You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

676 lines
26 KiB
Plaintext

{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "c2e5ae80-979a-454d-b87b-667ad732d63b",
"metadata": {},
"outputs": [],
"source": [
"import numpy as np\n",
"from numpy import linalg as LA\n",
"import pandas as pd\n",
"import math\n",
"import scipy.optimize as optimize\n",
"import spiceypy"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "19159865-290b-4ad6-ad47-3a34d79dbb85",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Collecting poliastro\n",
" Using cached poliastro-0.17.0-py3-none-any.whl.metadata (11 kB)\n",
"Requirement already satisfied: astropy<6,>=5.0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from poliastro) (5.3.4)\n",
"Collecting astroquery>=0.3.9 (from poliastro)\n",
" Using cached astroquery-0.4.7-py3-none-any.whl.metadata (7.2 kB)\n",
"Requirement already satisfied: jplephem in ./miniconda3/envs/py39/lib/python3.9/site-packages (from poliastro) (2.22)\n",
"Requirement already satisfied: matplotlib!=3.0.1,>=2.0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from poliastro) (3.8.0)\n",
"Requirement already satisfied: numba>=0.53.0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from poliastro) (0.59.1)\n",
"Requirement already satisfied: numpy in ./miniconda3/envs/py39/lib/python3.9/site-packages (from poliastro) (1.23.5)\n",
"Requirement already satisfied: pandas in ./miniconda3/envs/py39/lib/python3.9/site-packages (from poliastro) (2.2.1)\n",
"Requirement already satisfied: plotly<6,>=4.0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from poliastro) (5.22.0)\n",
"Requirement already satisfied: pyerfa in ./miniconda3/envs/py39/lib/python3.9/site-packages (from poliastro) (2.0.1.4)\n",
"Requirement already satisfied: scipy>=1.4.0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from poliastro) (1.12.0)\n",
"Requirement already satisfied: PyYAML>=3.13 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from astropy<6,>=5.0->poliastro) (6.0.1)\n",
"Requirement already satisfied: packaging>=19.0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from astropy<6,>=5.0->poliastro) (23.2)\n",
"Requirement already satisfied: requests>=2.19 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from astroquery>=0.3.9->poliastro) (2.31.0)\n",
"Requirement already satisfied: beautifulsoup4>=4.8 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from astroquery>=0.3.9->poliastro) (4.12.2)\n",
"Requirement already satisfied: html5lib>=0.999 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from astroquery>=0.3.9->poliastro) (1.1)\n",
"Collecting keyring>=15.0 (from astroquery>=0.3.9->poliastro)\n",
" Using cached keyring-25.2.0-py3-none-any.whl.metadata (20 kB)\n",
"Collecting pyvo>=1.1 (from astroquery>=0.3.9->poliastro)\n",
" Using cached pyvo-1.5.1-py3-none-any.whl.metadata (4.7 kB)\n",
"Requirement already satisfied: contourpy>=1.0.1 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from matplotlib!=3.0.1,>=2.0->poliastro) (1.2.0)\n",
"Requirement already satisfied: cycler>=0.10 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from matplotlib!=3.0.1,>=2.0->poliastro) (0.11.0)\n",
"Requirement already satisfied: fonttools>=4.22.0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from matplotlib!=3.0.1,>=2.0->poliastro) (4.25.0)\n",
