How to do this basic case thing?

1

I’m doing:
model:

sess=tf.Session()
session = ed.get_session()

mot1 = Dirichlet(tf.constant([1.0,1.0,1.0]), name='mot1')
qmot1 = Dirichlet(tf.nn.softplus(tf.Variable(tf.constant([1.0,1.0,1.0])), name="qmot1"))

probs = tf.reshape((tf.tile(tf.constant([1.0,5.0,15.0]),tf.constant([100]))),[100,3])
data = tf.constant(Categorical(probs=probs).eval())

probs1 = tf.reshape((tf.tile(mot1.eval(),tf.constant([100]))),[100,3])
w = Categorical(probs=probs1)

inference:

inference = ed.KLqp({mot1: qmot1}, data={data: w})
inference.initialize()
inference.n_iter = 1000

tf.global_variables_initializer().run()

for _ in range(inference.n_iter):
    info_dict = inference.update()
    inference.print_progress(info_dict)
    print(session.run(qmot1))
    print("______________")
    
inference.finalize()

I expect here, qmot1 to converge towards [1.0,5.0,15.0], but absolutely not, instead that prints senseless values. I’m surely missing some basics but dunno what.

2

Ok i had to use mot1 instead of mot1.eval()…
Also better to look at qmot1.mean() instead of qmot1…

sess=tf.Session()
session = ed.get_session()

session.run(tf.global_variables_initializer())

mot1 = Dirichlet(tf.constant([1.0,1.0,1.0]), name='mot1')
qmot1 = Dirichlet(tf.nn.softplus(tf.Variable(tf.constant([1.0,1.0,1.0])), name="qmot1"))

data = Categorical(probs=[1.0,5.0,15.0],sample_shape=[1,2000])
w = Categorical(probs=mot1, sample_shape=[1,2000],name="w")

inference = ed.KLqp({mot1: qmot1}, data={w: data})
inference.initialize()
inference.n_iter = 1000

tf.global_variables_initializer().run()

for _ in range(inference.n_iter):
    info_dict = inference.update()
    inference.print_progress(info_dict)
    print(session.run(qmot1.mean()))
    print("______________")
    
inference.finalize()