def compare_models(runs,
k,
model_generator,
base_output_folder,
test_prepositions=PREPOSITION_LIST):
study_info = StudyInfo("2019 study")
m = MultipleRuns(model_generator,
base_output_folder + "tables",
base_output_folder + "plots",
study_info,
test_prepositions=test_prepositions,
number_runs=runs,
k=k,
compare="y")
models_to_test = m.Generate_Models_all_scenes.models
t = TestModels(models_to_test, "all")
all_dataframe = t.score_dataframe.copy()
print(all_dataframe)
all_dataframe.to_csv(m.all_csv)
print(("Test Model k = " + str(k)))
m.validation()
m.output()
print(m.average_dataframe)
def main():
print(tf.__version__)
# should_be_one = [1.33, -0.60, 1.99, 0.40, -0.75, 2.60, -0.95, 0.45, 1.51, 1.84]
# print(tf.shape(should_be_one))
# print(tf.shape([should_be_one]))
study_info = StudyInfo("2019 study")
plot_all_csv()
def test_k_fold(self):
study_info = StudyInfo("2019 study")
m = MultipleRuns(GeneratePolysemeModels,
POLYSEMY_SCORES_FOLDER + "tables",
POLYSEMY_SCORES_FOLDER + "plots",
study_info,
number_runs=10,
k=10,
compare="y",
test_prepositions=POLYSEMOUS_PREPOSITIONS)
self.assertIsInstance(m.Generate_Models_all_scenes,
GeneratePolysemeModels)
self.assertIsInstance(m.Generate_Models_all_scenes.features_to_remove,
list)
self.assertEqual(m.Generate_Models_all_scenes.features_to_remove,
Configuration.object_specific_features.copy())
self.assertIsInstance(m.model_name_list, list)
test_folds = m.get_validation_scene_split()
while not m.folds_check(test_folds):
test_folds = m.get_validation_scene_split()
self.assertEqual(len(test_folds), 10)
for f1 in test_folds:
self.assertNotEqual(len(f1), 0)
for f2 in test_folds:
if f1 != f2:
f1_set = set(f1)
f2_set = set(f2)
self.assertFalse(f1_set & f2_set)
def output_model_params():
s_info = StudyInfo("2019 study")
all_scenes = s_info.scene_name_list
generated_polyseme_models = GenerateOSModels(all_scenes, all_scenes, s_info, PREPOSITION_LIST)
# for m in generated_polyseme_models.models:
# for preposition in PREPOSITION_LIST:
# m.output_typicalities(preposition)
#
# m.output_unsatisfied_constraints()
generated_polyseme_models.distinct_supervised_model.output_polyseme_info()
generated_polyseme_models.sense_model.output_polyseme_info()
def verify_sup_model():
# Show the model trains well
study_info = StudyInfo("2019 study")
scene_list = study_info.scene_name_list
features_to_remove = Configuration.object_specific_features.copy()
train_scenes, test_scenes = scene_list, scene_list
model = SupervisedNeuralTypicalityModel(train_scenes,
test_scenes,
study_info,
features_to_remove,
train_test_proportion=0.8,
number_of_epochs=1000)
t = TestModels([model], "all")
all_dataframe = t.score_dataframe.copy()
print(all_dataframe)
def verify_dnn_model():
# Show the model trains well
study_info = StudyInfo("2019 study")
scene_list = study_info.scene_name_list
preposition_models_dict = get_standard_preposition_parameters(scene_list)
dnn_model = NeuralNetworkCategorisationModel(preposition_models_dict,
scene_list,
study_info,
make_plots=True,
train_test_proportion=0.8,
number_of_epochs=1000)
t = TestModels([dnn_model], "all")
all_dataframe = t.score_dataframe.copy()
print(all_dataframe)
def test_polyseme_rank_info(self):
'''
:return:
'''
study_info = StudyInfo("2019 study")
all_scenes = study_info.scene_name_list
generated_polyseme_models = GeneratePolysemeModels(
all_scenes, all_scenes, study_info)
# Check ranks
generated_polyseme_models.non_shared.output_polyseme_info(
base_folder=output_folder)
model_name = GeneratePolysemeModels.distinct_model_name
for preposition in POLYSEMOUS_PREPOSITIONS:
new_rank_csv = output_folder + POLYSEMY_MODEL_PROPERTY_FOLDER + model_name + "/ranks/" + preposition + " -ranks.csv"
new_rank_df = pd.read_csv(new_rank_csv)
original_rank_df = pd.read_csv(get_original_csv(new_rank_csv))
cols = new_rank_df.columns.tolist()
redordered_orig_df = original_rank_df[cols]
assert_frame_equal(new_rank_df, redordered_orig_df)
def test_initial_model(self):
cluster_numbers = KMeansPolysemyModel.cluster_numbers
self.assertEqual(cluster_numbers["on"], 8)
self.assertEqual(cluster_numbers["in"], 4)
self.assertEqual(cluster_numbers["over"], 4)
self.assertEqual(cluster_numbers["under"], 4)
study_info = StudyInfo("2019 study")
all_scenes = study_info.scene_name_list
generated_polyseme_models = GeneratePolysemeModels(
all_scenes,
all_scenes,
study_info,
test_prepositions=POLYSEMOUS_PREPOSITIONS)
p_models = generated_polyseme_models.models
archive_all_csv = archive_folder + "2019 study/polysemy/scores/all_test.csv"
original_dataframe = pd.read_csv(archive_all_csv, index_col=0)
print(original_dataframe)
t = TestModels(p_models, "all")
new_dframe = t.score_dataframe
print(new_dframe)
# reindex original as it contains shared aswell but new doesn't
new_reindexed, original_reindexed = dropcolumns_reindexlike(
new_dframe, original_dataframe)
try:
assert_frame_equal(new_reindexed, original_reindexed)
except AssertionError as e:
print(e)
# first check basic parametres
for model in p_models:
self.assertEqual(len(model.all_feature_keys), 16)
if hasattr(model, "preposition_model_dict"):
for p in model.test_prepositions:
self.assertEqual(
len(model.preposition_model_dict[p].feature_keys), 10)
# Check typicalities
for model in p_models:
for preposition in POLYSEMOUS_PREPOSITIONS:
typ_csv = output_folder + model.study_info.base_polysemy_folder + "config typicalities/typicality-" + preposition + ".csv"
model.output_typicalities(preposition, input_csv=typ_csv)
for preposition in POLYSEMOUS_PREPOSITIONS:
# Remove Kmeans column as it is not deterministic
new_typicality_csv = output_folder + generated_polyseme_models.study_info.base_polysemy_folder + "config typicalities/typicality-" + preposition + ".csv"
new_typicality_df = pd.read_csv(new_typicality_csv,
usecols=[0, 1, 2, 3, 4, 5, 6])
original_typicality_df = pd.read_csv(
get_original_csv(new_typicality_csv),
usecols=[0, 1, 2, 3, 4, 5, 6])
columns_to_check = [
'scene', 'figure', 'ground', 'Distinct Prototype',
'Baseline Model'
]
print(original_typicality_df.columns.tolist())
print(new_typicality_df.columns.tolist())
# reindex original as it contains shared aswell but new doesn't
original_reindexed = original_typicality_df[columns_to_check]
new_reindexed = new_typicality_df[columns_to_check]
assert_frame_equal(new_reindexed, original_reindexed)
print(original_reindexed.columns.tolist())
print(new_reindexed.columns.tolist())