"Requirement already satisfied: kiwisolver>=1.0.1 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from matplotlib!=3.0.1,>=2.0->poliastro) (1.4.4)\n",
"Requirement already satisfied: pillow>=6.2.0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from matplotlib!=3.0.1,>=2.0->poliastro) (10.2.0)\n",
"Requirement already satisfied: pyparsing>=2.3.1 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from matplotlib!=3.0.1,>=2.0->poliastro) (3.0.9)\n",
"Requirement already satisfied: python-dateutil>=2.7 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from matplotlib!=3.0.1,>=2.0->poliastro) (2.8.2)\n",
"Requirement already satisfied: importlib-resources>=3.2.0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from matplotlib!=3.0.1,>=2.0->poliastro) (6.1.1)\n",
"Requirement already satisfied: llvmlite<0.43,>=0.42.0dev0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from numba>=0.53.0->poliastro) (0.42.0)\n",
"Requirement already satisfied: tenacity>=6.2.0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from plotly<6,>=4.0->poliastro) (8.3.0)\n",
"Requirement already satisfied: pytz>=2020.1 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from pandas->poliastro) (2023.3.post1)\n",
"Requirement already satisfied: tzdata>=2022.7 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from pandas->poliastro) (2023.3)\n",
"Requirement already satisfied: soupsieve>1.2 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from beautifulsoup4>=4.8->astroquery>=0.3.9->poliastro) (2.5)\n",
"Requirement already satisfied: six>=1.9 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from html5lib>=0.999->astroquery>=0.3.9->poliastro) (1.16.0)\n",
"Requirement already satisfied: webencodings in ./miniconda3/envs/py39/lib/python3.9/site-packages (from html5lib>=0.999->astroquery>=0.3.9->poliastro) (0.5.1)\n",
"Requirement already satisfied: zipp>=3.1.0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from importlib-resources>=3.2.0->matplotlib!=3.0.1,>=2.0->poliastro) (3.17.0)\n",
"Requirement already satisfied: jaraco.classes in ./miniconda3/envs/py39/lib/python3.9/site-packages (from keyring>=15.0->astroquery>=0.3.9->poliastro) (3.4.0)\n",
"Requirement already satisfied: jaraco.functools in ./miniconda3/envs/py39/lib/python3.9/site-packages (from keyring>=15.0->astroquery>=0.3.9->poliastro) (4.0.1)\n",
"Requirement already satisfied: jaraco.context in ./miniconda3/envs/py39/lib/python3.9/site-packages (from keyring>=15.0->astroquery>=0.3.9->poliastro) (5.3.0)\n",
"Requirement already satisfied: importlib-metadata>=4.11.4 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from keyring>=15.0->astroquery>=0.3.9->poliastro) (7.0.1)\n",
"Requirement already satisfied: SecretStorage>=3.2 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from keyring>=15.0->astroquery>=0.3.9->poliastro) (3.3.3)\n",
"Requirement already satisfied: jeepney>=0.4.2 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from keyring>=15.0->astroquery>=0.3.9->poliastro) (0.8.0)\n",
"Requirement already satisfied: charset-normalizer<4,>=2 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from requests>=2.19->astroquery>=0.3.9->poliastro) (2.0.4)\n",
