data-estate / examples /quickstart.py
EMTIAZZ's picture
Refreshed synthetic data (new seed, company, id scheme, counts)
d385479 verified
Raw
History Blame Contribute Delete
3.06 kB
#!/usr/bin/env python3
"""
Data Estate — quickstart.
Usage:
python quickstart.py [ROOT]
ROOT defaults to the parent folder of this script (so it works in-place inside the dataset).
It loads the structured tables, joins across the CRM/ERP/claims "systems", then shows how to
reach a scanned form, an email thread, and a call transcript for the SAME claim.
"""
import os, sys, json
import pandas as pd
ROOT = sys.argv[1] if len(sys.argv) > 1 else os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
def path(*a): return os.path.join(ROOT, *a)
print(f"== Data Estate :: root = {ROOT} ==\n")
# 1) structured tables -------------------------------------------------------
cust = pd.read_csv(path("structured", "crm_customers.csv"))
pol = pd.read_csv(path("structured", "erp_policies.csv"))
claims = pd.read_csv(path("structured", "claims.csv"))
print(f"customers={len(cust):>6} policies={len(pol):>6} claims={len(claims):>6}")
# 2) join across the three "systems" (note the column-name mismatch) ---------
full = (claims
.merge(cust, on="customer_id", how="left")
.merge(pol, left_on="policy_no", right_on="PolicyNo", how="left"))
print(f"joined claims+customer+policy -> {full.shape[0]} rows x {full.shape[1]} cols")
# 3) how many customers look like near-duplicates ----------------------------
dup_keys = cust["customer_id"].astype(str).str.startswith("CU99").sum()
print(f"near-duplicate customer rows (CU99*): {dup_keys}")
# 4) find a customer that appears across ALL modalities, then trace them -----
docs = json.load(open(path("documents_index.json")))
emails = json.load(open(path("emails_index.json")))
calls = json.load(open(path("calls_index.json")))
doc_custs = {d["customer_id"] for d in docs}
mail_custs = {e["customer_id"] for e in emails}
call_custs = {c["customer_id"] for c in calls}
fully_linked = doc_custs & mail_custs & call_custs
cust_id = sorted(fully_linked)[0] if fully_linked else docs[0]["customer_id"]
sample = next(d for d in docs if d["customer_id"] == cust_id)
e = next(x for x in emails if x["customer_id"] == cust_id)
c = next(x for x in calls if x["customer_id"] == cust_id)
claim_no = sample["claim_no"]
print(f"\n-- tracing customer {cust_id} across ALL sources --")
print(" claims :", list(claims[claims.customer_id == cust_id].claim_no))
print(" scanned form :", sample["scan"], "(open with PIL / feed to OCR)")
print(" pdf :", sample["pdf"])
print(" email thread :", e["file"])
print(" transcript :", c["file"])
row = claims[claims["claim_no"] == claim_no].iloc[0]
print(" claim detail :", dict(row[["policy_no", "customer_id", "line", "status", "amount_claimed"]]))
# 5) verify the scan opens ---------------------------------------------------
try:
from PIL import Image
im = Image.open(path(sample["scan"]))
print(f"\n scan opens OK -> {im.size} {im.mode}")
except Exception as ex:
print(" (PIL not available or scan missing:", ex, ")")
print("\nDone. See README.md for the full data dictionary and per-question hints.")