Loading deep_chem/utils/preprocess.py +1 −6 Original line number Diff line number Diff line Loading @@ -49,15 +49,10 @@ def undo_normalization(y_orig, y_pred): """Undo the applied normalization transform.""" old_mean = np.mean(y_orig) old_std = np.std(y_orig) return y_orig * old_std + old_mean return y_pred * old_std + old_mean def undo_transform_outputs(y_raw, y_pred, output_transforms): """Undo transforms on y_pred, W_pred.""" print "undo_transform_outputs()" print "output_transforms" print output_transforms print "y_raw" print y_raw if output_transforms == ["log"]: return np.exp(y_pred) elif output_transforms == ["normalize"]: Loading deep_chem/utils/test/__init__.py 0 → 100644 +0 −0 Empty file added. deep_chem/utils/test/test_preprocess.py +26 −0 Original line number Diff line number Diff line Loading @@ -8,6 +8,7 @@ __license__ = "LGPL" import numpy as np import unittest from deep_chem.utils.preprocess import balance_positives from deep_chem.utils.preprocess import undo_transform_outputs def ensure_balanced(y, W): """Helper function that ensures postives and negatives are balanced.""" Loading Loading @@ -39,3 +40,28 @@ class TestPreprocess(unittest.TestCase): pos_weight += Wbal[sample_ind, target_ind] assert np.isclose(pos_weight, neg_weight) def test_undo_transform_outputs(self): # Test undo-log y_raw = np.ones(10) y_pred = np.log(y_raw) output_transforms = ["log"] assert np.array_equal(y_raw, undo_transform_outputs(y_raw, y_pred, output_transforms)) # Test undo-normalization y_raw = np.random.randint(0, 10, size=(10,)) mean = np.mean(y_raw) std = np.std(y_raw) y_pred = (y_raw-mean)/std output_transforms = ["normalize"] y_ret = undo_transform_outputs(y_raw, y_pred, output_transforms) assert np.allclose(y_raw, y_ret) # Test undo log-normalization y_raw = np.random.randint(1, 10, size=(10,)) y_pred = np.log(y_raw) mean = np.mean(y_pred) std = np.std(y_pred) y_pred = (y_pred - mean)/std output_transforms = ["log", "normalize"] y_ret = undo_transform_outputs(y_raw, y_pred, output_transforms) assert np.allclose(y_raw, y_ret) Loading
deep_chem/utils/preprocess.py +1 −6 Original line number Diff line number Diff line Loading @@ -49,15 +49,10 @@ def undo_normalization(y_orig, y_pred): """Undo the applied normalization transform.""" old_mean = np.mean(y_orig) old_std = np.std(y_orig) return y_orig * old_std + old_mean return y_pred * old_std + old_mean def undo_transform_outputs(y_raw, y_pred, output_transforms): """Undo transforms on y_pred, W_pred.""" print "undo_transform_outputs()" print "output_transforms" print output_transforms print "y_raw" print y_raw if output_transforms == ["log"]: return np.exp(y_pred) elif output_transforms == ["normalize"]: Loading
deep_chem/utils/test/test_preprocess.py +26 −0 Original line number Diff line number Diff line Loading @@ -8,6 +8,7 @@ __license__ = "LGPL" import numpy as np import unittest from deep_chem.utils.preprocess import balance_positives from deep_chem.utils.preprocess import undo_transform_outputs def ensure_balanced(y, W): """Helper function that ensures postives and negatives are balanced.""" Loading Loading @@ -39,3 +40,28 @@ class TestPreprocess(unittest.TestCase): pos_weight += Wbal[sample_ind, target_ind] assert np.isclose(pos_weight, neg_weight) def test_undo_transform_outputs(self): # Test undo-log y_raw = np.ones(10) y_pred = np.log(y_raw) output_transforms = ["log"] assert np.array_equal(y_raw, undo_transform_outputs(y_raw, y_pred, output_transforms)) # Test undo-normalization y_raw = np.random.randint(0, 10, size=(10,)) mean = np.mean(y_raw) std = np.std(y_raw) y_pred = (y_raw-mean)/std output_transforms = ["normalize"] y_ret = undo_transform_outputs(y_raw, y_pred, output_transforms) assert np.allclose(y_raw, y_ret) # Test undo log-normalization y_raw = np.random.randint(1, 10, size=(10,)) y_pred = np.log(y_raw) mean = np.mean(y_pred) std = np.std(y_pred) y_pred = (y_pred - mean)/std output_transforms = ["log", "normalize"] y_ret = undo_transform_outputs(y_raw, y_pred, output_transforms) assert np.allclose(y_raw, y_ret)
