Where can I get training data for a player prop model?
Short answer:Use a dataset with one row per graded line that carries the opening and closing line, the price at close, game context and the result; Underdog API's /api/v1/training-set returns exactly that for Underdog Fantasy lines as JSON or CSV, with voids left out unless you ask for them.
Underdog API on 15 Nov 2026
| Graded results | 12,669 graded lines in NHL, NBA, college football, NFL and college basketball, games 6 to 14 Nov 2026 (paid plans) |
|---|---|
| Line history | Every recorded line and multiplier move since 6 Nov 2026 (paid plans) |
| Cheapest paid plan | Developer, $49 a month for 10,000 credits |
Updated November 15, 2026. Our numbers are from the production database at 06:07 UTC on 15 Nov 2026; other providers' facts are paraphrased from their public pages on the date shown.
What is in a row
Opening and closing line, both side multipliers at close, payout tier, opponent, home or away, rest days, minutes, the actual value and the result. Every row is a line that was recorded, never a reconstructed one.
The graded base is 12,669 lines (games 6 to 14 Nov 2026) and grows with every game day.
Compared with other providers
| Underdog API (this site) | PropLine | The Odds API | ParlayAPI | |
|---|---|---|---|---|
| Graded results | 12,669 graded lines in NHL, NBA, college football, NFL and college basketball, games 6 to 14 Nov 2026 (paid plans) | Graded props from Hobby; 90-day resolved archive on Pro ($19) | Not stated on its pricing page | Not stated on its pricing page |
| Line history | Every recorded line and multiplier move since 6 Nov 2026 (paid plans) | Historical line movement and opening and closing lines from Hobby ($9) | Historical odds on every paid plan | Lookback by plan: 168 hours on Starter up to 87,600 hours on Scale, where records exist |
| Cheapest paid plan | Developer, $49 a month for 10,000 credits | Hobby, $9 a month, 5,000 requests a day; Pro $19, Streaming Lite $39, Streaming $79 | $30 a month for 20,000 credits; $59, $119 and $249 plans above it | Starter, $5 a month for 20,000 credits; Pro $20, Business $40, Enterprise $100, Scale $200 |
Read on 15 Nov 2026 unless a cell gives another date. Plans change; confirm on each provider's page. This site is independent and not affiliated with any company named here.
Where this API is worse
- Recording began on 6 Nov 2026, so history is days deep, not seasons. Providers that started earlier have longer archives.
- Graded results cover five leagues so far (NHL 5,668, NBA 2,153, college football 1,841, NFL 1,564, college basketball 1,443); tennis, esports and CFL lines are not graded yet.
Try it against the real API
Set UNDERDOG_API_KEY to a free key from sign-up. Every path below is documented in the API reference; costs are on credits.
import os, requests
r = requests.get(
"https://underdogapi.com/api/v1/training-set?league=NBA&stat=points&limit=100",
headers={"Authorization": f"Bearer {os.environ['UNDERDOG_API_KEY']}"},
timeout=30,
)
r.raise_for_status()
body = r.json()
for row in body["data"]:
print(row)
print(body["meta"]) # observed_at, next_cursor, plan limitscurl -s 'https://underdogapi.com/api/v1/training-set?league=NHL&format=csv' \
-H "Authorization: Bearer $UNDERDOG_API_KEY"Questions
Is the training set free?
No; Free gets a 402 with a few labelled rows to inspect the schema. Developer pages the full set.
Is there a Python starter?
Yes: /build-an-nba-prop-model walks through a first model.
Underdog API is an independent data service. It is not affiliated with, endorsed by or operated by Underdog Fantasy or any company named on this page.