Visualizing Models and Data

Let’s revisit the starter snippet from the homepage of our documentation.

Line by line, we’ll walk through how Visualizing your Models and Datasets works!

First, let’s import the necessary libraries:

from olorenchemengine.visualizations import *
from olorenchemengine import *
import olorenchemengine as oce

Second, we’ll use a pandas dataframe of csv data given by the line

df = oce.ExampleDataset

:: to train the model, but visualizations do not allow us to pass in a pandas dataframe. Instead, it must be of type ‘BaseDataset’

with this line,

test_dataset = oce.BaseDataset(name='purple', data=df.to_csv(), structure_col='Smiles', property_col='pChEMBL Value')

See how it all comes together below.

## Loading in a dataset

# df is a Pandas Dataframe with the following columns:
# "Smiles" (structure)
# "pChEMBL Value" (property to be predicted)
df = oce.ExampleDataset
test = oce.BaseDataset(name='purple', data=df.to_csv(), structure_col='Smiles', property_col='pChEMBL Value')

Next, we’ll train, fit, save, and load our model in the same as before:

model = oce.BaseBoosting([
            oce.RandomForestModel(oce.DescriptastorusDescriptor("rdkit2dnormalized"),n_estimators=1000),
            oce.BaseTorchGeometricModel(oce.TLFromCheckpoint("default"), preinitialized=True),
            oce.RandomForestModel(oce.OlorenCheckpoint("default"),n_estimators=1000)])


## Training the model
model.fit(df["Smiles"], df["pChEMBL Value"])

## Predicting property of new compounds
y_pred = model.predict(["CC(=O)OC1=CC=CC=C1C(=O)O"])

## Saving the model
oce.save(model, "model.oce")

## Loading the model
loaded_model = oce.load("model.oce")

Now, we can use the loaded_model to Visualize our with VisualizeModelSim, a method that “ Visualizes a model’s predicted vs true plot on given dataset, where each point is colored by a compound’s similarity to the train set”

Note: notice how we use test_dataset, which is oce.ExampleDataset converted into a BaseDataset object.

vis =  oce.VisualizeModelSim(model=loaded_model,dataset=test_dataset)

Finally, every Visualization has a render_ipynb method that can be used to render the visualization. So, we can do the following:

vis.render_ipynb()

:: And your visualization is viewable in a jupyter notebook.

“””