def upload_dataset_to_project(con: h2o.Client, project_key: str,
dataset_file: str, dataset_type: str):
"""
Uploads the data provided in dataset_file path to Driverless AI and links to the project. If the project already
has a dataset of the specified type and filename linked, then it is not re-uploaded. For the uploaded dataset, the
dataset_key of the newly uploaded dataset is returned. If it is not uploaded, then key of the dataset matching the
file name is returned.
:param con: Connection to H2O Driverless AI
:param project_key: Key of the project to link the dataset to
:param dataset_file: File path of the dataset to upload and link to project
:param dataset_type: Either 'Training' or 'Testing'
:return: dataset_key
"""
file_name = os.path.basename(dataset_file)
datasets = con.get_datasets_for_project(project_key, dataset_type)
dataset = next((x for x in datasets if x.name == file_name), None)
if dataset is None:
dataset = con.upload_dataset_sync(file_path=dataset_file)
con.link_dataset_to_project(project_key=project_key,
dataset_key=dataset.key,
dataset_type=dataset_type)
return dataset.key
def test_debug_pyclient():
from h2oai_client import Client
pd.set_option('display.max_rows', 50)
pd.set_option('display.max_columns', 500)
pd.set_option('display.width', 1000)
# Login info
dai_url = "http://****:12345"
dai_user = "******"
dai_pwd = "****"
# Data Information
data_file_name = "****.csv"
y = "****"
# Transformers information
transformer_file_name = "****.py"
transformers_noncustom = []
transformers_custom_nontesting = []
# All Offical Transformers
transformers_noncustom = ['CVCatNumEncode', 'CVTargetEncode'
, 'CatOriginalTransformer', 'ClusterDistTransformer'
, 'ClusterIdTransformer', 'ClusterTETransformer', 'DatesTransformer'
, 'EwmaLagsTransformer', 'FrequentTransformer', 'InteractionsTransformer'
, 'IsHolidayTransformer', 'LagsAggregatesTransformer', 'LagsInteractionTransformer'
, 'LagsTransformer', 'LexiLabelEncoder', 'NumCatTETransformer', 'NumToCatTETransformer'
, 'NumToCatWoEMonotonicTransformer', 'NumToCatWoETransformer', 'OneHotEncodingTransformer'
, 'OriginalTransformer', 'SortedLETransformer', 'StrFeatureTransformer', 'TextClustDistTransformer'
, 'TextClustTETransformer', 'TextLinModelTransformer', 'TextTransformer', 'TruncSVDNumTransformer'
, 'WeightOfEvidenceTransformer']
# Any Installed Custom Transformers you don't want to test
transformers_custom_nontesting = ['MyLogTransformer']
all_nontest_transformers = transformers_noncustom + transformers_custom_nontesting
# STEP ZERO: Connect to Driverless AI
h2oai = Client(dai_url, dai_user, dai_pwd)
# STEP ONE: Load data set (and related tasks)
# view all data sets in DAI
all_data_sets = h2oai.list_datasets(0, 100)
all_data_sets = pd.DataFrame({
'key': list(map(lambda x: x.key, all_data_sets))
, 'name': list(map(lambda x: x.name, all_data_sets))})
print("PRE-LOADED DATASETS:")
print(all_data_sets)
# check if data was pre-loaded - if so use that data set - if not load data
if data_file_name in all_data_sets['name'].values:
print("\nData already loaded ", data_file_name)
data_key = all_data_sets[all_data_sets["name"] == data_file_name]["key"][0]
data_load_job = h2oai.get_dataset_job(data_key).entity
else:
print("\nLoading file ", data_file_name)
data_load_job = h2oai.upload_dataset_sync(data_file_name)
data_key = data_load_job.key
# STEP TWO: Load custom transformer (and related tasks)
# probably not good to just upload every time
# no function to delete from python, only from ssh-ing in
# rm tmp/contrib/transformers/[function]_randomletters_content.py
print("\nUploading Transformer ", transformer_file_name)
my_transformer = h2oai.upload_custom_recipe_sync(transformer_file_name)
# returns true or false - exit if fails - check DAI UI for error message
if my_transformer:
print("\nTransformer uploaded successfully\n")
else:
print("\nTransformer uploaded failed, exiting program.\n")
sys.exit()
# STEP THREE: Run experiment (and related tasks)
print("\nStarting Experiment\n")
experiment = h2oai.start_experiment_sync(
dataset_key=data_key
, target_col=y
, is_classification=True
, accuracy=1
, time=1
, interpretability=10
, scorer="F1"
, score_f_name=None
, config_overrides="""
feature_brain_level=0
exclude_transformers={dont_use}
""".format(dont_use=all_nontest_transformers)
)
# experiment = h2oai.get_model_job("lomotare").entity
# STEP FOUR: Check the transformation was used
# Download Summary
summary_path = h2oai.download(src_path=experiment.summary_path, dest_dir=".")
dir_path = "h2oai_experiment_summary_" + experiment.key
import zipfile
with zipfile.ZipFile(summary_path, 'r') as z:
z.extractall(dir_path)
# View Features
features = pd.read_table(dir_path + "/features.txt", sep=',', skipinitialspace=True)
print(features)
# STEP FIVE: Transform data and ensure it looks as expected
transform = h2oai.fit_transform_batch_sync(model_key=experiment.key
, training_dataset_key=data_key
, validation_dataset_key=None
, test_dataset_key=None
, validation_split_fraction=0.25
, seed=1234
, fold_column=None)
# Download the training and validation transformed data
transform_train_path = h2oai.download(src_path=transform.training_output_csv_path, dest_dir=".")
transform_validate_path = h2oai.download(src_path=transform.validation_output_csv_path, dest_dir=".")
transform_train = pd.read_table(transform_train_path, sep=',', skipinitialspace=True)
transform_validate = pd.read_table(transform_validate_path, sep=',', skipinitialspace=True)
print(transform_train.head())
print(transform_validate.head())
# STEP 1000: Clean up
os.remove(summary_path)
os.remove(transform_train_path)
os.remove(transform_validate_path)
shutil.rmtree(dir_path)