negatives = [j for j in range(n_molecules) if w[j, i] > 0 and y[j, i] == 0]
task_molecules.append(
np.concatenate(
[positives[:5], negatives[:5], positives[5:10], negatives[5:10]]))
// Create the model to train.
After Change
positives = [j for j in range(n_molecules) if w[j, i] > 0 and y[j, i] == 1]
negatives = [j for j in range(n_molecules) if w[j, i] > 0 and y[j, i] == 0]
np.random.shuffle(positives)
np.random.shuffle(negatives)
mols = sum((list(x) for x in zip(positives, negatives)), [])
task_molecules.append(mols)
// Create the model to train. We use a simple fully connected network with
