New reticulate

.Rbuildignore

@@ -10,3 +10,5 @@
 ^\.positai$
 ^\.claude$
 ^todo\.txt$
+^README_cache$
+^\.DS_Store$

DESCRIPTION

@@ -1,6 +1,6 @@
 Package: walking
 Title: Segments Accelerometry Data into Walking using the Python `forest` Module
-Version: 0.5.1
+Version: 0.6.0
 Authors@R: 
     person(given = "John",
            family = "Muschelli",

R/aaa_utils.R

@@ -3,6 +3,10 @@ oak_base = function() {
     stop(
       "Python package 'forest' is not installed in the active reticulate environment. ",
       "Install it yourself before calling walking functions that use forest.\n",
+      "If installation fails with \"clang++: error: unsupported option '-fopenmp'\", ",
+      "that is a Python toolchain/OpenMP problem, not a walking problem.\n",
+      "Use a Python environment where forest and numba/llvmlite already resolve cleanly, ",
+      "then point reticulate at that environment.\n",
       "See https://github.com/onnela-lab/forest/issues/293 and ",
       "https://github.com/numba/llvmlite/issues/1389."
     )

R/find_walking.R

@@ -69,7 +69,9 @@ find_walking = function(
   if (verbose) {
     message("Bout is Preprocessed")
   }
-  vm_bout = pp_out$vm_bout$vm
+  vm_bout = pp_out$vm_bout
+  vm = vm_bout$vm
+  vm_bout$vm = NULL
   # step_frequency = do.call(reticulate::tuple, as.list(step_frequency))
 
   oak = oak_base()
@@ -80,7 +82,7 @@ find_walking = function(
   #            fs = sample_rate)
 
   cadence_bout = oak$find_walking(
-    vm_bout = vm_bout,
+    vm_bout = vm,
     fs = sample_rate_analysis,
     min_amp = min_amplitude,
     step_freq = step_frequency,
@@ -91,9 +93,8 @@ find_walking = function(
   if (verbose) {
     message("OAK: Find walking is done")
   }
-  vm_bout = pp_out$vm_bout
+
   vm_bout$steps = cadence_bout
-  vm_bout$vm = NULL
   vm_bout = as.data.frame(vm_bout)
   # need to remove 1D aspect
   vm_bout$time = c(vm_bout$time)

R/preprocess_bout.R

@@ -180,11 +180,17 @@ preprocess_bout_r = function(data, sample_rate = 10L) {
   vm_bout_interp = np$sqrt(x_bout_interp**2 +
                              y_bout_interp**2 +
                              z_bout_interp**2) - 1
+  rm(x_bout_interp)
+  rm(y_bout_interp)
+  rm(z_bout_interp)
 
   vm_bout = list(
     t_bout_interp,
     vm_bout_interp
   )
+  rm(vm_bout_interp)
+  rm(t_bout_interp)
+
   process_vm_bout(vm_bout, tz = orig_tz, sample_rate = sample_rate)
 }
 

README.Rmd

@@ -11,6 +11,9 @@ knitr::opts_chunk$set(
   fig.path = "man/figures/README-",
   out.width = "100%"
 )
+reticulate::py_require(
+  "git+https://github.com/onnela-lab/forest@45fb41038bd46c25d9e6a4442aa74fa03b501317", 
+  python_version = "3.11")
 ```
 
 # walking
@@ -33,10 +36,13 @@ devtools::install_github("muschellij2/walking")
 `walking` no longer tries to install `forest` for you. Install `forest` in
 your own Python environment first, then load `walking` from R.
 
-If you hit the LLVM mismatch described in
+If you hit `clang++: error: unsupported option '-fopenmp'` or the LLVM
+mismatch described in
 [onnela-lab/forest#293](https://github.com/onnela-lab/forest/issues/293) and
 [numba/llvmlite#1389](https://github.com/numba/llvmlite/issues/1389), use a
-Python environment where `forest` and its dependencies already resolve cleanly.
+Python environment where `forest` and its dependencies already resolve cleanly,
+then point `reticulate` at that environment with `use_condaenv()` or
+`use_python()`.
 
 ## Example
 
@@ -54,33 +60,56 @@ print(res)
 ## Potential Conflicts
 
 ### Running `forest` and [`stepcount`](https://github.com/jhuwit/stepcount). 
-The two Python modules (`forest` and `stepcount`) can be be installed in the same `conda` environment, but if they are not, this will lead to an error message.  The options.  One solution is to run them in two separate R sessions (recommended). 
+The two Python modules (`forest` and `stepcount`) can have different conflicts of python dependencies and python versions.  You can do specific call-outs with `callr` to try to resolve this:
 
-Alternatively, you can try to install `forest` in the `stepcount` conda environment, such as:
+## Running Walking with Different Python Environments
 
-```{r, eval = FALSE}
-envname = "stepcount2"
-stepcount::conda_create_stepcount(envname = envname)
-# if you have RETICULATE_PYTHON set
-stepcount::unset_reticulate_python()
-stepcount::use_stepcount_condaenv(envname = envname)
-```
-
-and then run examples such as:
+If you want to use both walking estimation from `forest` and `stepcount`, you can run them in separate R processes using `callr` to avoid Python package conflicts. 
+Here's how you can do it:
 
-```{r, eval = FALSE}
-library(walking)
+```{r, cache = TRUE}
+library(callr)
 library(stepcount)
-envname = "stepcount2"
-
-# if you have RETICULATE_PYTHON set
-stepcount::unset_reticulate_python()
-stepcount::use_stepcount_condaenv(envname = envname)
-
 csv_file = system.file("test_data_bout.csv", package = "walking")
-x = readr::read_csv(csv_file)
-colnames(x)[colnames(x) == "UTC time"] = "time"
-res = find_walking(data = x)
+data = readr::read_csv(csv_file)
+colnames(data)[colnames(data) == "UTC time"] = "time"
+
+stepcount_callr = function(data,
+                           ...) {
+
+  reticulate::py_require("stepcount==3.11.0", python_version = "3.10")
+  sc <- reticulate::import("stepcount")
+  stepcount::stepcount_check()
+
+  res = stepcount::stepcount(data, ...)
+  return(res)
+}
+
+# 2. Run the isolated background R process
+result <- callr::r(
+  func = stepcount_callr,
+  show = TRUE,
+  args = list(data = data) # Safely injects data into the process
+)
+head(result)
+
+forest_callr = function(data,
+                        ...) {
+  reticulate::py_require(
+    "git+https://github.com/onnela-lab/forest@45fb41038bd46c25d9e6a4442aa74fa03b501317", 
+    python_version = "3.11")
+  fr = reticulate::import("forest")
+  oak = fr$oak$base
+  oak
+  res = walking::estimate_steps_forest(data, ...)
+  return(res)
+}
+
+# 2. Run the isolated background R process
+fresult <- callr::r(
+  func = forest_callr,
+  show = TRUE,
+  args = list(data = data) # Safely injects data into the process
+)
+head(fresult)
 ```
-
-Remember, however, best practices for Python is "Always create a separate virtual environment for each project" and sometimes one for each "goal".