Fair Performance Comparison of Simple Postflop and PioSolver

Go back

Follow us on Twitter

Follow us

The test was performed on the 25 BB HU limp spot, board Js 8c 4c.

Input ranges:

Post image

Screen of input ranges from Simple Postflop Solver

Configuration of bets sizings:

Post image

NB: in Pio and SPF the functionality of rounding bets to All-in works differently.

In SPF bet is compared to the stack that is left in the beginning of the street, in Pio bet comparing to the starting stack in the hand:

Post image

generated tree in SPF contains bet 12bb on the river cause 12/19<67%

Post image

generated tree in Pio doesn't contain bet 12bb on the river cause 17/24>67%

We adjusted the tree in SPF so that it absolutely corresponded to the tree in Pio.

SPF treehttps://drive.google.com/file/d/1n0mRsFYoacRVvqxBGzlo2-f8ShY4DXN6/view?usp=sharing

PioSolver treehttps://docs.google.com/document/d/175YuL3ocd1XIXQxs8Il4xZTOPiQ7xPT5RbAd2am_CFs/edit?usp=sharing

Performance comparison was run on the same server with the same conditions (server with an Intel® Xeon® processor W-2145 Octa-Core Skylake W)

Raw data of the experiment resultshttps://drive.google.com/file/d/19-Peewr0EltxBKUVkLJjyG7OoQnHpgpv/view?usp=sharing

PioSolver has 1 algorithm for solving tree, Simple Postflop Solver has 3 different algorithms (A1, A2, A3) and SPF A1 algorithm could use different degrees of tree compression in the RAM.

The results of the experiment on the chart (on the X axis - time in minutes, on the Y axis - exploitability of solution in blinds per 1 hand):

Post image

logarithm applied to the scale of vertical axis

From the chart we can make the following Conclusions: the A3 SPF algorithm using the same time for solving tree as the PIO, shows the accuracy of the solution 2-3 times better. Algorithm A2 SPF performed better convergence, algorithms A1 and A1 with compression x2 performed a result similar to Pio.

Chart for the required RAM per calculation:

Post image

From this graph we can conclude that the SPF has an algorithm (A1 compression x4), which allows calculating a complex tree with fairly good accuracy, using a small amount of RAM.

PS: to perform calculations in PioSolver using more than 16 cores, the user needs to purchase a license for $1099*, while Simple Postflop allows to use any number of cores in the* $299 version.