DeepChem Model Saving/Reloading Triage

See original GitHub issue

DeepChem has a large collection of models not all of which can be saved/reloaded. #2151 adds some first correctness tests for some of our models, but there are many more we need to fix. Let’s use this thread to track the needed work

dc.models.AtomicConvModel
dc.models.ChemCeption
dc.models.CNN
dc.models.DTNNModel
dc.models.DAGModel
dc.models.GraphConvModel
dc.models.MPNNModel
dc.models.WeaveModel
dc.models.MultitaskClassifier
dc.models.MultitaskRegressor
dc.model.MultitaskFitTransformRegressor
dc.models.MultitaskIRVClassifier
dc.models.ProgressiveMultitaskClassifier
dc.models.ProgressiveMultitaskRegressor
dc.models.RobustMultitaskClassifier
dc.models.RobustMultitaskRegressor
dc.models.ScScoreModel
dc.models.SeqToSeq
dc.models.Smiles2Vec
dc.models.TextCNNModel
dc.models.CGCNNModel
dc.models.GATModel
dc.models.SklearnModel
dc.models.GBDTModel (XGBoost)
dc.models.GBDTModel (LightGBM)
dc.models.SingletaskToMultitask
dc.models.NormalizingFlowModel
dc.models.GAN
dc.models.GAN (MixGan)
dc.models.WGAN
dc.metalearning.MAML
dc.rl.A2C
dc.rl.PPO

Some of these models may already work and we just need to test them. Others have subtle saving/reloading issues. It’s a priority for me to get this figured out since we can’t use DeepChem models in any critical settings without saving/loading guarantees. For any folks interested in helping, feel free to grab a model and try to save/reload it! For examples of how to add save/reload tests see #2151. The idea is that you should train a model on a toy dataset, save it to disk, reload it, then check that the original model and the reloaded model make the same predictions.

I’ll post in this issue which model I’m currently working on debugging/adding tests for. If you’d like to help, please drop a note here with the model you’re investigating

Issue Analytics

State:
Created 3 years ago
Comments:13 (8 by maintainers)

Top GitHub Comments

2reactions

hsjang001205commented, Oct 21, 2020

@rbharath I found the part where the error occurred. I will try to more debug and PR.

0reactions

nesanderscommented, Mar 5, 2021

Thank you @rbharath . I can confirm in my recent testing that the MWE I posted above passes when run with 2.4.0.

Note to anyone using the MWE - you must change the model_instance attribute to model for it to work with the new version.

Top Results From Across the Web

Model Classes — deepchem 2.6.2.dev documentation

The place to get started is this “model cheatsheet” that lists various types of custom DeepChem models. Note that some wrappers like SklearnModel...

DeepChem Minutes 9/11/2020 - Community

Bharath this last week has been working on this PR trying to test the saving/reloading for DeepChem models. Unfortunately, it turns out that ......

The Ability to Follow Commands, Age, Pulse Rate, Systolic ...

A neural network model (NN-CAPSO) was developed using the ability to follow commands (whether GCS-motor was <6), age, pulse rate, systolic blood pressure...

deepchem/Lobby - Gitter

Use from deepchem.models.graph_models import GraphConvModel instead. しゃお. @Jinniuzuo_dier_twitter ... Is there an example for saving/reloading a model?

Emergency department triage prediction of clinical outcomes ...

In the critical care outcome prediction, all four machine learning models outperformed the reference model (e.g., AUC, 0.86 [95%CI 0.85–0.87] in ...