Naive Baseline Model: No difference between folds?

[TheresaSchmidt]

As far as I can see, there's no difference between the folds. (We would expect different train/dev/test splits.)

alignment-models/train.py

Lines 812 to 916 in 8165592

	def run_naive_folds_train( self,
	model,
	num_folds
	):
	"""
	Running 10 fold cross validation for naive baseline
	Parameters
	----------
	model : NaiveModel object
	Naive Baseline model
	num_folds : Int
	"""
	dish_list = os.listdir(folder)
	dish_list = [dish for dish in dish_list if not dish.startswith(".")]
	dish_list.sort()
	fold_result_df = pd.DataFrame(
	columns=[
	"Fold",
	"Test_Accuracy",
	"Correct_Predictions",
	"Num_Actions",
	]
	) # , "Test_Dish1_accuracy", "Test_Dish2_accuracy"])
	destination_folder = destination_folder4
	overall_predictions = 0
	overall_actions = 0
	for fold in range(num_folds):
	start = datetime.now()
	saved_file_path = os.path.join(
	destination_folder, "model" + str(fold + 1) + ".pt"
	) # Model saved path
	train_dish_list = dish_list.copy()
	print("Fold [{}/{}]".format(fold + 1, num_folds))
	print("-------Training-------")
	self.basic_training(
	model,
	train_dish_list,
	saved_file_path,
	)
	overall_predictions += total_correct_predictions
	overall_actions += total_actions
	fold_result = {
	"Fold": fold + 1,
	"Test_Accuracy": test_accuracy,
	"Correct_Predictions": total_correct_predictions,
	"Num_Actions": total_actions,
	} # ,
	# "Test_Dish1_accuracy" : test_accuracy_list[0][2],
	# "Test_Dish2_accuracy" : test_accuracy_list[1][2]}
	fold_result_df = fold_result_df.append(fold_result, ignore_index=True)
	end = datetime.now()
	elapsedTime = end - start
	elapsed_duration = divmod(elapsedTime.total_seconds(), 60)
	print(
	"Time elapsed: {} mins and {:.2f} secs".format(
	elapsed_duration[0], elapsed_duration[1]
	)
	)
	print("--------------")
	overall_accuracy = overall_predictions * 100 / overall_actions
	print("Overall Model Accuracy: {:.2f}".format(overall_accuracy))
	fold_result = {
	"Fold": 'Overall',
	"Test_Accuracy": overall_accuracy,
	"Correct_Predictions": overall_predictions,
	"Num_Actions": overall_actions,
	}
	fold_result_df = fold_result_df.append(fold_result, ignore_index=True)
	save_result_path = os.path.join(destination_folder, "fold_results.tsv")
	results_file_path = os.path.join(
	destination_folder, "model_result.tsv"
	) # Model saved path
	# Saving the results
	fold_result_df.to_csv(save_result_path, sep="\t", index=False, encoding="utf-8")
	print("Fold Results saved in ==>" + save_result_path)

[irisferrazzo]

Hi @TheresaSchmidt , only the alignment model "base" & "with-feature" have been adapted. The other versions are still in the original version.