"Requirement already satisfied: idna<4,>=2.5 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from requests>=2.19->astroquery>=0.3.9->poliastro) (3.4)\n",
"Requirement already satisfied: urllib3<3,>=1.21.1 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from requests>=2.19->astroquery>=0.3.9->poliastro) (1.26.18)\n",
"Requirement already satisfied: certifi>=2017.4.17 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from requests>=2.19->astroquery>=0.3.9->poliastro) (2024.2.2)\n",
"Requirement already satisfied: cryptography>=2.0 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from SecretStorage>=3.2->keyring>=15.0->astroquery>=0.3.9->poliastro) (42.0.5)\n",
"Requirement already satisfied: more-itertools in ./miniconda3/envs/py39/lib/python3.9/site-packages (from jaraco.classes->keyring>=15.0->astroquery>=0.3.9->poliastro) (10.2.0)\n",
"Requirement already satisfied: backports.tarfile in ./miniconda3/envs/py39/lib/python3.9/site-packages (from jaraco.context->keyring>=15.0->astroquery>=0.3.9->poliastro) (1.1.1)\n",
"Requirement already satisfied: cffi>=1.12 in ./miniconda3/envs/py39/lib/python3.9/site-packages (from cryptography>=2.0->SecretStorage>=3.2->keyring>=15.0->astroquery>=0.3.9->poliastro) (1.16.0)\n",
"Requirement already satisfied: pycparser in ./miniconda3/envs/py39/lib/python3.9/site-packages (from cffi>=1.12->cryptography>=2.0->SecretStorage>=3.2->keyring>=15.0->astroquery>=0.3.9->poliastro) (2.21)\n",
"Using cached poliastro-0.17.0-py3-none-any.whl (162 kB)\n",
"Using cached astroquery-0.4.7-py3-none-any.whl (5.3 MB)\n",
"Using cached keyring-25.2.0-py3-none-any.whl (38 kB)\n",
"Using cached pyvo-1.5.1-py3-none-any.whl (910 kB)\n",
"Installing collected packages: pyvo, keyring, astroquery, poliastro\n",
"Successfully installed astroquery-0.4.7 keyring-25.2.0 poliastro-0.17.0 pyvo-1.5.1\n",
"Note: you may need to restart the kernel to use updated packages.\n"
]
}
],
"source": [
"pip install poliastro"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "eb4d9629-fbbe-42cc-bbfe-5fb11aff2302",
"metadata": {},
"outputs": [],
"source": [
"from astropy import units as u\n",
"from astropy import constants as const\n",
"from astropy.time import Time\n",
"from astropy.coordinates import get_body_barycentric_posvel\n",
"from poliastro.bodies import Sun, Earth\n",
"from poliastro.twobody import Orbit"
]
},
{
"cell_type": "code",
"execution_count": 19,
"id": "c5a3cf10-9e10-4d54-8c08-f8b5cac2b0de",
"metadata": {},
"outputs": [],
"source": [
"folder='/home/unipi/v.vichi3/Desktop/'\n",
"df=pd.read_csv(folder+'neos_dataframe.csv')"
]
},
{
"cell_type": "code",
"execution_count": 20,
"id": "51bac7c4-0fea-4dd9-b152-6ab13b5ed0b5",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>Unnamed: 0</th>\n",
" <th>spkid</th>\n",
" <th>pha</th>\n",
" <th>H</th>\n",
" <th>epoch_mjd</th>\n",
" <th>e</th>\n",
" <th>a</th>\n",
" <th>i</th>\n",
" <th>om</th>\n",
" <th>w</th>\n",
" <th>ma</th>\n",
" <th>moid</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>0</td>\n",
" <td>20000433</td>\n",
" <td>N</td>\n",
" <td>10.41</td>\n",
" <td>60400</td>\n",
" <td>0.2227</td>\n",
" <td>1.458</td>\n",
" <td>10.83</td>\n",
" <td>304.28</td>\n",
" <td>178.90</td>\n",
" <td>334.73</td>\n",
