def test_result_length():
null_model = ModelInterface("NULL")
df = pd.DataFrame({
"id": [1, 2, 3],
"text": ["the cat", "the hat", "the mat"]
})
result = explain_predictions(null_model, df, "text", None)
assert len(result) == len(
df), "Explain function returns results for all records"
def test_single_word_model():
jellybean_model = SingleWordModel("JellyBeanModel", "TEXT", "jellybean")
df = pd.DataFrame({
"ID": [1, 2],
"TEXT": ["bob eats jellybeans", "jane likes to swim"]
})
result = explain_predictions(jellybean_model, df, "TEXT", None)
assert len(result) == len(
df), "Explain function returns results for all records"
assert result[0][
0] == 1, "First record contains discriminative word that perfectly explains output."
assert result[1][0] == 0, "Second record cannot be determined"
def test_result_attributes():
null_model = ModelInterface("NULL")
df = pd.DataFrame({"id": [1, 2], "text": ["the cat", "the hat"]})
result = explain_predictions(null_model, df, "text", None)
assert str(
type(result)) == "<class 'list'>", "Explain function returns a list"
assert str(type(
result[0])) == "<class 'tuple'>", "Returned list contains tuples"
#assert str(type(result[0][0])) == "<class 'float'>", "First element is a number"
assert isinstance(result[0][0], float) == True, "First element is a number"
assert str(type(
result[0][1])) == "<class 'str'>", "Second element is a string"
def test_exceptions():
null_model = ModelInterface("NULL")
df = pd.DataFrame({"id": [1, 2], "text": ["the cat", "the hat"]})
thrown = False
try:
result = explain_predictions(null_model, df, 9, None)
except:
thrown = True
assert thrown == True, "Exception should be thrown when column name is not a string"
thrown = False
try:
result = explain_predictions(null_model, df, "textnot", None)
except:
thrown = True
assert thrown == True, "Exception should be thrown when column is not present in dataframe"
thrown = False
try:
result = explain_predictions("garbage", df, "text", None)
except:
thrown = True
assert thrown == True, "Exception should be thrown when model does not implement predict"
def test_multiple_sentences_model():
jellybean_model = SingleWordModel("JellyBeanModel", "TEXT", "jellybean")
df = pd.DataFrame({
"ID": [1, 2],
"TEXT": ["I eat a jellybean. Bob eats a fig.", "Jane likes to swim"]
})
result = explain_predictions(jellybean_model, df, "TEXT", None)
assert len(result) == len(
df), "Explain function returns results for all records"
assert result[0][
0] == 1, "First record contains word that perfectly explains output."
assert result[1][0] == 0, "Second record cannot be determined"
record_one = result[0]
print("record one:", record_one)
record_one_text = result[0][1]
print("record one text:", record_one_text)
assert record_one_text.__contains__(
"{{1.0}}"), "One sentence with full contribution"
assert record_one_text.__contains__(
"jellybean{{1.0}}"), "jellybean has full contribution"
def test_multi_word_model():
mood_model = MultiWordModel("MoodsModel", "TEXT", ["happy", "sad"])
df = pd.DataFrame({
"ID": [1, 2],
"TEXT": [
"bob is very happy today",
"jane is happy most mornings, but sometimes sad after school"
]
})
result = explain_predictions(mood_model, df, "TEXT", None)
assert len(result) == len(
df), "Explain function returns results for all records"
record_one = result[0]
record_one_text = result[0][1]
assert record_one_text.__contains__("happy{{0.5}}"), "happy contribution"
record_two = result[1]
record_two_text = result[1][1]
assert record_two_text.__contains__(
"{{0.5}}"), "Words with partial contribution."
assert record_two_text.__contains__(
"sad{{0.5}}"), "sad has half contribution"