GraphDataModeler
from neo4j_runway import GraphDataModeler
This class is responsible for generating a graph data model via communication with an LLM. It handles prompt generation, model generation history as well as access to the generated data models.
Attributes
----------
llm : BaseLLM
The LLM used to generate data models.
discovery : Union[str, Discovery], optional
Either a string containing the LLM generated
discovery or a Discovery object that has been run.
user_input : Union[Dict[str, str], UserInput], optional
Either a dictionary with keys general_description
and column names with descriptions or a UserInput
object.
model_iterations : int
The number of times a data model has been returned.
model_history : List[DataModel]
A list of all valid models generated.
current_model : DataModel
The most recently generated or loaded data model.
Class Methods
init
Takes an LLM instance and Discovery information. Either a Discovery object can be provided, or each field can be provided individually.
Parameters
----------
llm : BaseLLM
The LLM used to generate data models.
discovery : Union[str, Discovery], optional
Either a string containing the LLM generated
discovery or a Discovery object that has been run.
If a Discovery object is provided then the remaining
discovery attributes don't need to be provided, by
default ""
user_input : Union[Dict[str, str], UserInput], optional
Either a dictionary with keys general_description
and column names with descriptions or a UserInput
object, by default dict()
data_dictionary : Dict[str, Any], optional
A data dictionary. If single-file input, then the
keys will be column names and the values are
descriptions.
If multi-file input, the keys are file names and
each contain a nested dictionary of column name keys
and description values.
This argument will take precedence over any data
dictionary provided via the Discovery object.
This argument will take precedence over the
allowed_columns argument. By default None
allowed_columns : List[str], optional
A list of allowed columns for modeling. Can be used
only for single-file inputs. By default = list()
create_initial_model
Generate the initial model.
You may access this model with the get_model method
and providing version=1.
Parameters
----------
max_retries : int, optional
The max number of retries for generating the initial
model, by default 3
use_advanced_data_model_generation_rules, optional
Whether to include advanced data modeling rules, by
default True
allow_duplicate_properties : bool, optional
Whether to allow a property to exist on multiple
node labels or relationship types, by default False
enforce_uniqueness : bool, optional
Whether to error if a node has no unique identifiers
(unique or node key).
Setting this to false may be detrimental during code
generation and ingestion. By default True
allow_parallel_relationships : bool, optional
Whether to allow parallel relationships to exist in
the data model, by default False
apply_neo4j_naming_conventions : bool, optional
Whether to apply Neo4j naming conventions to the
generated Data Model, by default True
Returns
-------
DataModel
The generated data model.
get_model
Get the data model version specified. By default will return the most recent model. Version are 1-indexed.
Parameters
----------
version : int, optional
The model version, 1-indexed, by default -1
as_dict : bool, optional
whether to return as a Python dictionary. Will
otherwise return a DataModel object, by default
False
Returns
-------
Union[DataModel, Dict[str, Any]]
The data model in desired format.
Examples
--------
>>> gdm.get_model(1) == gdm.model_history[0]
True
iterate_model
Iterate on the current model. A data model must exist in
the model_history property to run.
Parameters
----------
iterations : int, optional
How many times to perform model generation. Each
successful iteration will be appended to the
GraphDataModeler model_history.
For example if a value of 2 is provided, then two
successful models will be appended to the
model_history. Model generation will use the same
prompt for each generation attempt. By default 1
corrections : Union[str, None], optional
What changes the user would like the LLM to address
in the next model, by default None
max_retries : int, optional
The max number of retries for generating the initial
model, by default 3
use_advanced_data_model_generation_rules, optional
Whether to include advanced data modeling rules, by
default True
allow_duplicate_properties : bool, optional
Whether to allow a property to exist on multiple
node labels or relationship types, by default False
enforce_uniqueness : bool, optional
Whether to error if a node has no unique identifiers
(unique or node key).
Setting this to false may be detrimental during code
generation and ingestion. By default True
allow_parallel_relationships : bool, optional
Whether to allow parallel relationships to exist in
the data model, by default False
apply_neo4j_naming_conventions : bool, optional
Whether to apply Neo4j naming conventions to the
generated Data Model, by default True
Returns
-------
DataModel
The most recently generated data model.
load_model
Append a new data model to the end of the
model_history.
This will become the new current_model.
Parameters
----------
data_model : DataModel
The new data model.
Raises
------
ValueError
If the data_model is not an instance of DataModel.
Class Properties
allowed_columns
The allowed columns for model generation. If multi-file, then a dictionary with file name keys and list of columns for values. If single-file, then a list of columns.
Returns
-------
Dict[str, List[str]]
The allowd columns for data model generation.
Raises
------
AssertionError
When no _data_dictionary attribute is initialized in
the GraphDataModeler class.
current_model
Get the most recently created or loaded data model.
Returns
-------
DataModel
The current data model.
current_model_viz
Visualize the most recent model with Graphviz.
Returns
-------
Digraph
The object to visualize.
is_multifile
Whether data is multi-file or not.
Returns
-------
bool
True if multi-file detected, else False