" <td>0.1500</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>1</td>\n",
" <td>20000719</td>\n",
" <td>N</td>\n",
" <td>15.59</td>\n",
" <td>60400</td>\n",
" <td>0.5469</td>\n",
" <td>2.636</td>\n",
" <td>11.58</td>\n",
" <td>183.85</td>\n",
" <td>156.22</td>\n",
" <td>102.37</td>\n",
" <td>0.2010</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>2</td>\n",
" <td>20000887</td>\n",
" <td>N</td>\n",
" <td>13.88</td>\n",
" <td>60400</td>\n",
" <td>0.5710</td>\n",
" <td>2.472</td>\n",
" <td>9.40</td>\n",
" <td>110.42</td>\n",
" <td>350.48</td>\n",
" <td>289.48</td>\n",
" <td>0.0803</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>3</td>\n",
" <td>20001036</td>\n",
" <td>N</td>\n",
" <td>9.26</td>\n",
" <td>60400</td>\n",
" <td>0.5328</td>\n",
" <td>2.665</td>\n",
" <td>26.69</td>\n",
" <td>215.50</td>\n",
" <td>132.48</td>\n",
" <td>321.69</td>\n",
" <td>0.3450</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>4</td>\n",
" <td>20001221</td>\n",
" <td>N</td>\n",
" <td>17.38</td>\n",
" <td>60400</td>\n",
" <td>0.4352</td>\n",
" <td>1.920</td>\n",
" <td>11.88</td>\n",
" <td>171.31</td>\n",
" <td>26.68</td>\n",
" <td>197.64</td>\n",
" <td>0.1080</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" Unnamed: 0 spkid pha H epoch_mjd e a i om \\\n",
"0 0 20000433 N 10.41 60400 0.2227 1.458 10.83 304.28 \n",
"1 1 20000719 N 15.59 60400 0.5469 2.636 11.58 183.85 \n",
"2 2 20000887 N 13.88 60400 0.5710 2.472 9.40 110.42 \n",
"3 3 20001036 N 9.26 60400 0.5328 2.665 26.69 215.50 \n",
"4 4 20001221 N 17.38 60400 0.4352 1.920 11.88 171.31 \n",
"\n",
" w ma moid \n",
"0 178.90 334.73 0.1500 \n",
"1 156.22 102.37 0.2010 \n",
"2 350.48 289.48 0.0803 \n",
"3 132.48 321.69 0.3450 \n",
"4 26.68 197.64 0.1080 "
]
},
"execution_count": 20,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df.head()"
]
},
{
"cell_type": "code",
"execution_count": 21,
"id": "c68024ad-7917-401b-9f72-3ba3641a7bff",
"metadata": {},
"outputs": [],
"source": [
"#Mean anomaly to true anomaly\n",
"def kepler_equation(M,E,e):\n",
" #M=mean anomaly, E=#eccentric anomaly, e=eccentricity\n",
" return E-e*np.sin(E)-M\n",
"def mean_to_true_an(M,e):\n",
" E=optimize.newton(kepler_equation, M, args=(M,e))\n",
" nu=math.atan2(math.sin(E)*math.sqrt(1-e**2),math.cos(E)-e) #true anomaly in radians\n",
" return math.degrees(nu)"
]
},
{
"cell_type": "code",
"execution_count": 22,
"id": "f8f1dbef-9764-4732-9cfd-fb5cbebb6e5a",
"metadata": {},
"outputs": [],
"source": [
"#Perform the conversion for each element of the DataFrame\n",
"N=df.shape[0]\n",
"nu=np.zeros((N,))\n",
"for el in range(N):\n",
" nu[el]=mean_to_true_an(df['ma'][el],df['e'][el])\n",
"df['nu']=nu"
]
},
{
"cell_type": "code",
"execution_count": 23,
"id": "8188f5ba-609c-45f9-89f1-549d9b71fb0c",
"metadata": {},
"outputs": [],
"source": [
"#Keplerian to Cartesian elements\n",
"def kep_to_car(a,e,i,om,w,nu):\n",
" orb=Orbit.from_classical(Sun, a<<u.AU, e<<u.one, i<<u.deg, om<<u.deg, w<<u.deg, nu<<u.deg)\n",
" r=orb.r.to(u.AU)\n",
" v=orb.v.to(u.AU/u.d)\n",
" return r,v"
]
},
{
"cell_type": "code",
"execution_count": 24,
"id": "12f68b4e-3682-4050-866d-836cb32ae0ea",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"0 completed\n",
"10000 completed\n",
"20000 completed\n",
"30000 completed\n"
]
}
],
"source": [
