0 converted to small double/float when used in a power function #775
Conversation
|
Hi @sstein93 You are right. This is about avoiding You can turn this off by |
|
I have thought about refactoring this code for a while. In theory, different models should be generated based on if you want this protection or not (since it adds a bit of overhead). Then the powers and the like could probably be only fixed if there is an issue (though that would take a bit more overhead...) |
|
Thank you @mattfidler! |
|
@mattfidler I just tried the example above and adding Do you know if this is due to a specific version of an R or C dependency? Thanks, Here is my sessionInfo in case that is helpful: |
|
This is likely based on the version of |
|
Hi @sstein93, The pull request this is being changed into probably behaves a bit library(rxode2)
#> rxode2 2.1.3.9000 using 8 threads (see ?getRxThreads)
#> no cache: create with `rxCreateCache()`
test_rxode_0 <- rxode2({
fn(0) <- 0
is_0_0 <- 0
check_0_exp_1.2 <- 0^1.2
check_near0_exp_1.2 <- (1.490116e-08)^1.2
check_0_exp_0.67 <- 0^0.67
check_near0_exp_0.67 <- (1.490116e-08)^0.67
check_0_exp_1 <- 0^1
check_near0_exp_near0 <- (1.490116e-08)^(1.490116e-08)
check_near0_exp_0 <- (1.490116e-08)^0
check_0_exp_0 <- 0^0
check_0_exp_near0 <- 0^(1.490116e-08)
check_0_div_0 <- 0/0
check_0_plus_0 <- 0 + 0
d/dt(fn) <- 1
})
test_events <- et(seq(0,10, by = 1))
test_sim <- rxSolve(test_rxode_0, events = test_events, returnType="data.frame",
safeZero=TRUE, safePow=TRUE, safeLog=TRUE, useStdPow=FALSE)
print(test_sim)
#> time check_0_exp_1.2 check_near0_exp_1.2 check_0_exp_0.67
#> 1 0 0 4.053817e-10 0
#> 2 1 0 4.053817e-10 0
#> 3 2 0 4.053817e-10 0
#> 4 3 0 4.053817e-10 0
#> 5 4 0 4.053817e-10 0
#> 6 5 0 4.053817e-10 0
#> 7 6 0 4.053817e-10 0
#> 8 7 0 4.053817e-10 0
#> 9 8 0 4.053817e-10 0
#> 10 9 0 4.053817e-10 0
#> 11 10 0 4.053817e-10 0
#> check_near0_exp_0.67 check_0_exp_1 check_near0_exp_near0 check_near0_exp_0
#> 1 5.702397e-06 0 0.9999997 1
#> 2 5.702397e-06 0 0.9999997 1
#> 3 5.702397e-06 0 0.9999997 1
#> 4 5.702397e-06 0 0.9999997 1
#> 5 5.702397e-06 0 0.9999997 1
#> 6 5.702397e-06 0 0.9999997 1
#> 7 5.702397e-06 0 0.9999997 1
#> 8 5.702397e-06 0 0.9999997 1
#> 9 5.702397e-06 0 0.9999997 1
#> 10 5.702397e-06 0 0.9999997 1
#> 11 5.702397e-06 0 0.9999997 1
#> check_0_exp_0 check_0_exp_near0 check_0_div_0 check_0_plus_0 fn
#> 1 1 0 0 0 0
#> 2 1 0 0 0 1
#> 3 1 0 0 0 2
#> 4 1 0 0 0 3
#> 5 1 0 0 0 4
#> 6 1 0 0 0 5
#> 7 1 0 0 0 6
#> 8 1 0 0 0 7
#> 9 1 0 0 0 8
#> 10 1 0 0 0 9
#> 11 1 0 0 0 10
test_sim <- rxSolve(test_rxode_0, events = test_events, returnType="data.frame",
safeZero=FALSE, safePow=FALSE, safeLog=FALSE, useStdPow=TRUE)
# This tests some of the differences between the solves:
m <- rxode2({
nan0 = 0/0
nanLog = log(-1)
nanPow = 0^-1
rPow = 0^-1.5
})
et <- et(1)
# No safe zero in place, use C's pow() function
rxSolve(m, et, safeZero=FALSE, safePow=FALSE, safeLog=FALSE, useStdPow = TRUE, returnType="data.frame")
#> time nan0 nanLog nanPow rPow
#> 1 1 NaN NaN Inf Inf
# No safe zero in place, use R's pow() function
rxSolve(m, et, safeZero=FALSE, safePow=FALSE, safeLog=FALSE, useStdPow = FALSE, returnType="data.frame")
#> time nan0 nanLog nanPow rPow
#> 1 1 NaN NaN Inf Inf
# Safe zero for everything
rxSolve(m, et, safeZero=TRUE, safePow=TRUE, safeLog=TRUE, useStdPow = FALSE, returnType="data.frame")
#> time nan0 nanLog nanPow rPow
#> 1 1 0 -36.04365 4.5036e+15 3.022315e+23Created on 2024-08-28 with reprex v2.1.1 The rules are updated to:
If this successfully passes the checks, I will merge into the main I also plan to release at the end of September 2024 due to a CRAN |
|
Thanks for reporting @sstein93 |
|
Thanks @mattfidler !! |
|
No problem. |
Hello,
I found that 0 is converted to a small double (1.490116e-08) when used in the base of a power function inside an rxode2 model. This has caused large effects in my model. Particularly, when the amount of drug is 0 in the central compartment, I still see the "effect" compartment change, even though the derivative should be exactly 0 when the central compartment concentration is exactly 0. However, when I run the model with no drug added (i.e. central compartment is always exactly 0) the drug effect compartment changes 10-15% over the simulated 30 hours.
Here is a reproducible example where I've set some variables inside the rxode2 model definition. All of the variables that are 0^x should be 0 in the model results, however you can see in the output below that they are not.
Thanks!
Shayna