From 81c668cd6718ad9423673d8620ae6e83b498f53b Mon Sep 17 00:00:00 2001
From: lucas_miranda <lucasmiranda42@gmail.com>
Date: Wed, 19 May 2021 16:48:34 +0200
Subject: [PATCH] Replaced for loop with vectorised mapping on ClusterOverlap
regularization layer
---
deepof/model_utils.py | 5 ++++-
deepof_experiments.smk | 13 +++++++++----
tests/test_train_utils.py | 4 +++-
3 files changed, 16 insertions(+), 6 deletions(-)
diff --git a/deepof/model_utils.py b/deepof/model_utils.py
index 5f957fdf..783d0857 100644
--- a/deepof/model_utils.py
+++ b/deepof/model_utils.py
@@ -602,7 +602,10 @@ class ClusterOverlap(Layer):
random_idxs = np.random.choice(random_idxs, self.samples)
get_local_neighbourhood_entropy = partial(
- get_neighbourhood_entropy, tensor=encodings, clusters=hard_groups, k=self.k
+ get_neighbourhood_entropy,
+ tensor=encodings,
+ clusters=hard_groups,
+ k=self.k,
)
purity_vector = tf.map_fn(
diff --git a/deepof_experiments.smk b/deepof_experiments.smk
index f34458aa..d2e82454 100644
--- a/deepof_experiments.smk
+++ b/deepof_experiments.smk
@@ -18,10 +18,11 @@ outpath = "/psycl/g/mpsstatgen/lucas/DLC/DLC_autoencoders/DeepOF/deepof/logs/"
warmup_epochs = [15]
warmup_mode = ["sigmoid"]
losses = ["ELBO"] # , "MMD", "ELBO+MMD"]
+overlap_loss = [0.1, 0.2, 0.5, 0.75, 1.]
encodings = [6] # [2, 4, 6, 8, 10, 12, 14, 16]
cluster_numbers = [15] # [1, 5, 10, 15, 20, 25]
latent_reg = ["variance"] # ["none", "categorical", "variance", "categorical+variance"]
-entropy_knn = [100]
+entropy_knn = [10]
next_sequence_pred_weights = [0.15]
phenotype_pred_weights = [0.0]
rule_based_pred_weights = [0.0]
@@ -51,10 +52,11 @@ rule deepof_experiments:
outpath + "train_models/trained_weights/"
"GMVAE_input_type={input_type}_"
"window_size={window_size}_"
- "NextSeqPred={nspredweight}_"
- "PhenoPred={phenpredweight}_"
- "RuleBasedPred={rulesweight}_"
+ "NSPred={nspredweight}_"
+ "PPred={phenpredweight}_"
+ "RBPred={rulesweight}_"
"loss={loss}_"
+ "overlap_loss={overlap_loss}_"
"loss_warmup={warmup}_"
"warmup_mode={warmup_mode}_"
"encoding={encs}_"
@@ -66,6 +68,7 @@ rule deepof_experiments:
input_type=input_types,
window_size=window_lengths,
loss=losses,
+ overlap_loss=overlap_loss,
warmup=warmup_epochs,
warmup_mode=warmup_mode,
encs=encodings,
@@ -134,6 +137,7 @@ rule train_models:
"PhenoPred={phenpredweight}_"
"RuleBasedPred={rulesweight}_"
"loss={loss}_"
+ "overlap_loss={overlap_loss}_"
"loss_warmup={warmup}_"
"warmup_mode={warmup_mode}_"
"encoding={encs}_"
@@ -153,6 +157,7 @@ rule train_models:
"--rule-based-prediction {wildcards.rulesweight} "
"--latent-reg {wildcards.latreg} "
"--loss {wildcards.loss} "
+ "--overlap_loss {wildcards.overlap_loss} "
"--kl-annealing-mode {wildcards.warmup_mode} "
"--kl-warmup {wildcards.warmup} "
"--mmd-annealing-mode {wildcards.warmup_mode} "
diff --git a/tests/test_train_utils.py b/tests/test_train_utils.py
index 07f21b26..8fb9450e 100644
--- a/tests/test_train_utils.py
+++ b/tests/test_train_utils.py
@@ -96,8 +96,10 @@ def test_autoencoder_fitting(
phenotype_prediction,
rule_based_prediction,
):
+
X_train = np.random.uniform(-1, 1, [20, 5, 6])
y_train = np.round(np.random.uniform(0, 1, [20, 1]))
+
if rule_based_prediction:
y_train = np.concatenate(
[y_train, np.round(np.random.uniform(0, 1, [20, 6]), 1)], axis=1
@@ -117,7 +119,7 @@ def test_autoencoder_fitting(
prun.deep_unsupervised_embedding(
preprocessed_data,
- batch_size=100,
+ batch_size=10,
encoding_size=2,
epochs=1,
kl_warmup=1,
--
GitLab