Variadic Keyword not handled correctly in `dot_digraph`

[thorwhalen]

Reproduce error:

from functools import partial

from meshed import DAG, FuncNode
from sklearn.decomposition import PCA
import numpy as np


X_train = np.random.random((100, 30))
y_train = None
X_run = np.random.random((20, 30))


def make_learner_(learner_class,
                  **learner_kwargs):
    model = learner_class
    # instantiate the learner
    learner_instance = model(**learner_kwargs)
    return learner_instance


def make_learner_node(learner_class,
                      name='make_learner',
                      output_name='learner',
                      **learner_kwargs):
    learner_ = partial(make_learner_, learner_class=learner_class)
    return FuncNode(func=learner_,
                    name=name,
                    out=output_name)


def make_train_learner_(learner, X_train, y_train, fitting_method_name='fit'):
    fitting_method = getattr(learner, fitting_method_name)
    fitted_learner = fitting_method(X_train, y_train)
    return fitted_learner


def make_train_learner_node(learner_bind='learner',
                            X_train_bind='X_train',
                            y_train_bind='y_train',
                            fitting_method_name='fit',
                            name='train_learner'):
    make_train_learner__ = partial(make_train_learner_, fitting_method_name=fitting_method_name)
    return FuncNode(func=make_train_learner__,
                    name=name,
                    bind={'learner': learner_bind, 'X_train': X_train_bind, 'y_train': y_train_bind},
                    out='fitted_learner')


def make_fitter_and_runner_nodes_(
        model,
        fit_bind='fit',
        transform_bind='transform',
        **model_kwargs,
):
    learner_node = make_learner_node(model, **model_kwargs)
    training_node = make_train_learner_node(learner_bind='learner')

    return learner_node, training_node  # WTF, Commenting the training node make the DAG d_2 fail to construct!


d_2 = DAG(make_fitter_and_runner_nodes_(model=PCA,
                                        transform_bind='score_samples'))

print(d_2(X_train, y_train).score_samples(X_train))

d_2.dot_digraph()

[thorwhalen]

Also (but different problem perhaps), adding two DAGs doesn't handle VARIADICS well:

Example:

d_1 = DAG(make_fitter_and_runner_nodes_(model=PCA))

d_2 = DAG(make_fitter_and_runner_nodes_(model=PCA,
                                        transform_bind='score_samples'))

print(d_2(X_train=X_train, y_train=y_train, X_run=X_train))
d_2 = ch_names(d_2, renamer='TEST')

d = d_1 + d_2
# ValueError: You can't have several variadic keywords: ('learner_kwargsTEST', 'learner_kwargs')

[sylvainbonnot]

Concerning, the first problem: the sequence of function calls is as follows

dot_digraph
--(calls)--> dot_digraph_body
--(calls)--> dot_lines_of_func_nodes(d_2.func_nodes, start_lines=()) 
--(calls)--> dot_lines_of_func_node(fnode) # where fnode = d_2.func_nodes[0]
--(calls)--> params = fnode.sig.ch_names(fnode.sig, **fnode.bind).params  
--(calls)--> self.ch_param_attrs('name', **changes_for_name)
--(calls)--> all_pk_self = self.modified(**{name: {'kind': PK} for name in self.names})

To resolve the problem, self.ch_params_attrs needs to be called with _allow_reordering=True instead of the default False

[sylvainbonnot]

In summary:
in i2.signatures, in the definition of "modified" function, change the default parameter _allow_reordering=True, as follows:

    def modified(self, _allow_reordering=True, **changes_for_name):