From 7663405dbef670d5ec356ca5ecc477f8e83ac3a8 Mon Sep 17 00:00:00 2001
From: lucas_miranda <lucasmiranda42@gmail.com>
Date: Tue, 28 Jul 2020 18:01:36 +0200
Subject: [PATCH] Enhanced performance with tf.function decorators
---
source/model_utils.py | 21 ++++++++++++++-------
source/models.py | 2 +-
source/utils.py | 4 +++-
3 files changed, 18 insertions(+), 9 deletions(-)
diff --git a/source/model_utils.py b/source/model_utils.py
index 1181d34d..2b1443d8 100644
--- a/source/model_utils.py
+++ b/source/model_utils.py
@@ -5,7 +5,6 @@ from keras import backend as K
from sklearn.metrics import silhouette_score
from tensorflow.keras.constraints import Constraint
from tensorflow.keras.layers import Layer
-import numpy as np
import tensorflow as tf
import tensorflow_probability as tfp
@@ -13,20 +12,27 @@ tfd = tfp.distributions
tfpl = tfp.layers
# Helper functions
-def far_away_uniform_initialiser(shape, minval=0, maxval=15, iters=10000000):
+@tf.function
+def far_away_uniform_initialiser(shape, minval=0, maxval=15, iters=100000):
"""
Returns a uniformly initialised matrix in which the columns are as far as possible
"""
- init_dist = 0
- for i in range(iters):
- temp = np.random.uniform(minval, maxval, shape)
- dist = np.abs(np.linalg.norm(np.diff(temp)))
+
+ init = tf.random.uniform(shape, minval, maxval)
+ init_dist = tf.abs(tf.norm(tf.math.subtract(init[1:], init[:1])))
+ i = 0
+
+ while tf.less(i, iters):
+ temp = tf.random.uniform(shape, minval, maxval)
+ dist = tf.abs(tf.norm(tf.math.subtract(temp[1:], temp[:1])))
if dist > init_dist:
init_dist = dist
init = temp
- return init.astype(np.float32)
+ i += 1
+
+ return init
def compute_kernel(x, y):
@@ -40,6 +46,7 @@ def compute_kernel(x, y):
)
+@tf.function
def compute_mmd(tensors):
x = tensors[0]
diff --git a/source/models.py b/source/models.py
index 023d0fd8..db2febde 100644
--- a/source/models.py
+++ b/source/models.py
@@ -195,7 +195,7 @@ class SEQ_2_SEQ_GMVAE:
if self.prior == "standard_normal":
init_means = far_away_uniform_initialiser(
- [self.number_of_components, self.ENCODING], minval=0, maxval=5
+ shape=[self.number_of_components, self.ENCODING], minval=0, maxval=5
)
self.prior = tfd.mixture.Mixture(
diff --git a/source/utils.py b/source/utils.py
index fc169e4b..7d789844 100644
--- a/source/utils.py
+++ b/source/utils.py
@@ -785,7 +785,9 @@ def cluster_transition_matrix(
trans_normed = np.zeros([k, k]) + 1e-5
for t in trans.keys():
trans_normed[int(t[0]), int(t[1])] = np.round(
- trans[t] / (sum({i: j for i, j in trans.items() if i[0] == t[0]}.values()) + 1e-5), 3
+ trans[t]
+ / (sum({i: j for i, j in trans.items() if i[0] == t[0]}.values()) + 1e-5),
+ 3,
)
# If specified, returns the transition matrix as an nx.Graph object
--
GitLab