Tflite converter signature error

TensorFlow General Discussion
,
1

G’day & Namaste

Greatly appreciate the opportunity to submit a query on this forum.

My deepest gratitude to everyone in the Community. Especially the team marshalling the lite effort - your efforts are already giving hope to amputees hoping for low-energy powered protheses in India.

My first attempt at TensorFlow here is to convert a HiddenMarkovModel from core to lite. It’s my first, so a simple cloudy-day rainy-day example was considered. However, the following error is being encountered:

if not signature_keys:
raise ValueError(“Only support at least one signature key.”)

My code is as follows (& hopfully correct!)

import numpy as np
import tensorflow.compat.v2 as tf
tf.enable_v2_behavior()
import tensorflow_probability as tfp
from tensorflow_probability import distributions as tfd

from matplotlib import pylab as plt
%matplotlib inline
import scipy.stats

tfd = tfp.distributions

initial_distribution = tfd.Categorical(probs=[0.8, 0.2])

transition_distribution = tfd.Categorical(probs=[[0.7, 0.3],
                                                 [0.2, 0.8]])

observation_distribution = tfd.Normal(loc=[0., 15.], scale=[5., 10.])

hmm = tfd.HiddenMarkovModel(
    initial_distribution=initial_distribution,
    transition_distribution=transition_distribution,
    observation_distribution=observation_distribution,
    num_steps=7)

# Save the model.
tf.saved_model.save(hmm, '/sample_data/hmm')

# Convert the model
converter = tf.lite.TFLiteConverter.from_saved_model('/sample_data/hmm')
tflite_model = converter.convert()

Any & all advice would be useful, I am certain.

Regards,
g-sobers

3

G’day and Namaste All

I am encouraged to better title my post. But it isn’t editable. So here’s what might encourage you to respond:

Converter error: HiddenMarkovModel Core to Lite

Hope this prompts someone who has stopped by to send some advice my way.

Regards,
g_sobers