kmedved’s gists

kmedved / contemplative-llms.txt

Created January 6, 2025 21:03 — forked from Maharshi-Pandya/contemplative-llms.txt

"Contemplative reasoning" response style for LLMs like Claude and GPT-4o

	You are an assistant that engages in extremely thorough, self-questioning reasoning. Your approach mirrors human stream-of-consciousness thinking, characterized by continuous exploration, self-doubt, and iterative analysis.

	## Core Principles

	1. EXPLORATION OVER CONCLUSION
	- Never rush to conclusions
	- Keep exploring until a solution emerges naturally from the evidence
	- If uncertain, continue reasoning indefinitely
	- Question every assumption and inference

kmedved / NumbaDecisionTreeRegressor.py

Last active December 13, 2024 13:26

Faster DecisionTreeRegressor for use with Ngboost

	import numpy as np
	from numba import njit, prange
	from sklearn.base import BaseEstimator, RegressorMixin

	@njit(inline='always')
	def weighted_variance_from_sums(sum_w, sum_wy, sum_wy_sq):
	# Weighted variance: var_w = (sum_wy_sq / sum_w) - (sum_wy / sum_w)**2
	if sum_w <= 1e-14:
	return 0.0
	mean_w = sum_wy / sum_w

kmedved / fit_catboost

Created January 17, 2023 14:04

	def fit_catboost_model(df,
	features,
	label_col,
	sample_weight_col = None,
	raw_model = CatBoostRegressor(),
	cv_splitter = GroupShuffleSplit(n_splits=2, test_size = 0.3),
	group_id_col = None,
	verbose = False,
	split_type = None):
	"""

kmedved / fit_mrmr.py

Created January 5, 2023 15:53

kmedved / shorten_features_boruta.py

Created January 5, 2023 15:45

	def shorten_features_boruta(df,
	features,
	label_col,
	sample_weight_col = None,
	n_trials = 100,
	gpu = False,
	n_jobs = 8,
	classification_fl = False,
	sample_fl = True,
	base_model = None):

kmedved / shorten_features_catboost.py

Created January 5, 2023 15:41

	def shorten_features_catboost(df,
	features,
	label_col,
	n_estimators = 200,
	sample_weight_col = None,
	model = None,
	group_id = None,
	steps = 6,
	gpu = False,
	n_jobs = 8,

kmedved / weighted_chaterjee.py

Created April 29, 2022 16:07 — forked from bbstats/weighted_chaterjee.py

weighted_chaterjee.py

	from scipy.stats import rankdata
	import numpy as np

	def weighted_chaterjee_correlation(x, y, sample_weight=None):

	"""x and y must be 1d, probably"""

	n = x.size
	rk_x = rankdata(x, method="average")
	rk_y = rankdata(y, method="average")

kmedved / hebo_patch.py

Created January 4, 2022 21:30 — forked from bbstats/hebo_patch.py

Simple patch of HEBO to support sample weights (via fit_params argument)

kmedved / skboost.py

Created July 22, 2021 23:13

Skboost

	from threadpoolctl import threadpool_limits
	from sklearn.experimental import enable_hist_gradient_boosting
	from sklearn.ensemble import HistGradientBoostingRegressor

	class SkBoost(HistGradientBoostingRegressor):
	def __init__(
	self,
	loss="least_squares",
	learning_rate=0.1,
	max_iter=100,

kmedved / scrape_pbpstats

Created July 20, 2021 12:37

PBPstats API Endpoints

	def get_pbpstats(nba_game_id, stats_type = 'box'):

	# Convert the game ID

	nba_game_id = str(nba_game_id)

	if len(nba_game_id) < 10:
	nba_game_id = '00' + nba_game_id

	# Figure out the right URL