Credit score
An agent builds the profile and you watch the MLflow champion's pipeline run step by step, or validate and score a CSV in batch.
A real credit model, step by step
It's the mlops-platform pipeline: the MLflow @champion, trained on German Credit and served by the internal scoring service. In live mode, you build a profile and watch each step run on the server. In batch mode, you validate and score a whole CSV.
…
Build the profile
Chat with the agent or fill in the form. Whatever you don't know, the pipeline imputes.
R$ 9.70 per DM Why reais, if the data is German?
credit-g comes from a 1970s German bank, with amounts in Deutsche Marks (DM). Before training, mlops-platform converts the dataset to Brazilian reais at R$ 9.70 per DM: the amount is multiplied by the rate and the balance and savings brackets become R$ brackets (DM 200 becomes R$ 1,940). The model was retrained that way and only knows reais. The rate is by purchasing power: €1 = DM 1.95583 (fixed parity), German prices rose about 3.25× since 1975 and the euro is near R$ 5.85. Changing the scale doesn't change what the model learns: the AUC is the same as the Deutsche Mark version.
The pipeline, step by step
The steps take about 100 ms on the server; the page shows one at a time, with the time measured there.
- 1Raw data
- 2Data transform
- 3Normalization
- 4Live fit
- 5Inference
- 6What weighed
Result
Probability of default and the decision by the champion's cutoff.
What weighed the most
Generate a test file or use yours
Two sample CSVs, generated in your browser: a clean one and one with planted errors.
Columns and accepted values
One column per variable (the header is required; "id" is optional and comes back in the result). Comma or semicolon. Up to 250 rows and 50 KB.
| purpose | new car · used car · furniture/equipment · radio/tv · domestic appliance · repairs · education · retraining · business · other |
| credit_amount | number >0–10000000 |
| duration | integer 1–120 |
| installment_commitment | integer 1–4 |
| checking_status | <0 · 0<=X<1940 · >=1940 · no checking |
| savings_status | <970 · 970<=X<4850 · 4850<=X<9700 · >=9700 · no known savings |
| credit_history | no credits/all paid · all paid · existing paid · delayed previously · critical/other existing credit |
| existing_credits | integer 0–10 |
| other_payment_plans | none · bank · stores |
| other_parties | none · guarantor · co applicant |
| employment | unemployed · <1 · 1<=X<4 · 4<=X<7 · >=7 |
| job | unemp/unskilled non res · unskilled resident · skilled · high qualif/self emp/mgmt |
| age | integer 18–100 |
| housing | own · rent · for free |
| residence_since | integer 1–4 |
| property_magnitude | real estate · life insurance · car · no known property |
| num_dependents | integer 0–10 |
| own_telephone | yes · none |
Upload and validate
The file is treated as untrusted data: read with Python's csv module, no eval, with size and row limits. Validating doesn't use the quota.
Score it
Only the valid rows go to the model. The whole batch counts as 1 use.
What happens in each step
- 1
Raw data
The agent (or the form) sends a JSON with 18 fields, uncorrected in the browser: a number may arrive as text and a field may be empty. Sex and foreign-worker status aren't even asked, because they're left out of the model.
- 2
Data transform
The service casts the types, fills missing values with the credit-g median or mode and runs the 11 categoricals through the champion's own OneHotEncoder. A category rare in training falls into the "infrequent" bucket.
- 3
Normalization
The 7 numeric variables go through the champion's StandardScaler: z = (x − mean) / sd, with the training mean and sd. That puts months and reais on the same scale.
- 4
Live fit
To show the training is real, a logistic regression is trained on the spot on 750 rows and measured on 250 it didn't see (AUC). It decides nothing: it's the baseline.
- 5
Inference
The champion, the model that passed the MLflow gates, computes the log-odds; the sigmoid turns it into a probability of default and the cutoff decides between approving and sending to review.
- 6
What weighed
Each variable is swapped for its typical credit-g value and the model runs again. The difference, in percentage points, shows what pushed the risk up or down.
German Credit (OpenML credit-g, UCI) comes from a 1970s German bank, with amounts in Deutsche Marks, converted to reais (R$ 9.70 per DM) before training. Sex and foreign-worker status are left out of the model, and the performance gap between age groups is one of the promotion gates. The answer comes from the @champion in MLflow.