print('Confusion Matrix')
print('================')
cm = confusion_matrix(test_y, pred_y)
print(cm)
# plt.figure()
# plot_confusion_matrix(cm)
#
## plt.show()
print('\nCalculating predictions on Crowdflower')
print('--------------------------------------')
cf_x, cf_y = prep.get_crowdflower(1000)
pred_cf = pipe.predict(cf_x)
print()
print(classification_report(cf_y, pred_cf))
print()
print("Accuracy score")
print("==============")
print(accuracy_score(cf_y, pred_cf))
print()
print('Confusion Matrix')
print('================')
cm = confusion_matrix(cf_y, pred_cf)
tick_marks = np.arange(len(CM_LABELS))
plt.xticks(tick_marks, CM_LABELS, rotation=45)
plt.yticks(tick_marks, CM_LABELS)
plt.tight_layout()
plt.ylabel("True label")
plt.xlabel("Predicted label")
NUM_PER_CATEGORY = 1000
print("loading data....")
print(str(NUM_PER_CATEGORY) + " examples per category")
start_time = time.time()
train_x, train_y = prep.get_data(NUM_PER_CATEGORY)
train_x = train_x.astype("float32")
test_x, test_y = prep.get_crowdflower(NUM_PER_CATEGORY)
test_x = test_x.astype("float32")
print("Total data load time:")
print("---------------------")
print(time.time() - start_time)
os.system('say "data is loaded"')
# Consider trying different values for output_layers
print("\nstarting nn trained on twitter with logit @ -2....")
tf = OverfeatTransformer(output_layers=[-2])
clf = LogisticRegression()
# clf = SVC()
# clf = RandomForestClassifier()
pipe = make_pipeline(tf, clf)