TensorFlow 2.0 の Variable を ndarray に変換するには tensorflow.Variable.numpy() を使います。

import tensorflow

x = tensorflow.Variable(tensorflow.random.uniform([4, 4]))
print(x)

y = x.numpy()
print(y)

$ # macOS Mojave (10.14) + TensorFlow 2.0.0b1
$ python3 sample.py
<tf.Variable 'Variable:0' shape=(4, 4) dtype=float32, numpy=
array([[0.4963708 , 0.5265337 , 0.645676  , 0.36317182],
       [0.88785803, 0.6770747 , 0.34829807, 0.7676312 ],
       [0.64356697, 0.1472162 , 0.6687521 , 0.6949041 ],
       [0.9588698 , 0.0772624 , 0.69979465, 0.9192028 ]], dtype=float32)>
[[0.4963708  0.5265337  0.645676   0.36317182]
 [0.88785803 0.6770747  0.34829807 0.7676312 ]
 [0.64356697 0.1472162  0.6687521  0.6949041 ]
 [0.9588698  0.0772624  0.69979465 0.9192028 ]]

参考資料