Bump numpy from 1.19.5 to 1.21.0

README.md

@@ -3,11 +3,3 @@ This repository contains code for a method for clustering multivariate time seri
 The use of the method is not restricted to clinical data. It can generally be used for multivariate time series data. 
 
 In addition to variational autoencoders with gaussian mixture priors, the code allows to train ordinary variational autoencoders (multivariate gaussian prior) and ordinary autoencoders (without prior), for all available time series models (LSTM, GRUs and Transformers).
-
-The code was written using
-
-(1) Python 3.6 and Tensorflow 1.10.1 (directory tensorflow1), and
-
-(2) Python 3.8 and Tensorflow 2.4.0 (directory tensorflow2).
-
-Note that only the Tensorflow 2.4.0 version gives the option for training transformer networks in addition to LSTMs/GRUs.

tensorflow2/vader/layers.py → VaDER/layers.py

@@ -1,6 +1,15 @@
 import tensorflow as tf
 import numpy as np
-from utils import positional_encoding, scaled_dot_product_attention
+from scipy.optimize import linear_sum_assignment as linear_assignment
+from sklearn import metrics
+from scipy.stats import multivariate_normal
+import warnings
+from sklearn.mixture import GaussianMixture
+import tensorflow_addons as tfa
+import abc
+
+from VaDER.utils import get_angles, positional_encoding, create_padding_mask, create_look_ahead_mask, \
+    scaled_dot_product_attention, create_masks
 
 class ImputationLayer(tf.keras.layers.Layer):
     def __init__(self, A_init):
@@ -112,7 +121,6 @@ def call(self, x, training, mask):
 
         return out2
 
-
 class TransformerDecoderLayer(tf.keras.layers.Layer):
     def __init__(self, d_model, num_heads, dff, rate=0.1):
         super(TransformerDecoderLayer, self).__init__()
@@ -146,7 +154,6 @@ def call(self, x, enc_output, training,
 
         return out3, attn_weights_block1, attn_weights_block2
 
-
 class TransformerEncoder(tf.keras.layers.Layer):
     def __init__(self, num_layers, D, d_model, num_heads, dff, maximum_position_encoding, rate=0.1):
         super(TransformerEncoder, self).__init__()
@@ -174,7 +181,6 @@ def call(self, x, training, mask):
 
         return x
 
-
 class TransformerDecoder(tf.keras.layers.Layer):
     def __init__(self, num_layers, D, d_model, num_heads, dff, maximum_position_encoding, rate=0.1):
         super(TransformerDecoder, self).__init__()

tensorflow2/vader/utils.py → VaDER/utils.py

@@ -1,5 +1,12 @@
 import tensorflow as tf
 import numpy as np
+from scipy.optimize import linear_sum_assignment as linear_assignment
+from sklearn import metrics
+from scipy.stats import multivariate_normal
+import warnings
+from sklearn.mixture import GaussianMixture
+import tensorflow_addons as tfa
+import abc
 
 def get_angles(pos, i, d_model):
     angle_rates = 1 / np.power(10000, (2 * (i//2)) / np.float32(d_model))
@@ -64,7 +71,6 @@ def scaled_dot_product_attention(q, k, v, mask):
 
     return output, attention_weights
 
-
 def create_masks(inp):
     # does not matter which i we take for inp[:,:,i], so just take i=0
     tar = inp[:, :-1, 0]

tensorflow2/vader/vader.py → VaDER/vader.py

@@ -1,11 +1,14 @@
 import tensorflow as tf
+import numpy as np
 from scipy.optimize import linear_sum_assignment as linear_assignment
 from sklearn import metrics
 from scipy.stats import multivariate_normal
-import numpy as np
 import warnings
 from sklearn.mixture import GaussianMixture
-from vadermodel import VaderRNN, VaderFFN, VaderTransformer
+import tensorflow_addons as tfa
+import abc
+
+from VaDER.vadermodel import VaderModel, VaderRNN, VaderFFN, VaderTransformer
 
 class VADER:
     '''

tensorflow2/vader/vadermodel.py → VaDER/vadermodel.py

@@ -1,9 +1,19 @@
 import tensorflow as tf
+import numpy as np
+from scipy.optimize import linear_sum_assignment as linear_assignment
+from sklearn import metrics
+from scipy.stats import multivariate_normal
+import warnings
+from sklearn.mixture import GaussianMixture
 import tensorflow_addons as tfa
 import abc
-import numpy as np
-from layers import ImputationLayer, RnnDecodeTransformLayer, GmmLayer, TransformerEncoder, TransformerDecoder
-from utils import create_masks
+
+from VaDER.layers import ImputationLayer, RnnDecodeTransformLayer, GmmLayer, MultiHeadAttention, \
+    point_wise_feed_forward_network, TransformerEncoderLayer, TransformerDecoderLayer, \
+    TransformerEncoder, TransformerDecoder
+from VaDER.utils import get_angles, positional_encoding, create_padding_mask, create_look_ahead_mask, \
+    scaled_dot_product_attention, create_masks
+
 
 class VaderModel(tf.keras.Model):
     def __init__(self, X, W, D, K, I, cell_type, n_hidden, recurrent, output_activation, cell_params=None):

tensorflow2/test_vader.py → main.py

@@ -1,20 +1,17 @@
 import numpy as np
 import time
 import os
-import sys
-import cProfile
 import tensorflow as tf
 tf.config.run_functions_eagerly(False)
-from tensorflow2.vader.vader import VADER
-# from tensorflow2.prog import Transformer
+from VaDER.vader import VADER
 
 save_path = os.path.join('test_vader', 'vader.ckpt')
 
 np.random.seed(123)
 
 # generating some simple random data [ns * 2 samples, nt - 1 time points, 2 variables]
 nt = int(8)
-ns = int(5e2    )
+ns = int(5e2)
 sigma = 0.5
 mu1 = -2
 mu2 = 2
@@ -60,6 +57,7 @@
 # "cell_params". These are hyperparameters to the transformer architecture, and are interpreted as in
 # https://www.tensorflow.org/tutorials/text/transformer
 # The use of dropout can be debated, due to the regularizing properties of the variational layer
+# Also note that a Transformer is not a great model choice for this data...
 vader = VADER(X_train=X_train, W_train=W_train, y_train=y_train, save_path=save_path, n_hidden=[12, 2], k=4,
               learning_rate=1e-3, output_activation=None, recurrent=True, cell_type="Transformer", batch_size=64,
               cell_params={'d_model': 4, 'num_layers': 1, 'num_heads': 1, 'dff': 16, 'rate': 0.0})

requirements.txt

@@ -0,0 +1,6 @@
+numpy==1.21.0
+scikit_learn==1.0.2
+scipy==1.7.3
+setuptools==57.0.0
+tensorflow==2.7.1
+tensorflow_addons==0.16.1

setup.py

@@ -0,0 +1,23 @@
+from setuptools import setup
+
+setup(
+    name='VaDER',
+    version='0.1.0',
+    author='Johann de Jong',
+    author_email='johanndejong@gmail.com',
+    packages=['VaDER'],
+    #    package_dir={'': '.'},
+    #    scripts=['bin/script1', 'bin/script2'],
+    url='',
+    license='LICENSE',
+    description='Clustering multivariate time series with missing values',
+    long_description=open('README.md').read(),
+    install_requires=[
+        "numpy >= 1.19.5",
+        "scikit_learn >= 1.0.2",
+        "scipy >= 1.7.3",
+        "setuptools >= 57.0.0",
+        "tensorflow >= 2.7.1",
+        "tensorflow_addons >= 0.16.1"
+    ]
+)