"#Perform the conversion for each element of the DataFrame\n",
"car=np.zeros((N,6))\n",
"for el in range(N):\n",
" car[el,:3],car[el,3:]=kep_to_car(df['a'][el],df['e'][el],df['i'][el],df['om'][el],df['w'][el],df['nu'][el])\n",
" if (el%10000==0):\n",
" print(f\"{el} completed\")\n",
"car_df=pd.DataFrame(car, columns=['x','y','z','vx','vy','vz'])\n",
"car_df['EPOCH']=df['epoch_mjd']"
]
},
{
"cell_type": "code",
"execution_count": 26,
"id": "5a5f5a81-4b07-4a8a-9d47-39508a3b8525",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>x</th>\n",
" <th>y</th>\n",
" <th>z</th>\n",
" <th>vx</th>\n",
" <th>vy</th>\n",
" <th>vz</th>\n",
" <th>EPOCH</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>-1.043768</td>\n",
" <td>-0.948369</td>\n",
" <td>-0.267177</td>\n",
" <td>0.007095</td>\n",
" <td>-0.012608</td>\n",
" <td>-0.000237</td>\n",
" <td>60400</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>0.009877</td>\n",
" <td>2.220705</td>\n",
" <td>-0.453875</td>\n",
" <td>-0.010286</td>\n",
" <td>0.006398</td>\n",
" <td>-0.001450</td>\n",
" <td>60400</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>-0.590315</td>\n",
" <td>0.914974</td>\n",
" <td>0.038736</td>\n",
" <td>-0.018393</td>\n",
" <td>-0.008600</td>\n",
" <td>0.003350</td>\n",
" <td>60400</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>0.888349</td>\n",
" <td>1.323366</td>\n",
" <td>-0.282285</td>\n",
" <td>-0.008231</td>\n",
" <td>0.011640</td>\n",
" <td>-0.007167</td>\n",
" <td>60400</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>2.682621</td>\n",
" <td>0.150702</td>\n",
" <td>-0.116604</td>\n",
" <td>0.000993</td>\n",
" <td>0.007884</td>\n",
" <td>-0.001671</td>\n",
" <td>60400</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" x y z vx vy vz EPOCH\n",
"0 -1.043768 -0.948369 -0.267177 0.007095 -0.012608 -0.000237 60400\n",
"1 0.009877 2.220705 -0.453875 -0.010286 0.006398 -0.001450 60400\n",
"2 -0.590315 0.914974 0.038736 -0.018393 -0.008600 0.003350 60400\n",
"3 0.888349 1.323366 -0.282285 -0.008231 0.011640 -0.007167 60400\n",
"4 2.682621 0.150702 -0.116604 0.000993 0.007884 -0.001671 60400"
]
},
"execution_count": 26,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"car_df.head()"
]
},
{
"cell_type": "code",
"execution_count": 27,
"id": "cc6906b5-f2f6-4855-814e-c49c01beb2d8",
"metadata": {},
"outputs": [],
"source": [
"#Cartesian to Attributable elements\n",
"def car_to_att(r_sun, v_sun, epoch):\n",
" #r_sun, v_sun are the Cartesian heliocentric elements of the asteroid\n",
" #epoch is in MJD\n",
" earth_posvel=get_body_barycentric_posvel('earth',Time(epoch,format='mjd'))\n",
" r_earth=earth_posvel[0].xyz\n",
" v_earth=earth_posvel[1].xyz\n",
"\n",
" #Cartesian geocentric elements\n",
" r=r_sun-r_earth\n",
" v=v_sun-v_earth\n",
"\n",
" #Rotation to the equatorial reference frame\n",
" roteceq=spiceypy.spiceypy.pxform(\"ECLIPJ2000\",\"J2000\",0)\n",
" r=roteceq.dot(r)\n",
" v=roteceq.dot(v)\n",
"\n",
" rho=LA.norm(r)\n",
" dz=r[0]**2+r[1]**2\n",
" epsilon=np.finfo(float).eps\n",
" if (dz.value<100*epsilon):\n",
" ra=0\n",
" else: ra=np.arctan2(r[1],r[0]) #in radians\n",
" dec=np.arcsin(r[2]/rho)\n",
"\n",
" #Computation of first derivatives\n",
" dadx=np.zeros((3,1))<<u.one/u.AU\n",
" dddx=np.zeros((3,1))<<u.one/u.AU\n",
" dadx[0]=-r[1]/dz\n",
