def test_train_mlp_mock():
directory = "/tmp"
file_path = os.path.join(directory, const.MLPMODEL_FILE)
mlp_clf = MLPMultiClass(
model_dir=directory,
dest="output.intents",
debug=False,
)
train_df = pd.DataFrame([
{
"data": "yes",
"labels": "_confirm_"
},
{
"data": "yea",
"labels": "_confirm_"
},
{
"data": "no",
"labels": "_cancel_"
},
{
"data": "nope",
"labels": "_cancel_"
},
])
mlp_clf.train(train_df)
assert mlp_clf.labels_num == 2
# This copy loads from the same directory that was trained previously.
# So this instance would have read the saved mlp model.
mlp_clf_copy = MLPMultiClass(
model_dir=directory,
dest="output.intents",
debug=False,
)
mlp_clf_copy.load()
assert isinstance(mlp_clf_copy.model_pipeline, sklearn.pipeline.Pipeline)
assert mlp_clf_copy.valid_mlpmodel is True
os.remove(file_path)
def test_train_mlp_gridsearch_mock():
directory = "/tmp"
file_path = os.path.join(directory, const.MLPMODEL_FILE)
USE = "use"
fake_args = {
const.TRAIN: {
const.NUM_TRAIN_EPOCHS: 10,
const.USE_GRIDSEARCH: {
USE: True,
const.CV: 2,
const.VERBOSE_LEVEL: 2,
const.PARAMS: {
"activation": ["relu", "tanh"],
"hidden_layer_sizes": [(1, ), (2, 1)],
"ngram_range": [(1, 1)],
"max_iter": [1, 2],
},
},
},
const.TEST: {},
const.PRODUCTION: {},
}
mlp_clf = MLPMultiClass(
model_dir=directory,
dest="output.intents",
debug=False,
args_map=fake_args,
)
train_df = pd.DataFrame([
{
"data": "yes",
"labels": "_confirm_"
},
{
"data": "yea",
"labels": "_confirm_"
},
{
"data": "no",
"labels": "_cancel_"
},
{
"data": "nope",
"labels": "_cancel_"
},
])
mlp_clf.train(train_df)
assert mlp_clf.labels_num == 2
# This copy loads from the same directory that was trained previously.
# So this instance would have read the saved mlp model.
mlp_clf_copy = MLPMultiClass(
model_dir=directory,
dest="output.intents",
debug=False,
)
mlp_clf_copy.load()
assert mlp_clf_copy.valid_mlpmodel is True
mlp_clf_copy.train(
train_df) # When trying to train an already trained model
os.remove(file_path)