def test_grid_distance(): reset_all() with open('training/' + os.listdir('training/')[56]) as f: raw_task = json.load(f) task = tuplefy_task(raw_task) input_grid = task['train'][0]['input'] output_grid = task['train'][0]['output'] same_shape = ((8, 8, 0, 8, 8, 0, 0), (0, 8, 0, 0, 8, 0, 0), (8, 8, 8, 8, 8, 8, 0), (0, 0, 0, 0, 0, 0, 0)) different_shape = ((8, 8, 0, 8, 8, 0, 0), (0, 8, 0, 0, 8, 0, 0), (8, 8, 8, 8, 8, 8, 0), (0, 8, 0, 0, 0, 0, 0)) # print(base_entity_finder.grid_distance(output_grid, same_shape), base_entity_finder.grid_distance(output_grid, different_shape)) assert base_entity_finder.grid_distance( output_grid, same_shape) < base_entity_finder.grid_distance( output_grid, different_shape) color_0 = Property(lambda x: frozenset({0}), np.log(10) - 1, name=f'color {0}', output_types=frozenset({'color'}), entity_finder=base_entity_finder) take_color = Property(lambda x: x.entity.colors(), name='the colors', output_types=frozenset({'color'}), entity_finder=base_entity_finder, nll=1) select_not_0 = Selector.make_property_selector(take_color, color_0, False) # min_x = Property(lambda x: x.entity.min_coord(axis=1), np.log(4), # name='the largest x coordinate', # output_types=frozenset({'x_coordinate'}), # entity_finder=base_entity_finder) # max_x = Property(lambda x: x.entity.max_coord(axis=1), np.log(4), # name='the largest x coordinate', # output_types=frozenset({'x_coordinate'}), # entity_finder=base_entity_finder) # x_length = Property.create_distance_property(max_x, min_x) x_length = Property( lambda x: x.entity.max_coord(axis=1) - x.entity.min_coord(axis=1) + 1, np.log(2), name='the x length', output_types=frozenset({'x_length'}), entity_finder=base_entity_finder) zero = Property(lambda x: 0, 1, name='0', output_types=frozenset({'y_length'}), entity_finder=base_entity_finder) train_input = task['train'][0]['input'] train_output = task['train'][0]['output'] entities = base_entity_finder(train_input) assert x_length(entities[1], train_input) == 3 x_length_vect = Property.xy_length_to_vector(zero, x_length) # print(select_not_0.select(entities)) _, prediction = move(select_not_0.select(entities), train_input, x_length_vect, copy=True) assert base_entity_finder.grid_distance(train_output, prediction) < \ base_entity_finder.grid_distance(train_output, train_input)
def test_case_34(): with open('training/' + os.listdir('training/')[34]) as f: raw_task = json.load(f) base_entity_finder = EntityFinder( lambda grid: find_components(grid, directions=ALL_DIRECTIONS)) task = tuplefy_task(raw_task) inp = task['train'][0]['input'] out = task['train'][0]['output'] entities = base_entity_finder(inp) color_8 = Property(lambda x: frozenset({8}), np.log(10) - 1, name=f'color {8}', output_types=frozenset({'color'}), entity_finder=base_entity_finder) color_0 = Property(lambda x: frozenset({0}), np.log(10) - 1, name=f'color {0}', output_types=frozenset({'color'}), entity_finder=base_entity_finder) take_color = Property(lambda x: x.entity.colors(), name='the colors', output_types=frozenset({'color'}), entity_finder=base_entity_finder, nll=1, requires_entity=True) select_8 = Selector.make_property_selector(take_color, color_8, True) select_not_8 = Selector.make_property_selector(take_color, color_8, False) select_not_0 = Selector.make_property_selector(take_color, color_0, False) select_not_0.nll = np.log(2) select_not_0_nor_8 = Selector.intersect(select_not_0, select_not_8) selected_entities = select_not_0_nor_8.select(entities) collision = Relation( lambda entity1, entity2: next( iter(collision_directions(entity1, entity2, adjustment=1))) if len(collision_directions(entity1, entity2)) == 1 else None, nll=1 + np.log(2), name='the unique collision vector to', output_types=frozenset({'vector'})) collision_with_8 = Property.from_relation_selector(collision, select_8, base_entity_finder) move_into_8 = Transformer( lambda entities, grid: move(entities, vector_property=collision_with_8, copy=True, extend_grid=False), nll=collision_with_8.nll + np.log(2), name=f"{'copy' if True else 'move'} them by ({collision_with_8})") new_entities, new_grid = move_into_8.transform(selected_entities, inp) assert new_grid == out my_entity_finder = base_entity_finder.compose(select_not_0_nor_8) my_predictor = Predictor(my_entity_finder, move_into_8) for case in task['train'] + task['test']: assert my_predictor.predict(case['input']) == case['output'] my_predictor_2 = Predictor(base_entity_finder, move_into_8)
