Commit bda995ce authored by lucas_miranda's avatar lucas_miranda
Browse files

Rolled back to KNN instead of radius

parent 45a3beea
Pipeline #96822 canceled with stages
in 4 minutes and 28 seconds
......@@ -18,7 +18,7 @@ losses = ["ELBO"] # , "MMD", "ELBO+MMD"]
encodings = [6] # [2, 4, 6, 8, 10, 12, 14, 16]
cluster_numbers = [25] # [1, 5, 10, 15, 20, 25]
latent_reg = ["none", "categorical", "variance", "categorical+variance"]
entropy_min_n = [2, 5, 8, 12]
entropy_knn = [20, 50, 80, 100]
pheno_weights = [0.01, 0.1, 0.25, 0.5, 1.0, 2.0, 4.0, 10.0, 100.0]
......@@ -35,12 +35,12 @@ rule deepof_experiments:
# ),
expand(
"/psycl/g/mpsstatgen/lucas/DLC/DLC_autoencoders/DeepOF/deepof/logs/latent_regularization_experiments/trained_weights/"
"GMVAE_loss={loss}_encoding={encs}_k={k}_latreg={latreg}_entropymink={entmin}_final_weights.h5",
"GMVAE_loss={loss}_encoding={encs}_k={k}_latreg={latreg}_entropyknn={entknn}_final_weights.h5",
loss=losses,
encs=encodings,
k=cluster_numbers,
latreg=latent_reg,
entmin=entropy_min_n,
entknn=entropy_knn,
),
# expand(
# "/psycl/g/mpsstatgen/lucas/DLC/DLC_autoencoders/DeepOF/deepof/logs/pheno_classification_experiments/trained_weights/"
......@@ -89,7 +89,7 @@ rule latent_regularization_experiments:
output:
trained_models=os.path.join(
outpath,
"latent_regularization_experiments/trained_weights/GMVAE_loss={loss}_encoding={encs}_k={k}_latreg={latreg}_entropymink={entmin}_final_weights.h5",
"latent_regularization_experiments/trained_weights/GMVAE_loss={loss}_encoding={encs}_k={k}_latreg={latreg}_entropyknn={entknn}_final_weights.h5",
),
shell:
"pipenv run python -m deepof.train_model "
......@@ -106,7 +106,7 @@ rule latent_regularization_experiments:
"--mmd-warmup 20 "
"--montecarlo-kl 10 "
"--encoding-size {wildcards.encs} "
"--entropy-min-n {wildcards.entmin} "
"--entropy-knn {wildcards.entknn} "
"--batch-size 256 "
"--window-size 24 "
"--window-step 12 "
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment