EXAMPLE_PREDS_COL
)
# download all the things
napi = NumerAPI()
current_round = napi.get_current_round()
# Tournament data changes every week so we specify the round in their name. Training
# and validation data only change periodically, so no need to download them every time.
print('Downloading dataset files...')
Path("./v4").mkdir(parents=False, exist_ok=True)
napi.download_dataset("v4/train.parquet")
napi.download_dataset("v4/validation.parquet")
napi.download_dataset("v4/live.parquet", f"v4/live_{current_round}.parquet")
napi.download_dataset("v4/validation_example_preds.parquet")
napi.download_dataset("v4/features.json")
print('Reading minimal training data')
# read the feature metadata and get a feature set (or all the features)
with open("v4/features.json", "r") as f:
feature_metadata = json.load(f)
# features = list(feature_metadata["feature_stats"].keys()) # get all the features
# features = feature_metadata["feature_sets"]["small"] # get the small feature set
features = feature_metadata["feature_sets"]["medium"] # get the medium feature set
# read in just those features along with era and target columns
read_columns = features + [ERA_COL, DATA_TYPE_COL, TARGET_COL]
# a value of 10 means use every 10th row
downsample_cross_val = 20
downsample_full_train = 2
# if model_selection_loop=True get OOS performance for training_data
# and use that to select best model
# if model_selection_loop=False, just predict on tournament data using existing models and model config
model_selection_loop = True
model_config_name = "advanced_example_model"
napi = NumerAPI()
current_round = napi.get_current_round()
Path("./v4").mkdir(parents=False, exist_ok=True)
napi.download_dataset("v4/train.parquet")
napi.download_dataset("v4/features.json")
print("Entering model selection loop. This may take awhile.")
if model_selection_loop:
model_config = {}
print('reading training_data')
training_data = pd.read_parquet('v4/train.parquet')
# keep track of some prediction columns
ensemble_cols = set()
pred_cols = set()
# pick some targets to use
possible_targets = [
c for c in training_data.columns if c.startswith("target_")
