apoorvalal

automated reclaiming RAM from macos' pesky background services

Edit ramcleaner.sh to kill the processes that tend to balloon RAM usage. medianalysisd is the main culprit; it often gobbles up 3-4GB RAM in random spikes, which is prohitive on the base mac mini 16GB that I run as a home-server.

This bash script searches for the service by prgrepping active processes and kills what it finds. When OS services are observationally equivalent to viruses, they ought to be treated accordingly.

Schedule it by running crontab -e and entering the following, which runs the cleanup script every 2 minutes and logs its actions to ~/.ramcleaner.log.

*/2 * * * * /bin/bash ramcleaner.sh >/dev/null 2>&1

apoorvalal

#!/Users/alal/miniforge3/envs/llm/bin/python
# this needs to point to a python virtualenv with ollama-python
# your system should also have ollama installed 
# if you have an nvidia gpu, install lshw first before installing ollama

import argparse
import ollama

def main(msg, mod='gemma3:27b', outfile=None):
    response = ollama.chat(model=mod, messages=[

apoorvalal

#!/bin/bash

# Directory to search
DIRECTORY=$1

# Additional extensions to search, if provided
EXTENSIONS="tex"

if [ ! -z "$2" ]; then
  EXTENSIONS="$EXTENSIONS|$2"

apoorvalal

# %%
import functools
from typing import Callable, TypeVar, Any
import pandas as pd
import numpy as np


T = TypeVar("T")

# %%

apoorvalal

library(momentfit); library(car); library(tictoc)
set.seed(42)
# %%
dgp = \(n=500, k = 2){
  X = matrix(rnorm(n * 2), n, 2)
  Y1 = X[, 1] + X[, 1]^2 + runif(n, -0.5, 0.5)
  Y0 = X[, 2] + X[, 2]^2 + runif(n, -1, 1)
  Z = rbinom(n, 1, 0.6)
  Y = Z * Y1 + (1-Z) * Y0
  data.frame(Y, Z, X)

apoorvalal

from joblib import Parallel, delayed
import numpy as np
import pandas as pd


class LinearMediation:
    def __init__(self):
        pass

    def fit(self, X, W, y, store=True):

apoorvalal

\documentclass[%
  article,
  % beamer,
  beameroptions={ignorenonframetext,14pt},
  articleoptions={a4paper,12pt},
  also={trans,handout,article}
  ]{beamerswitch}
\handoutlayout{nup=3plus,border=1pt}
\articlelayout{maketitle,frametitles=none}
\mode<article>{\usepackage[hmargin=2cm,vmargin=2cm]{geometry}}

apoorvalal

import numpy as np
import pandas as pd
from scipy.stats import norm
from sklearn.model_selection import cross_val_predict, KFold

# learners
from xgboost import XGBRegressor
from glum import GeneralizedLinearRegressorCV
from sklearn.kernel_ridge import KernelRidge

apoorvalal

library(estimatr)
data(auto)
# %% FWL regression coefficient
auto$ytil = lm(price ~ displacement, auto)$resid
auto$x2til = lm(weight ~ displacement, auto)$resid
(fwlest = lm_robust(ytil ~ x2til, auto, se_type = "HC0")
  %>% summary %>% .$coefficients %>% .[2, 1:2])
# %%
(fullest =
  lm_robust(price ~ weight + displacement, auto, se_type = "HC0") %>%

apoorvalal

import numpy as np
from scipy.linalg import lstsq
np.random.seed(42)
# %%
def ols(X, y, vcov = 'HC1', driver = 'gelsy'):
    """
    Fast, minimal implementation of least squares regression with robust SEs

    Args:
        X:          n X p array of covariates