get into the optimizer on a deeper level. Best of luck in Week 1 and we’ll see you at the top of the leaderboards!