first_output = None
for i, data in enumerate(test_loader):
inputs, output = data
inputs = inputs
inputs_text = test_loader.dataset.bow_column_name.tensor_to_text(inputs)
prediction = network(inputs).data
prediction_text = test_loader.dataset.bow_column_data_type.tensor_to_text(
prediction)
actual = output
actual_text = test_loader.dataset.bow_column_data_type.tensor_to_text(
actual)
if (prediction_text == actual_text):
print.success(inputs_text, "successfully interpeted as",
prediction_text)
successes += 1
else:
print.fail(
inputs_text,
"interpeted as",
prediction_text,
"SHOULD BE",
actual_text,
)
fails += 1
print("Summary successes", successes, "/", successes + fails)
train_datatypes, return_counts=True)
print.info("Lets look at the statistics of the training dataset")
sort_index = np.argsort(-train_unique_datatypes_count)
y_pos = np.arange(len(train_unique_datatypes))
plt.bar(y_pos,
train_unique_datatypes_count[sort_index],
align='center',
alpha=0.5)
plt.xticks(y_pos, train_unique_datatypes[sort_index], rotation=90)
plt.subplots_adjust(bottom=0.35)
plt.ylabel('Usage')
plt.title('Data type')
plt.show()
# How many matches if we select a random datatype?
correctGuessRandom = list(
filter(
lambda item: item['datatype'] == np.random.choice(
train_unique_datatypes), test))
print.info("Random datatype")
print.success("{0:.0%}".format(len(correctGuessRandom) / len(test)), "matches")
# How many matches if we just assume string?
correctGuessString = list(
filter(lambda item: item['datatype'] == "string", test))
print.info("Assuming string")
print.success("{0:.0%}".format(len(correctGuessString) / len(test)), "matches")
