import streamlit as st st.title('Numerai Example Script') # content below adapted from # https://github.com/numerai/example-scripts/blob/master/example_model.py # import pandas as pd from lightgbm import LGBMRegressor import gc import json from pathlib import Path import os from numerapi import NumerAPI from utils import ( save_model, load_model, neutralize, get_biggest_change_features, validation_metrics, ERA_COL, DATA_TYPE_COL, TARGET_COL, EXAMPLE_PREDS_COL ) IS_RUNNING_IN_HUGGING_FACE = os.environ.get('HF_ENDPOINT') is not None 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. Path("./v4").mkdir(parents=False, exist_ok=True) @st.cache def get_dataset_path(): if IS_RUNNING_IN_HUGGING_FACE: from datasets import load_dataset_builder ds_builder = load_dataset_builder("Numerati/numerai-datasets") return ds_builder.cache_dir else: return "./v4" @st.cache def download_dataset(): print('download_dataset') if IS_RUNNING_IN_HUGGING_FACE: napi.download_dataset("v4/train.parquet") napi.download_dataset("v4/validation.parquet") napi.download_dataset("v4/validation_example_preds.parquet") napi.download_dataset("v4/features.json") napi.download_dataset("v4/live.parquet", f"v4/live_{current_round}.parquet") print('done download_dataset') @st.cache def load_dataset(feature_set: str): dataset_path = get_dataset_path() print(f'load_dataset with feature_set {feature_set} and path {dataset_path}') # read the feature metadata and get a feature set (or all the features) with open(f"{dataset_path}/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"][feature_set] # 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] # note: sometimes when trying to read the downloaded data you get an error about invalid magic parquet bytes... # if so, delete the file and rerun the napi.download_dataset to fix the # corrupted file training_data = pd.read_parquet(f'{dataset_path}/train.parquet', columns=read_columns) validation_data = pd.read_parquet(f'{dataset_path}/validation.parquet', columns=read_columns) live_data = pd.read_parquet(f'v4/live_{current_round}.parquet', columns=read_columns) # pare down the number of eras to every 4th era # every_4th_era = training_data[ERA_COL].unique()[::4] # training_data = training_data[training_data[ERA_COL].isin(every_4th_era)] # getting the per era correlation of each feature vs the target all_feature_corrs = training_data.groupby(ERA_COL).apply( lambda era: era[features].corrwith(era[TARGET_COL]) ) # find the riskiest features by comparing their correlation vs # the target in each half of training data; we'll use these later riskiest_features = get_biggest_change_features(all_feature_corrs, 50) # "garbage collection" (gc) gets rid of unused data and frees up memory gc.collect() print('done with feature_set', feature_set) return training_data, validation_data, live_data, features, riskiest_features feature_set = st.selectbox( 'Which feature set should be used?', ('small', 'medium', 'fncv3_features', 'v2_equivalent_features', 'v3_equivalent_features')) data_load_state = st.text('Loading data...') download_dataset() training_data, validation_data, live_data, features, riskiest_features = load_dataset(feature_set) data_load_state.text('Loading data...done!') st.subheader('Raw data') st.write(training_data.head()) st.subheader('Model Configuration') n_estimators = st.slider('n_estimators', 100, 10000, 2000) learning_rate = st.slider('learning_rate', 0.0001, 0.1, 0.01) max_depth = st.slider('max_depth', 2, 20, 5) params = {"n_estimators": n_estimators, "learning_rate": learning_rate, "max_depth": max_depth, "num_leaves": 2 ** 5, "colsample_bytree": 0.1 } model_name = f"model_target" @st.cache def get_model_and_fit(model_name, *params): print('get_model_and_fit') model = load_model(model_name) if not model: with st.spinner('Wait model training...'): print(f"model not found, creating new one") model = LGBMRegressor(**params) # train on all of train and save the model so we don't have to # train next time model.fit(training_data.filter(like='feature_', axis='columns'), training_data[TARGET_COL]) print(f"saving new model: {model_name}") save_model(model, model_name) st.success('Done model training!') gc.collect() print('done get_model_and_fit') has_model_preds = False @st.cache def get_model_preds(model_name, *params): print('get_model_preds') model = load_model(model_name) has_model_preds = False nans_per_col = live_data[live_data["data_type"] == "live"][features].isna().sum() # check for nans and fill nans if nans_per_col.any(): total_rows = len(live_data[live_data["data_type"] == "live"]) print(f"Number of nans per column this week: {nans_per_col[nans_per_col > 0]}") print(f"out of {total_rows} total rows") print(f"filling nans with 0.5") live_data.loc[:, features] = live_data.loc[:, features].fillna(0.5) else: print("No nans in the features this week!") # double check the feature that the model expects vs what is available to prevent our # pipeline from failing if Numerai adds more data and we don't have time # to retrain! model_expected_features = model.booster_.feature_name() if set(model_expected_features) != set(features): print(f"New features are available! Might want to retrain model {model_name}.") validation_data.loc[:, f"preds_{model_name}"] = model.predict( validation_data.loc[:, model_expected_features]) live_data.loc[:, f"preds_{model_name}"] = model.predict( live_data.loc[:, model_expected_features]) gc.collect() # neutralize our predictions to the riskiest features validation_data[f"preds_{model_name}_neutral_riskiest_50"] = neutralize( df=validation_data, columns=[f"preds_{model_name}"], neutralizers=riskiest_features, proportion=1.0, normalize=True, era_col=ERA_COL ) live_data[f"preds_{model_name}_neutral_riskiest_50"] = neutralize( df=live_data, columns=[f"preds_{model_name}"], neutralizers=riskiest_features, proportion=1.0, normalize=True, era_col=ERA_COL ) model_to_submit = f"preds_{model_name}_neutral_riskiest_50" # rename best model to "prediction" and rank from 0 to 1 to meet upload # requirements validation_data["prediction"] = validation_data[model_to_submit].rank(pct=True) live_data["prediction"] = live_data[model_to_submit].rank(pct=True) validation_prediction_fname = f"validation_predictions_{current_round}.csv" validation_data["prediction"].to_csv(validation_prediction_fname) live_data["prediction"].to_csv(f"live_predictions_{current_round}.csv") validation_preds = pd.read_parquet(f'{get_dataset_path()}/validation_example_preds.parquet') validation_data[EXAMPLE_PREDS_COL] = validation_preds["prediction"] # get some stats about each of our models to compare... # fast_mode=True so that we skip some of the stats that are slower to calculate print('start validation_metrics') validation_stats = validation_metrics(validation_data, [model_to_submit, f"preds_{model_name}"], example_col=EXAMPLE_PREDS_COL, fast_mode=True, target_col=TARGET_COL) st.markdown(validation_stats[["mean", "sharpe"]].to_markdown()) # st.write(f''' # Done! Next steps: # 1. Go to numer.ai/tournament (make sure you have an account) # 2. Submit validation_predictions_{current_round}.csv to the diagnostics tool # 3. Submit tournament_predictions_{current_round}.csv to the "Upload Predictions" button # ''') has_model_preds = True st.button('Start model training', on_click=get_model_and_fit, args=[model_name, params]) st.button('Start model evaluation', on_click=get_model_preds, args=[model_name, params]) if has_model_preds: st.download_button('Validation data for diagnostics tool', validation_data["prediction"], validation_prediction_fname)