Save and Load

In [1]:

%load_ext autoreload
%autoreload 2

import datamol as dm

%load_ext autoreload %autoreload 2 import datamol as dm

Any Molfeat featurizer should be serializable. This makes it easy to share a specific instantiation of a featurizer. Any featurizer can thus be created from - or saved to - a state dictionary. We support native support to save these state dicts to both YAML and JSON.

Example¶

In [2]:

from molfeat.trans import MoleculeTransformer

featurizer = MoleculeTransformer("ecfp")

from molfeat.trans import MoleculeTransformer featurizer = MoleculeTransformer("ecfp")

In [3]:

print(featurizer.to_state_yaml())

print(featurizer.to_state_yaml())

_molfeat_version: 0.8.2.dev26+g8b6a1c9.d20230401
args:
  dtype: null
  featurizer: ecfp
  n_jobs: 1
  parallel_kwargs: null
  verbose: false
name: MoleculeTransformer

In [4]:

print(featurizer.to_state_json())

print(featurizer.to_state_json())

{"name": "MoleculeTransformer", "args": {"featurizer": "ecfp", "n_jobs": 1, "verbose": false, "dtype": null, "parallel_kwargs": null}, "_molfeat_version": "0.8.2.dev26+g8b6a1c9.d20230401"}

In [5]:

print(featurizer.to_state_dict())

print(featurizer.to_state_dict())

{'name': 'MoleculeTransformer', 'args': {'featurizer': 'ecfp', 'n_jobs': 1, 'verbose': False, 'dtype': None, 'parallel_kwargs': None}, '_molfeat_version': '0.8.2.dev26+g8b6a1c9.d20230401'}

In [6]:

print(featurizer.to_state_json())

print(featurizer.to_state_json())

{"name": "MoleculeTransformer", "args": {"featurizer": "ecfp", "n_jobs": 1, "verbose": false, "dtype": null, "parallel_kwargs": null}, "_molfeat_version": "0.8.2.dev26+g8b6a1c9.d20230401"}

In [7]:

# You can also directly save to files:
import tempfile

with tempfile.TemporaryDirectory() as tmpdir:
    path = dm.fs.join(tmpdir, "state_dict.yml")
    featurizer.to_state_yaml_file(path)
    featurizer = MoleculeTransformer.from_state_yaml_file(path)

# You can also directly save to files: import tempfile with tempfile.TemporaryDirectory() as tmpdir: path = dm.fs.join(tmpdir, "state_dict.yml") featurizer.to_state_yaml_file(path) featurizer = MoleculeTransformer.from_state_yaml_file(path)

Loop over multiple featurizers¶

One implication of this, is that despite the various interfaces, you can easily loop over multiple featurizers.

In [8]:

FEATURIZERS = [
    {"name": "MoleculeTransformer", "args": {"featurizer": "ecfp", "n_jobs": 1, "verbose": False, "dtype": None, "parallel_kwargs": None}, "_molfeat_version": "0.0.1"},
    {"name": "MoleculeTransformer", "args": {"featurizer": "maccs", "n_jobs": 1, "verbose": False, "dtype": None, "parallel_kwargs": None}, "_molfeat_version": "0.0.1"},
    {"name": "MoleculeTransformer", "args": {"featurizer": "fcfp", "n_jobs": 1, "verbose": False, "dtype": None, "parallel_kwargs": None}, "_molfeat_version": "0.0.1"},
]

FEATURIZERS = [ {"name": "MoleculeTransformer", "args": {"featurizer": "ecfp", "n_jobs": 1, "verbose": False, "dtype": None, "parallel_kwargs": None}, "_molfeat_version": "0.0.1"}, {"name": "MoleculeTransformer", "args": {"featurizer": "maccs", "n_jobs": 1, "verbose": False, "dtype": None, "parallel_kwargs": None}, "_molfeat_version": "0.0.1"}, {"name": "MoleculeTransformer", "args": {"featurizer": "fcfp", "n_jobs": 1, "verbose": False, "dtype": None, "parallel_kwargs": None}, "_molfeat_version": "0.0.1"}, ]

In [9]:

for state_dict in FEATURIZERS:
    featurizer = MoleculeTransformer.from_state_dict(state_dict)
    # ... Work your magic!

for state_dict in FEATURIZERS: featurizer = MoleculeTransformer.from_state_dict(state_dict) # ... Work your magic!