brandonlind
commented
2 years ago
- Some of the intro language is confusing, is the purpose of the script to train the GF, make a prediction from GF, or test the GF? e.g.
"fit trained models from gradient forests to the climate of a transplant location" "Given a trained gradient forest, fit model to input climate data,
garden_data." Confusing: "fit to common gardens" I think what I find confusing is the use of the word "fit" here, which is used in reference to training, but you use it here regarding making a prediciton?
the way I've been using train/fit is like this - i use the genetic/climate data to train a model (eg GF) then I use future climate to fit that model (this gives a prediction). After fitting comes validation where I test the fitted model (ie where I test the prediction) using known fitness.
- The file directory and setup needs explantion: garden_dir = makedir(op.join(DIR, 'fitting/gradient_forests/garden_files')) fitting_dir = makedir(op.join(op.dirname(garden_dir), 'fitting_outfiles')) training_outdir = op.join(DIR, 'training/gradient_forests/training_outfiles')
Note I use a different directory structure for the command line script (see README in 01_src) that I think makes more sense than this layout. In this layout there is a fitting directory and a training directory where each has subfolders regarding GF, RONA etc. In the new layout for the command line scripts, the method is the top layer and fitting/training/validation are subdirectories of the method.
notebook directory layout (note a lot of the times I'll print out full paths when the code actually saves things so i don't have to do mental gymnastics with the op.join / op.dirname stuff):
DIR
|
|------ fitting
| |
| |------ gradient_forests
| | |------garden_files # garden_dir = make_dir(op.join(DIR, 'fitting/gradient_forests/garden_files'))
| | |
| | |------fitting_outfiles # fitting_dir = makedir(op.join(op.dirname(garden_dir), 'fitting_outfiles'))
|
|------ training
| |
| |------ gradient_forests
| | |------garden_files # training_outdir = op.join(DIR, 'training/gradient_forests/training_outfiles')
- Maybe it was because I was not reviewing the final script, but the output from testing the fitting script wasn't what I was expecting tting/gradient_forests/fitting_outfiles/1231094_pooled_all_100_gradient_forest_offset.txt")
offset 1 0.716106 2 0.716106 3 0.716105 4 0.716105 5 0.716105 96 0.000645 97 0.000509 98 0.000360 99 0.000221 100 0.000000 100 rows × 1 columns
Why is this 100 rows x 1 column? Each source location (100) should have an offset prediction to a different common garden (100), so I was expecting a 100 x 100 matrix.
This is because I parallelized the fitting to each common garden, so I haven't combined fitted files (and I can therefore read them in in parallel as well). So for this file that you referenced, are offsets to garden 100 (100 is in the filename). Note also that I almost always use row names instead of default indices (it looks like the output is the first 5 and last 5 rows of the file), I'm always scared indices will get shuffled so I almost always will use names in place for look-ups etc - here the row names are subpopIDs.
- Let's make sure to plot some visualizations of the CI curves for each environment and overall CI
TODO
DrK-Lo
"fit trained models from gradient forests to the climate of a transplant location" "Given a trained gradient forest, fit model to input climate data,
garden_data." Confusing: "fit to common gardens" I think what I find confusing is the use of the word "fit" here, which is used in reference to training, but you use it here regarding making a prediciton?The file directory and setup needs explantion: garden_dir = makedir(op.join(DIR, 'fitting/gradient_forests/garden_files')) fitting_dir = makedir(op.join(op.dirname(garden_dir), 'fitting_outfiles')) training_outdir = op.join(DIR, 'training/gradient_forests/training_outfiles')
Maybe it was because I was not reviewing the final script, but the output from testing the fitting script wasn't what I was expecting tting/gradient_forests/fitting_outfiles/1231094_pooled_all_100_gradient_forest_offset.txt")
offset 1 0.716106 2 0.716106 3 0.716105 4 0.716105 5 0.716105 96 0.000645 97 0.000509 98 0.000360 99 0.000221 100 0.000000 100 rows × 1 columns
Why is this 100 rows x 1 column? Each source location (100) should have an offset prediction to a different common garden (100), so I was expecting a 100 x 100 matrix.