If dtype == 'float', clean_data will cast the contents of data to floats

ds_name can be the name of the dataset in data.tabular_ds_info.json, or a path to a .ds_info.json file

Grab the container id for this db's docker container, starting container if needed. If you can't connect, make sure the store_lock file is re

Helper function that adds columns relevant to a date in the column `field_name` of `df`.

tensorboard hparam logger needs a dict with all values as ints, floats, or strings. This function flattens a dict into that format.

Make sure `df[field_name]` is of the right date type.

Returns the model's prediction of the class of the input given the output of self.forward

Helper function that adds trigonometric date/time features to a date in the column `field_name` of `df`.

Hands the job of running create_datapoint_from_database in parallel over to the linux scheduler and xargs, for speed and memory improveme

Return feature names of date/time cycles as produced by `cyclic_dt_features`.

Find and return a good learning rate for this model with this optimizer. ***THIS MESSES UP THE MODEL AND OPTIMIZER - YOU NEED TO RESET THEM

Runs the GNN component of the model and returns logits for output classes. :param g: BatchedDGLGraph with g.ndata[h] initialized to

Uses the tabular models in self.node_initializers to encode the raw database features (datetimes, text, etc.) of each table, such tha

Fixes the column types of the dataframe output by featuretools to match those of features (why this isn't done automatically I have no idea..

Adapted from https://sgugger.github.io/how-do-you-find-a-good-learning-rate.html Does a single epoch of training, sweeping through learning r

Returns a trunated copy of edge_list, node_types, edge_types, and features. Removes all nodes with index >= max_nodes_per_graph

predictions should be pandas dataframe with columns ['dp_id', 'prob']