def test_composite_selections(): with open('training/' + os.listdir('training/')[205]) as f: raw_cases = json.load(f) cases = tuplefy_task(raw_cases) color_0 = Property(lambda x: frozenset({0}), np.log(2), name=f'color {0}', output_types=frozenset({'color'}), entity_finder=base_entity_finder) color_5 = Property(lambda x: frozenset({5}), np.log(10) - 1, name=f'color {5}', output_types=frozenset({'color'}), entity_finder=base_entity_finder) take_color = Property(lambda x: x.entity.colors(), name='the colors', output_types=frozenset({'color'}), entity_finder=base_entity_finder, nll=1) select_not_0 = Selector.make_property_selector(take_color, color_0, False) select_not_5 = Selector.make_property_selector(take_color, color_5, False) select_not_0_nor_5 = select_not_0.intersect(select_not_5) entity_finder = base_entity_finder.compose(select_not_0_nor_5, True) select_5 = Selector.make_property_selector(take_color, color_5) center_y = Property(lambda x: x.entity.center(axis=0), nll=np.log(2), name='the center y coordinate', output_types=frozenset({'y_coordinate'}), entity_finder=base_entity_finder, requires_entity=True) center_x = Property(lambda x: x.entity.center(axis=1), nll=np.log(2), name='the center x coordinate', output_types=frozenset({'x_coordinate'}), entity_finder=base_entity_finder, requires_entity=True) center_5y = center_y.add_selector(select_5) length_5y = Property.create_distance_property(center_5y, center_y) center_5x = center_x.add_selector(select_5) length_5x = Property.create_distance_property(center_5x, center_x) vect_prop = Property.xy_length_to_vector(length_5y, length_5x) move_to_5 = Transformer( lambda entities, grid, copy=True: move( entities, vector_property=vect_prop, copy=copy, extend_grid=False), nll=vect_prop.nll + np.log(2), name=f"{'copy' if True else 'move'} them by ({vect_prop})") my_predictor = Predictor(entity_finder, move_to_5) for case in cases['train']: assert my_predictor.predict(case['input']) == case['output']
def test_sequential(): with open('training/' + os.listdir('training/')[56]) as f: raw_task = json.load(f) task = tuplefy_task(raw_task) input_grid = task['train'][0]['input'] output_grid = task['train'][0]['output'] color_0 = Property(lambda x: frozenset({0}), np.log(10) - 1, name=f'color {0}', output_types=frozenset({'color'}), entity_finder=base_entity_finder) take_color = Property(lambda x: x.entity.colors(), name='the colors', output_types=frozenset({'color'}), entity_finder=base_entity_finder, nll=1) select_not_0 = Selector.make_property_selector(take_color, color_0, False) x_length = Property( lambda x: x.entity.max_coord(axis=1) - x.entity.min_coord(axis=1) + 1, np.log(2), name='the x length', output_types=frozenset({'x_length'}), entity_finder=base_entity_finder) zero = Property(lambda x: 0, 1, name='0', output_types=frozenset({'y_length'}), entity_finder=base_entity_finder) x_length_vect = Property.xy_length_to_vector(zero, x_length) copy_move_x_length = Transformer( lambda entities, grid: move(entities, grid, x_length_vect, copy=True), name=f'copy them by ({x_length_vect})') my_entity_finder = base_entity_finder.compose(select_not_0) cropper = Transformer(crop_entities, nll=np.log(2), name='crop them') single_predictor = Predictor(my_entity_finder, copy_move_x_length, parallel=False) predictor_1 = Predictor(my_entity_finder, copy_move_x_length) predictor_2 = Predictor(my_entity_finder, cropper) sequential_predictor = Predictor([my_entity_finder, my_entity_finder], [copy_move_x_length, cropper], parallel=False) composed_predictor = predictor_1.compose(predictor_2, parallel=False) train_input = task['train'][0]['input'] train_output = task['train'][0]['output'] print(composed_predictor) assert sequential_predictor.predict(train_input) == train_output assert composed_predictor.predict(train_input) == train_output
