From 61bb9c12ba7b9e09bc0d0898218223cc1f33e959 Mon Sep 17 00:00:00 2001
From: lucas_miranda <lucasmiranda42@gmail.com>
Date: Fri, 12 Feb 2021 00:48:39 +0100
Subject: [PATCH] Added latent regularization control to
deepof.data.coordinates.deep_unsupervised_embedding()
---
deepof/train_model.py | 3 ++-
deepof/train_utils.py | 23 ++++++++++++++++++-----
2 files changed, 20 insertions(+), 6 deletions(-)
diff --git a/deepof/train_model.py b/deepof/train_model.py
index 638f21c6..7380650c 100644
--- a/deepof/train_model.py
+++ b/deepof/train_model.py
@@ -287,7 +287,7 @@ project_coords = project(
animal_ids=tuple([animal_id]),
arena="circular",
arena_dims=tuple([arena_dims]),
- enable_iterative_imputation=True,
+ enable_iterative_imputation=False,
exclude_bodyparts=exclude_bodyparts,
exp_conditions=treatment_dict,
path=train_path,
@@ -359,6 +359,7 @@ if not tune:
trained_models = project_coords.deep_unsupervised_embedding(
(X_train, y_train, X_val, y_val),
+ epochs=1,
batch_size=batch_size,
encoding_size=encoding_size,
hparams=hparams,
diff --git a/deepof/train_utils.py b/deepof/train_utils.py
index 11ea712f..4b67d37e 100644
--- a/deepof/train_utils.py
+++ b/deepof/train_utils.py
@@ -94,6 +94,8 @@ def get_callbacks(
phenotype_class: float,
predictor: float,
loss: str,
+ reg_cat_clusters: bool,
+ reg_cluster_variance: bool,
logparam: dict = None,
outpath: str = ".",
) -> List[Union[Any]]:
@@ -103,6 +105,14 @@ def get_callbacks(
- cp_callback: for checkpoint saving,
- onecycle: for learning rate scheduling"""
+ latreg = "none"
+ if reg_cat_clusters and not reg_cluster_variance:
+ latreg = "categorical"
+ elif reg_cluster_variance and not reg_cat_clusters:
+ latreg = "variance"
+ elif reg_cat_clusters and reg_cluster_variance:
+ latreg = "categorical+variance"
+
run_ID = "{}{}{}{}{}{}_{}".format(
("GMVAE" if variational else "AE"),
("_Pred={}".format(predictor) if predictor > 0 and variational else ""),
@@ -110,6 +120,7 @@ def get_callbacks(
("_loss={}".format(loss) if variational else ""),
("_encoding={}".format(logparam["encoding"]) if logparam is not None else ""),
("_k={}".format(logparam["k"]) if logparam is not None else ""),
+ ("_latreg={}".format(latreg)),
(datetime.now().strftime("%Y%m%d-%H%M%S")),
)
@@ -251,11 +262,11 @@ def autoencoder_fitting(
log_history: bool,
log_hparams: bool,
loss: str,
- mmd_warmup,
- montecarlo_kl,
- n_components,
- output_path,
- phenotype_class,
+ mmd_warmup: int,
+ montecarlo_kl: int,
+ n_components: int,
+ output_path: str,
+ phenotype_class: float,
predictor: float,
pretrained: str,
save_checkpoints: bool,
@@ -290,6 +301,8 @@ def autoencoder_fitting(
phenotype_class=phenotype_class,
predictor=predictor,
loss=loss,
+ reg_cat_clusters=reg_cluster_variance,
+ reg_cluster_variance=reg_cluster_variance,
logparam=logparam,
outpath=output_path,
)
--
GitLab