How to save and load model

[Marcosuff]

Hello,

I'm trying to train a multitask regression using the same approach posted here:
https://github.com/deepchem/deepchem/blob/master/examples/notebooks/Multitask_Networks_on_MUV.ipynb.
Everything works fine until I need to save and load the model to make predictions. However when I load my model using the approach posted here
#650 (comment)
the evaluation results are completely different. Here's a sample output:

Best neural network prior to saving:

computed_metrics: [0.32803591086520073, 0.5407940796855545, 0.51435272945748, 1.5072726859368315, 0.8059894834144922, 0.43465885298644813, 0.7546208137228906, 0.5057226961061108, 0.5968139410991427, 0.4166696352270262, 0.40835529858]

Loaded model:

computed_metrics: [18.97499015453843, 18.11683325035692, 48.087994983626494,     36.00984590302385, 45.557838677120536, 36.11361848497406, 46.199146758789865,     54.76305977799556, 69.88862611897348, 37.10991454276001, 52.028279977709545]

As you can see the metrics are worse after loading the model.

Am I doing it the wrong way? Is there another way to save and load the model for predictions?

[rbharath]

We've converted to Keras as our backend, so we're just using the Keras save/load functionality. I believe that these errors (which may have been with our custom TensorGraph code) should no longer occur. I'm going to close this issue, but please re-open if you're still facing problems.

[sorenwacker]

Please add an example.

[rbharath]

So if you have a model, you can just call

model = GraphConvModel(model_dir="/place/to/save/your/model")
model.fit(dataset, nb_epoch=15)

And the model will automatically be saved to "/place/to/save/your/model". To reload the model you can do

model = GraphConvModel(model_dir="/place/to/save/your/model")
model.restore()

And you're all set!

[croqquettes02]

how can i reload the model that saved using model.restore()