def test_transformers_predictors(): with open('training/' + os.listdir('training/')[7]) as f: raw_case7 = json.load(f) case7 = tuplefy_task(raw_case7) inp = case7['train'][0]['input'] out = case7['train'][0]['output'] base_entity_finder = EntityFinder(find_components) entities = base_entity_finder(inp) take_color = Property(lambda x: x.entity.colors(), name='the colors', output_types=frozenset({'color'}), entity_finder=base_entity_finder, nll=1) color_2 = Property(lambda x, i=2: frozenset({2}), np.log(10) - 2, name=f'color {2}', output_types=frozenset({'color'}), entity_finder=base_entity_finder) color_8 = Property(lambda x, i=8: frozenset({8}), np.log(10) - 2, name=f'color {8}', output_types=frozenset({'color'}), entity_finder=base_entity_finder) select_8 = Selector.make_property_selector(take_color, color_8) select_2 = Selector.make_property_selector(take_color, color_2) max_ord = OrdinalProperty(lambda x: nth_ordered(x, 0, use_max=True), nll=0, name=f'take the {1} largest') find_collision_vect_to_8 = Property.from_relation_selector( collision_relation, select_8, entity_finder=base_entity_finder, ordinal_property=max_ord) my_transformer = Transformer( lambda entities, grid: move(entities, vector_property=find_collision_vect_to_8), name=f'move them by ({find_collision_vect_to_8})', nll=1 + np.log(2)) assert my_transformer.transform(select_2.select(entities))[1] == out select_2_finder = base_entity_finder.compose(select_2) my_predictor = Predictor(select_2_finder, my_transformer) assert my_predictor.predict(inp) == out
def test_case_29(): with open('training/' + os.listdir('training/')[29]) as f: raw_task = json.load(f) base_entity_finder = EntityFinder( lambda grid: find_components(grid, directions=ALL_DIRECTIONS)) trivial_selector = Selector(lambda entity, grid: True, name='') task = tuplefy_task(raw_task) inp = task['train'][0]['input'] out = task['train'][0]['output'] # print(task['train'][0]['input']) take_color = Property(lambda x: x.entity.colors(), name='the colors', output_types=frozenset({'color'}), entity_finder=base_entity_finder, nll=1, requires_entity=True) # color_2 = Property(lambda x, i=2: frozenset({2}), np.log(10) - 2, name=f'color {2}', # output_types=frozenset({'color'})) color_1 = Property(lambda x, i=2: frozenset({1}), np.log(10) - 1, name=f'color {1}', output_types=frozenset({'color'}), entity_finder=base_entity_finder) color_0 = Property(lambda x, i=2: frozenset({0}), np.log(10) - 1, name=f'color {0}', output_types=frozenset({'color'}), entity_finder=base_entity_finder) select_1 = Selector.make_property_selector(take_color, color_1) property_0 = Property(lambda x, i=0: i, nll=1, name=f'{0}', output_types=frozenset({ 'x_coordinate', 'y_coordinate', 'x_length', 'y_length', 'quantity' }), entity_finder=base_entity_finder) select_not_0 = Selector.make_property_selector(take_color, color_0, the_same=False) smallest_y = Property(lambda x: x.entity.max_coord(axis=0), 1 + np.log(4), name='the largest y coordinate', output_types=frozenset({'y_coordinate'}), entity_finder=base_entity_finder, requires_entity=True) min_y_of_blue = smallest_y.add_selector(select_1) distance_to_min_y_of_blue = Property.create_distance_property( min_y_of_blue, smallest_y) vector_to_min_y_of_blue = Property.xy_length_to_vector( distance_to_min_y_of_blue, property_0) move_transform = Transformer( lambda entities, grid, vector_prop=vector_to_min_y_of_blue: move( entities, vector_property=vector_prop), nll=vector_to_min_y_of_blue.nll + np.log(2), name=f'move them by ({vector_to_min_y_of_blue})') my_predictor = Predictor(base_entity_finder.compose(trivial_selector), move_transform) # .compose(select_not_0) # display_case(my_predictor.predict(inp)) # display_case(out) assert my_predictor.predict(inp) == out test_input = task['test'][0]['input'] test_output = task['test'][0]['output'] test_entities = base_entity_finder(test_input) assert len(test_entities) == 4 selected_finder = base_entity_finder.compose(select_not_0) # selected_finder(test_input) assert len(selected_finder(test_input)) == 3 assert my_predictor.predict(test_input) == test_output
