export your saved model from azure to local computer and evaluate its accuracy on the test set

export_and_evaluate_azure_tfestimator_model.py

#imports
import json
import numpy as np
from tensorflow.contrib import predictor

## transfer the saved model from azure blob storage to your local computer
ds = ws.get_default_datastore()

ds.download(target_path='outputs',
                   prefix='founder-classifier/outputs/model',
                   show_progress=True)
                   
## restore your tf.estimator model
saved_model_location='outputs/founder-classifier/outputs/model/1570442423'
loaded_model = predictor.from_saved_model(saved_model_location)

## load your test set
with open('./data/test.json', 'rb') as f:
    test_data=json.load(f)
features=np.reshape(test_data['images'], (-1,28,28,1))

## make predictions
predictions = loaded_model({'x': features})['classes']
true_labels=test_data['labels']

## compute the accuracy
count=0
for i in range(len(predictions)):
    if predictions[i]==true_labels[i]:
        count+=1
accuracy=count/len(predictions)
print('The accuracy on the test set is: ', accuracy)