prediction of an algorithm when no model/dataset exists or the cold start problem is a huge feature/issue on its own, for now, to prevent this from causing the below run time issue:

Traceback (most recent call last):
File "/app/.heroku/python/lib/python3.7/site-packages/flask/app.py", line 2446, in wsgi_app
response = self.full_dispatch_request()
File "/app/.heroku/python/lib/python3.7/site-packages/flask/app.py", line 1951, in full_dispatch_request
rv = self.handle_user_exception(e)
File "/app/.heroku/python/lib/python3.7/site-packages/flask/app.py", line 1820, in handle_user_exception
reraise(exc_type, exc_value, tb)
File "/app/.heroku/python/lib/python3.7/site-packages/flask/_compat.py", line 39, in reraise
raise value
File "/app/.heroku/python/lib/python3.7/site-packages/flask/app.py", line 1949, in full_dispatch_request
rv = self.dispatch_request()
File "/app/.heroku/python/lib/python3.7/site-packages/flask/app.py", line 1935, in dispatch_request
return self.view_functions[rule.endpoint](**req.view_args)
File "/app/src/api.py", line 102, in predict_fmlearn
tt_encoder = fml.get_encoders()[utils.TARGET_TYPE]
KeyError: 'Target Type'

Temporary Fix:
add a fail-safe to check for empty database and return a JSON object indicating no "Model has been Trained!"?

Proper Fix:
TODO: think of a way to solve the cold start problem?

Heroku Deployment Failed: reason - empty database.

when the application is deployed for the first time or the started after the database is cleared, the feature to load the data and train a model is breaking as the data frame loaded is an empty one and getting Xy for the model training fails with the error.

Traceback (most recent call last):
  File "app.py", line 24, in <module>
    from src.api import metrics_api
  File "/home/blake/code/fm-learn/src/api.py", line 22, in <module>
    fml.load_data()
  File "/home/blake/code/fm-learn/src/fmlearn.py", line 60, in load_data
    self._X, self._y = utils.get_Xy(self._df)
  File "/home/blake/code/fm-learn/src/utils/utils.py", line 60, in get_Xy
    y = df[[DATASET_HASH]]
  File "/home/blake/code/fm-learn/venv/lib/python3.7/site-packages/pandas/core/frame.py", line 2806, in __getitem__
    indexer = self.loc._get_listlike_indexer(key, axis=1, raise_missing=True)[1]
  File "/home/blake/code/fm-learn/venv/lib/python3.7/site-packages/pandas/core/indexing.py", line 1553, in _get_listlike_indexer
    keyarr, indexer, o._get_axis_number(axis), raise_missing=raise_missing
  File "/home/blake/code/fm-learn/venv/lib/python3.7/site-packages/pandas/core/indexing.py", line 1640, in _validate_read_indexer
    raise KeyError(f"None of [{key}] are in the [{axis_name}]")
KeyError: "None of [Index(['Dataset Hash'], dtype='object')] are in the [columns]"

Possible Fix:
Stop training the model if the data frame is empty? and train later once data is available.

Other Issues to be considered:

• #18 prediction of an algorithm when no model/dataset exists (cold start problem)
• #19 when to load the data and train model when said scenario occurs?

complete #TODO in predict() function.

The ideal case is the model is trained once (possibly at the server start?) with the existing data and then retrained again and again after a certain time frame has passed by reloading data from the DB.

Since there is no plan for using background processes to trigger an event to automatically retrain the model, an alternate way of achieving this would by checking if a set time has passed before predict() function is called via the API calls and then retraining the model.

This could be a possible workaround to background process though not an ideal one.

complete TODO in predict() function.

check the shape of the input data frame and the shape of the data frame used to train the model, possibly from the class variables?

NOTE: handle what would happen if the data is reloaded?
might have to fix #12 before this?

Add new API to compare and retrieve a list of stored data based on the hash value of the data-set sent to the server in the API call.

Also, add another API to retrieve only the best among the above data.

complete #TODO in load_data() function

Change the currently used key-based SHA256 to a one-time hash function.
Remove the process of encryption and decryption from this project, the current implementation will only handle matches based on hash values.
v1 of the project will not provide the possibility of retrieving the data back from the hash value.

Make sure to update sklearn based on these comments as well.

Based on comments from Joeran Beel, after meeting on 02-04-2020.

complete #TODO in train() function.

currently, as a workaround, the data is being loaded if is not already loaded. This might not be required, but it's better to have separate functions for future use?

Encrypt/Hash data-set which is being received from Client.
Current encryption is not being used in the code (dead code). Update the current code to use it or something better.

when to load the data and train model for the first time?

This scenario was discovered with #17 and should be partially solved with #18 though with the proposed temporary fix. But, the first time training of the model should occur when we have about MAX_NEW_RECORDS no of records in the database initially?

Thought in future this number could be configured.

Expose a new API to retrieve decrypted data-set

HTTP Method: GET
API URI: /metric/<id>/decrypt
Fucntion: To decrypt the requested message with id =