def create_predictor_queue(task, max_nll, base_entity_finder, allow_selector_pairs=False): for i, example in enumerate(task['train']): if len(base_entity_finder(example['input'])) == 0: return [] start_time = time.perf_counter() selector_list = list( selector_iterator(task, base_entity_finder, max_nll=max_nll - SELECTOR_MAX_NLL_CORRECTION)) selector_list.sort() print(f"selecting time = {time.perf_counter() - start_time}") if MAKE_PROPERTY_LIST: Property.property_list.sort() print(f"len(Property.property_list) = {len(Property.property_list)}") print(f'built selector list (1), length={len(selector_list)}') if allow_selector_pairs: for selector1, selector2 in itertools.combinations(selector_list, 2): if combine_pair_selector_nll( selector1, selector2) < max_nll - SELECTOR_MAX_NLL_CORRECTION: new_selector = selector1.intersect(selector2) if new_selector.validate_and_register( task, base_entity_finder, max_nll - SELECTOR_MAX_NLL_CORRECTION): selector_list.append(new_selector) if time.perf_counter() - start_time > MAX_SMALL_TIME: print('Out of time') return [] selector_list.sort() print(f'built selector list (2), length={len(selector_list)}') # print('Time after selectors created = ', time.perf_counter() - start_time) # Create distance properties out of coordinate properties Property.of_type['x_coordinate'].sort() Property.of_type['y_coordinate'].sort() # LENGTH PROPERTIES x_length_props = (prop1.create_distance_property( prop2, register=False) for prop1, prop2 in combine_sorted_queues(( Property.of_type['x_coordinate'], Property.of_type['x_coordinate']), max_nll - np.log(2)) if prop1.count != prop2.count and ( not prop1.is_constant or not prop2.is_constant)) y_length_props = (prop1.create_distance_property( prop2, register=False) for prop1, prop2 in combine_sorted_queues(( Property.of_type['y_coordinate'], Property.of_type['y_coordinate']), max_nll - np.log(2)) if prop1.count != prop2.count and ( not prop1.is_constant or not prop2.is_constant)) length_props = sorted(list(itertools.chain(x_length_props, y_length_props))) for length_prop in length_props: length_prop.validate_and_register( task, extra_validation=lambda output_signature: all( (value.is_integer() for value in output_signature))) if time.perf_counter() - start_time > MAX_SMALL_TIME: print('Out of time') return [] Property.of_type['x_length'].sort() Property.of_type['y_length'].sort() # Constructing point properties point_props = [ Property.create_point_property(prop1, prop2, register=False) for prop1, prop2 in combine_sorted_queues(( Property.of_type['y_coordinate'], Property.of_type['x_coordinate']), max_nll - 2 - POINT_PROP_COST) ] for point_prop in point_props: point_prop.validate_and_register(task) Property.of_type['point'].sort() if time.perf_counter() - start_time > MAX_SMALL_TIME: print('Out of time') return [] # Constructing vector properties # Create vectors from single lengths for axis, name in enumerate(['y_length', 'x_length']): for length in Property.of_type[name]: vect_prop = Property.length_to_vector(length, axis, register=False) vect_prop.validate_and_register(task) # Create vectors from pairs of points for source_pt, target_pt in combine_sorted_queues( (Property.of_type['point'], Property.of_type['point']), max_nll - np.log(2)): vect_prop = Property.points_to_vector(source_pt, target_pt, register=False) vect_prop.validate_and_register( task, extra_validation=lambda output_signature: all( (value[i].is_integer() for value in output_signature for i in range(2)))) if time.perf_counter() - start_time > MAX_SMALL_TIME: print('Out of time') return [] penalize_dim_change = True if all( (len(case['input']) == len(case['output']) and len(case['input'][0]) == len(case['output'][0]) for case in task['train'])) else False transformers = ( # 34 Transformer( lambda entities, grid, vector_prop=vector_prop, copy=copy: move( entities, vector_property=vector_prop, copy=copy, extend_grid=not penalize_dim_change), nll=vector_prop.nll + np.log(2), name=f"{'copy' if copy else 'move'} them by ({vector_prop})") for vector_prop in Property.of_type['vector'] for copy in [True, False] if vector_prop.nll + np.log(2) <= max_nll) if time.perf_counter() - start_time > MAX_SMALL_TIME: print('Out of time') return [] Property.of_type['color'].sort() # 35 composite_transformers = (Transformer(lambda entities, grid, offsets=offsets: crop_entities(entities, grid, offsets=offsets), nll=np.log(2) + sum( (abs(offset) for offset in offsets)) * np.log(2) + \ penalize_dim_change * DIM_CHANGE_PENALTY, name=f'crop them with offset {offsets}') for offsets in itertools.product([-1, 0, 1], repeat=4) if np.log(2) + sum((abs(offset) for offset in offsets)) * np.log(2) + \ penalize_dim_change * DIM_CHANGE_PENALTY < max_nll) if any(({entry for row in case['input'] for entry in row } == {entry for row in case['output'] for entry in row} for case in task['train'])): new_colors = False