# Although Mercury-ML does not require you to work with config files, it is encouraged as this allows you to make full
# use of the code abstraction capabilities that are on offer.
# We have included a few small utils that help with dealing with config files. In this example, we'll show how you can
# create and read from a referenced JSON file
# Let's first try to read our JSON file using the json library:
# Note the following entry: "filename": {"$ref": "#meta_info/model_object_name"}.
import json
with open("./01_reading_referenced_json_config.json", "r") as f:
config = json.load(f)
print(json.dumps(config, indent=2), "\n")
# Now let's use mercury_ml.common.utils.load_referenced_json_config (which builds on the jsonref library)
# Note the following entry: "filename": "my_model_123.h5". The line {"$ref": "#meta_info/model_object_name"} has been
# replaced by the value in meta_info/model_object_name.
from mercury_ml.common.utils import load_referenced_json_config
config = load_referenced_json_config(
"./01_reading_referenced_json_config.json")
print(json.dumps(config, indent=2))
from pip.operations import freeze
packages = []
for p in freeze.freeze():
packages.append(p)
return packages
# -
# ## Config
# #### Load config
config = utils.load_referenced_json_config(config_filepath)
print_dict(config)
# #### Set model_id
evaluation_session_id = str(uuid.uuid4().hex)
print(evaluation_session_id)
# #### Update config
#
# The function `utils.recursively_update_config(config, string_formatting_dict)` allows us to use string formatting to replace placeholder strings with acctual values.
#
# for example:
#
def test_load_referenced_json_config(filepath, expected):
config = load_referenced_json_config(filepath)
assert config == expected