" dadx[1]=r[0]/dz\n",
" dadx=dadx.reshape(1,-1) #transform dadx into a row vector\n",
" dddx[0]=-r[2]*(r[0]/(np.sqrt(dz)*rho**2))\n",
" dddx[1]=-r[2]*(r[1]/(np.sqrt(dz)*rho**2))\n",
" dddx[2]=np.sqrt(dz)/rho**2\n",
" dddx=dddx.reshape(1,-1)\n",
"\n",
" rho_dot=np.dot(r,v)/rho\n",
" ra_dot=np.dot(dadx,v) #in 1/day\n",
" dec_dot=np.dot(dddx,v) #in 1/day\n",
" cos_ra=np.cos(ra)\n",
" sin_ra=np.sin(ra)\n",
" sin_dec=np.sin(dec)\n",
" return [cos_ra.value,sin_ra.value,sin_dec.value, ra_dot[0].value, dec_dot[0].value]"
]
},
{
"cell_type": "code",
"execution_count": 28,
"id": "5b28583a-5474-418d-abdd-52713c7c03b2",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"0 completed\n",
"10000 completed\n",
"20000 completed\n",
"30000 completed\n"
]
}
],
"source": [
"#Perform the conversion for each element of the DataFrame\n",
"attr=np.zeros((N,5))\n",
"for el in range(N):\n",
" attr[el,:]=car_to_att(car_df.iloc[el,0:3]<<u.AU, car_df.iloc[el,3:6]<<u.AU/u.d, car_df['EPOCH'][el])\n",
" if (el%10000==0):\n",
" print(f\"{el} completed\")"
]
},
{
"cell_type": "code",
"execution_count": 29,
"id": "fabd950c-cd23-4fb8-8250-ec16dd2acc53",
"metadata": {},
"outputs": [],
"source": [
"attr_df=pd.DataFrame(attr,columns=['cos_ra','sin_ra','sin_dec','ra_dot','dec_dot'])"
]
},
{
"cell_type": "code",
"execution_count": 30,
"id": "b4e531f2-46da-4470-854a-d49fad2675c7",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>cos_ra</th>\n",
" <th>sin_ra</th>\n",
" <th>sin_dec</th>\n",
" <th>ra_dot</th>\n",
" <th>dec_dot</th>\n",
" <th>MOID</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>-0.084129</td>\n",
" <td>-0.996455</td>\n",
" <td>-0.606347</td>\n",
" <td>0.006616</td>\n",
" <td>0.006707</td>\n",
" <td>0.1500</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>0.392310</td>\n",
" <td>0.919833</td>\n",
" <td>0.229663</td>\n",
" <td>0.007516</td>\n",
" <td>0.004055</td>\n",
" <td>0.2010</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>0.387645</td>\n",
" <td>0.921809</td>\n",
" <td>0.459919</td>\n",
" <td>0.019881</td>\n",
" <td>0.011605</td>\n",
" <td>0.0803</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>0.791026</td>\n",
" <td>0.611783</td>\n",
" <td>0.166871</td>\n",
" <td>0.011227</td>\n",
" <td>0.003821</td>\n",
" <td>0.3450</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>0.996437</td>\n",
" <td>0.084345</td>\n",
" <td>0.023712</td>\n",
" <td>0.005317</td>\n",
" <td>0.003798</td>\n",
" <td>0.1080</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" cos_ra sin_ra sin_dec ra_dot dec_dot MOID\n",
"0 -0.084129 -0.996455 -0.606347 0.006616 0.006707 0.1500\n",
"1 0.392310 0.919833 0.229663 0.007516 0.004055 0.2010\n",
"2 0.387645 0.921809 0.459919 0.019881 0.011605 0.0803\n",
"3 0.791026 0.611783 0.166871 0.011227 0.003821 0.3450\n",
"4 0.996437 0.084345 0.023712 0.005317 0.003798 0.1080"
]
},
"execution_count": 30,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"attr_df['MOID']=df['moid']\n",
"attr_df.head()"
]
},
{
"cell_type": "code",
"execution_count": 31,
"id": "f65e0163-c9d8-492a-9095-0d7d34ffad5b",
"metadata": {},
"outputs": [],
"source": [
"attr_df.to_csv(folder+'neos_attr.csv',index=False)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.19"
}
},
"nbformat": 4,
"nbformat_minor": 5
}