fabian-s
commented
1 month ago Thanks, this already mostly works, but i see at least 3 major issues (some of them may have been present before as well...)
library(tf)
#>
#> Attaching package: 'tf'
#> The following objects are masked from 'package:stats':
#>
#> sd, var
x <- tf_rgp(3)
x_i <- tf_sparsify(x)
y <- tfb(x)
#> Percentage of input data variability preserved in basis representation
#> (per functional observation, approximate):
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> 99.90 99.90 99.90 99.93 99.95 100.00
y_pc <- tfb_fpc(x)
# 1. tfd-tfb ops don't seem to dispatch to the correct method / don't generate the right exception
x + y
#> Error in UseMethod("vec_arith.tfd", y): no applicable method for 'vec_arith.tfd' applied to an object of class "c('tfb_spline', 'tfb', 'tf', 'vctrs_vctr', 'list')"
tf:::vec_arith.tfd("+", x, y)
#> Error in UseMethod("vec_arith.tfd", y): no applicable method for 'vec_arith.tfd' applied to an object of class "c('tfb_spline', 'tfb', 'tf', 'vctrs_vctr', 'list')"
# why does the above not dispatch to this below ?!?
tf:::vec_arith.tfd.default("+", x, y)
#> Error in `tf:::vec_arith.tfd.default()`:
#> ! <tfd_reg> + <tfb_spline> is not permitted
# 2. numeric-fun ops should not cast tfb_fpc to tfb_spline
class(y_pc + y_pc)
#> [1] "tfb_fpc" "tfb" "tf" "vctrs_vctr" "list"
class(2 * y_pc) #!!!
#> [1] "tfb_spline" "tfb" "tf" "vctrs_vctr" "list"
# 3. ops for tfb should preserve attributes of the original tfb
# (ALWAYS for numeric-fun-ops, at least)
tf:::compare_tf_attribs(y, 2 * y)
#> domain basis basis_label basis_args basis_matrix arg
#> TRUE TRUE FALSE FALSE TRUE TRUE
#> family class
#> TRUE TRUE
tf:::compare_tf_attribs(y / y, y)
#> Warning: Fit captures <50% of input data variability for at least one function
#> -- consider increasing no. of basis functions 'k' or decreasing penalization.
#> Percentage of input data variability preserved in basis representation
#> (per functional observation, approximate):
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> -Inf -Inf -Inf -Inf -Inf -Inf
#> domain basis basis_label basis_args basis_matrix arg
#> TRUE TRUE FALSE FALSE TRUE TRUE
#> family class
#> TRUE TRUE
y + 2 * y
#> Error in vec_arith.tfb.tfb("+", e1, e2): all(compare_tf_attribs(x, y)) is not TRUE
# minor: better error messages for things like this
x + x_i
#> Error in fun_op(op, x, y): isTRUE(all.equal(tf_arg(x), tf_arg(y), check.attributes = FALSE)) is not TRUE
x + x == 2 * x
#> 1 2 3
#> TRUE TRUE TRUE
x * x == x^2
#> 1 2 3
#> TRUE TRUE TRUE
all( (tf_evaluations(x / x) |> unlist()) == 1)
#> [1] TRUE
tf:::compare_tf_attribs(x / x, x)
#> arg domain evaluator evaluator_name class
#> TRUE TRUE TRUE TRUE TRUE
x_i + x_i == 2 * x_i
#> 1 2 3
#> TRUE TRUE TRUE
x_i * x_i == x_i^2
#> 1 2 3
#> TRUE TRUE TRUE
all( (tf_evaluations(x_i / x_i) |> unlist()) == 1)
#> [1] TRUE
tf:::compare_tf_attribs(x_i / x_i, x_i)
#> arg domain evaluator evaluator_name class
#> TRUE TRUE TRUE TRUE TRUE
#-----------------------------------
y * y == y^2
#> 1 2 3
#> TRUE TRUE TRUE
all.equal((tf_evaluations(y / y) |> unlist()),
rep(1, length(y)*length(tf_arg(y))), check.attributes = FALSE)
#> Warning: Fit captures <50% of input data variability for at least one function
#> -- consider increasing no. of basis functions 'k' or decreasing penalization.
#> Percentage of input data variability preserved in basis representation
#> (per functional observation, approximate):
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> -Inf -Inf -Inf -Inf -Inf -Inf
#> [1] TRUE
tf:::compare_tf_attribs(y / y, y) #!!
#> Warning: Fit captures <50% of input data variability for at least one function
#> -- consider increasing no. of basis functions 'k' or decreasing penalization.
#> Percentage of input data variability preserved in basis representation
#> (per functional observation, approximate):
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> -Inf -Inf -Inf -Inf -Inf -Inf
#> domain basis basis_label basis_args basis_matrix arg
#> TRUE TRUE FALSE FALSE TRUE TRUE
#> family class
#> TRUE TRUE
#--------------------------
y_pc + y_pc == 2 * y_pc #!
#> [1] FALSE FALSE FALSE
all.equal(tf_evaluations(y_pc + y_pc), tf_evaluations(2 * y_pc)) # urks
#> [1] "Component \"1\": Mean relative difference: 0.01724136"
#> [2] "Component \"2\": Mean relative difference: 0.01293853"
#> [3] "Component \"3\": Mean relative difference: 0.01199792"next steps should be adding tests for all combinations of ops and classes and tackling the issues above.
m-muecke
Note: I haven't included the dispatch case for when one is logical, since I don't believe that ever make sense. But you had it previously implemented it.