What Classification Accuracy expected for hourly trend?

[ericleonardo]

Hi.. Please, what is the trend classification accuracy expected for hourly Forex data?

I'm trying to run the classification Main file on Windows, but there are some dependencies referring to Google Colab TPU.
To solve the dependence on "gs://cloud-tpu-checkpoints/bert/uncased_L-12_H-768_A-12", I got to download Bert files manually and put inside "Bert" folder in Windows.
But I still having TPU issue like on BERT_vectorization.py (line 349):
tpu_cluster_resolver = tf.contrib.cluster_resolver.TPUClusterResolver(TPU_ADDRESS)
AttributeError: module 'tensorflow' has no attribute 'contrib'

Thank you so much for sharing this amazing work...