How to build a player prop model in Python

import os, requests, pandas as pd

API = "https://fanduelapi.com/api/v1"
H = {"Authorization": f"Bearer {os.environ['FANDUEL_API_KEY']}"}

def rows(path, **params):
    r = requests.get(f"{API}/{path}", params=params, headers=H, timeout=30)
    r.raise_for_status()
    return r.json()["data"]

# One row per prop: open line and price, closing line and price, and the moves before the start.
train = pd.DataFrame(rows("training-set", league="NHL", **{"from": "2026-11-01T00:00:00Z"}))
# Graded outcomes come from /results: closing line against the box score. Join them on prop_id.
results = pd.DataFrame(rows("results", league="NHL", limit=200))[["prop_id", "actual", "result"]]
train = train.merge(results, on="prop_id")
train = train[train["result"].isin(["over", "under"])]     # pushes and voids are not outcomes
print(len(train), "graded props")

import numpy as np

df = train.copy()
df["y"] = (df["result"] == "over").astype(int)
df["line_move"] = df["close_line"] - df["open_line"]           # how far the line travelled
df["fair_move"] = df["close_fair_over"] - df["open_fair_over"]  # how far the price travelled, in probability
df["moves"] = df["line_moves"] + df["price_moves"]
df["market_p"] = df["close_fair_over"]                          # the market's own no-vig view at the close

X = df[["market_p", "line_move", "fair_move", "moves"]].fillna(0)
y = df["y"]

from sklearn.linear_model import LogisticRegression
from sklearn.metrics import brier_score_loss, log_loss

# Split by time, never at random: the future must not leak into training.
df = df.sort_values("starts_at")
cut = int(len(df) * 0.75)
train_idx, test_idx = df.index[:cut], df.index[cut:]

model = LogisticRegression().fit(X.loc[train_idx], y.loc[train_idx])
p = model.predict_proba(X.loc[test_idx])[:, 1]

market = df.loc[test_idx, "market_p"].clip(0.01, 0.99)
print("model  brier", brier_score_loss(y.loc[test_idx], p), "log loss", log_loss(y.loc[test_idx], p))
print("market brier", brier_score_loss(y.loc[test_idx], market), "log loss", log_loss(y.loc[test_idx], market))

# Closing line value: did the price you would have taken beat where the market closed?
def decimal(american):
    return 1 + (american / 100 if american > 0 else 100 / -american)

test = df.loc[test_idx].assign(p=p)
bets = test[test["p"] > test["market_p"] + 0.03]                # only where the model disagrees by 3 points
clv = (1 / bets["open_over_american"].map(decimal)) - (1 / bets["close_over_american"].map(decimal))
print(len(bets), "bets, mean CLV", round(clv.mean(), 4))