def test_right_answers(self):
input = np.arange(0, 8 * 20)
input1 = np.expand_dims(input.reshape(8, 20), 0).repeat(6, 0)
input2 = np.expand_dims(input.reshape(20, 8).T, 0).repeat(6, 0)
out1 = get_top_k_preds(input1, 3)
out2 = get_top_k_preds(input2, 3)
self.assertTrue(all((np.stack(i) - np.stack(out1[0])).sum() == 0 for i in out1), "Not stable across batch")
self.assertTrue(all((np.stack(i) - np.stack(out2[0])).sum() == 0 for i in out2), "Not stable across batch")
self.assertTrue(np.array_equal(np.array([[19, 19, 19, 19, 19, 19, 19, 19],
[19, 19, 19, 19, 19, 19, 19, 18],
[19, 19, 19, 19, 19, 19, 19, 17]]), np.stack(out1[0])))
self.assertTrue(np.array_equal(np.array([[19, 19, 19, 19, 19, 19, 19, 19],
[19, 19, 19, 19, 19, 19, 19, 18],
[19, 19, 19, 19, 19, 19, 18, 19]]), np.stack(out2[0])))
def unbatchify_top_k(self, batch, model_output, k):
"""
Returns a list of the k most probable outputs of the model (in order) and labels, all as lists of strings
"""
_, real_names = batch.label
predictions_labels = []
output_preds = get_top_k_preds(model_output.asnumpy(), k)
for i, topk in enumerate(output_preds):
topk_preds = []
for row in topk:
prediction = []
for pred in row:
if pred == self.data_encoder.all_node_name_subtokens[
'__PAD__']:
continue
if pred < len(self.data_encoder.all_node_name_subtokens):
prediction.append(self.data_encoder.
rev_all_node_name_subtokens[pred])
else:
prediction.append(
batch.data.graph_vocab_node_real_names[i]
[pred -
len(self.data_encoder.all_node_name_subtokens)])
topk_preds.append(prediction)
predictions_labels.append(
(topk_preds,
self.data_encoder.name_to_subtokens(real_names[i])))
return predictions_labels
def test_most_likely(self, input):
if not np.any(np.isnan(input)):
most_likely = np.zeros(input.shape[:2])
for i in range(input.shape[0]):
for j in range(input.shape[1]):
most_likely[i, j] = input[i, j, get_top_k_preds(input, 1)[i][0][j]]
self.assertTrue(np.array_equal(np.max(input, axis=2), most_likely))
def test_validate_inputs(self):
with self.assertRaises(AssertionError):
get_top_k_preds(np.zeros((50, 8, 100)), 101)
with self.assertRaises(AssertionError):
get_top_k_preds(np.zeros((50, 8, 100)), 0)
get_top_k_preds(np.zeros((50, 8, 100)), 100)