Model Signiture with MLflow
Model signiture is an object that allows us to specify the data type and shape that the model can work on.
You can build a model signiture using mlflow.types.schema
import mlflow
from mlflow_utils import create_mlflow_experiment
from mlflow.models.signature import ModelSignature
from mlflow.models.signature import infer_signature
from mlflow.types.schema import Schema
from mlflow.types.schema import ParamSchema
from mlflow.types.schema import ParamSpec
from mlflow.types.schema import ColSpec
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import make_classification
import pandas as pd
from typing import Tuple
def get_train_data() -> Tuple[pd.DataFrame]:
"""
Generate train and test data.
:return: x_train,y_train
"""
X, y = make_classification()
features = [f"feature_{i+1}" for i in range(X.shape[1])]
df = pd.DataFrame(X, columns=features)
df["label"] = y
return df[features], df["label"]X_train, y_train = get_train_data()
print(X_train.head())
# print(y_train.head())
print(X_train.info()) feature_1 feature_2 feature_3 feature_4 feature_5 feature_6 \
0 -1.201183 1.016247 0.048663 0.244420 -2.880241 -3.269429
1 -1.877101 -0.806603 -0.264468 0.404385 0.558561 -0.390860
2 1.334741 -0.080834 -0.495585 2.922658 -0.260945 1.771953
3 -0.450731 0.545779 -1.808946 -0.968887 -1.074346 2.231930
4 1.161051 0.131469 -0.743462 0.563626 -1.467326 -0.094547
feature_7 feature_8 feature_9 feature_10 feature_11 feature_12 \
0 -2.429654 0.530264 2.553249 -0.407145 -0.414696 1.151798
1 0.176100 0.309046 0.375332 0.744301 1.832984 -0.784057
2 -0.580651 0.116984 0.570811 -2.738946 0.921859 -0.612626
3 -0.978274 0.924715 1.299982 -0.581170 1.027152 -0.143539
4 -0.877583 -0.479286 0.830151 -0.337659 2.211519 1.227915
feature_13 feature_14 feature_15 feature_16 feature_17 feature_18 \
0 0.498910 -0.585375 1.922103 0.719760 -1.651078 1.339697
1 1.215857 -0.579073 -0.685584 2.194892 0.440435 1.973318
2 1.648948 0.065531 0.162212 -0.530821 -0.670817 -1.538897
3 0.506369 -0.375960 0.007354 -0.755527 0.170947 0.284366
4 -1.348091 -0.716956 0.657270 0.258002 1.125622 -1.368340
feature_19 feature_20
0 -0.117558 -0.806716
1 -1.021339 -0.669179
2 -1.286377 -1.706245
3 -0.298616 1.638584
4 1.214653 -1.409494
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 100 entries, 0 to 99
Data columns (total 20 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 feature_1 100 non-null float64
1 feature_2 100 non-null float64
2 feature_3 100 non-null float64
3 feature_4 100 non-null float64
4 feature_5 100 non-null float64
5 feature_6 100 non-null float64
6 feature_7 100 non-null float64
7 feature_8 100 non-null float64
8 feature_9 100 non-null float64
9 feature_10 100 non-null float64
10 feature_11 100 non-null float64
11 feature_12 100 non-null float64
12 feature_13 100 non-null float64
13 feature_14 100 non-null float64
14 feature_15 100 non-null float64
15 feature_16 100 non-null float64
16 feature_17 100 non-null float64
17 feature_18 100 non-null float64
18 feature_19 100 non-null float64
19 feature_20 100 non-null float64
dtypes: float64(20)
memory usage: 15.8 KB
NoneThe names of the data types should be standardized
data_map = {
'int64': 'integer',
'float64': 'double',
'bool': 'boolean',
'str': 'string',
"date": 'datetime'
}Define Input and Output Schema
cols_spec = []
for name, dtype in X_train.dtypes.to_dict().items():
cols_spec.append(ColSpec(name=name, type=data_map[str(dtype)]))
input_schema = Schema(inputs=cols_spec)
output_schema = Schema([ColSpec(name="label", type="integer")])
print(input_schema)
print(output_schema)['feature_1': double (required), 'feature_2': double (required), 'feature_3': double (required), 'feature_4': double (required), 'feature_5': double (required), 'feature_6': double (required), 'feature_7': double (required), 'feature_8': double (required), 'feature_9': double (required), 'feature_10': double (required), 'feature_11': double (required), 'feature_12': double (required), 'feature_13': double (required), 'feature_14': double (required), 'feature_15': double (required), 'feature_16': double (required), 'feature_17': double (required), 'feature_18': double (required), 'feature_19': double (required), 'feature_20': double (required)]
['label': integer (required)]Define Parametre schema
parameter = ParamSpec(name="model_name", dtype="string", default="model1")
param_schema = ParamSchema(params=[parameter])
print(parameter)
print(param_schema)'model_name': string (default: model1)
['model_name': string (default: model1)]Model Signature (input schema + output schema + parametre schema)
You can either specify the signature or infer the signature based on the data.
model_signature = ModelSignature(inputs=input_schema, outputs=output_schema, params=param_schema)
print(model_signature.to_dict())MODEL SIGNATURE
{'inputs': '[{"type": "double", "name": "feature_1", "required": true}, {"type": "double", "name": "feature_2", "required": true}, {"type": "double", "name": "feature_3", "required": true}, {"type": "double", "name": "feature_4", "required": true}, {"type": "double", "name": "feature_5", "required": true}, {"type": "double", "name": "feature_6", "required": true}, {"type": "double", "name": "feature_7", "required": true}, {"type": "double", "name": "feature_8", "required": true}, {"type": "double", "name": "feature_9", "required": true}, {"type": "double", "name": "feature_10", "required": true}, {"type": "double", "name": "feature_11", "required": true}, {"type": "double", "name": "feature_12", "required": true}, {"type": "double", "name": "feature_13", "required": true}, {"type": "double", "name": "feature_14", "required": true}, {"type": "double", "name": "feature_15", "required": true}, {"type": "double", "name": "feature_16", "required": true}, {"type": "double", "name": "feature_17", "required": true}, {"type": "double", "name": "feature_18", "required": true}, {"type": "double", "name": "feature_19", "required": true}, {"type": "double", "name": "feature_20", "required": true}]', 'outputs': '[{"type": "integer", "name": "label", "required": true}]', 'params': '[{"name": "model_name", "default": "model1", "shape": null, "type": "string"}]'}model_signature = infer_signature(X_train, y_train, params={"model_name": "model1"})
print("MODEL SIGNATURE")
print(model_signature.to_dict())MODEL SIGNATURE
{'inputs': '[{"type": "double", "name": "feature_1", "required": true}, {"type": "double", "name": "feature_2", "required": true}, {"type": "double", "name": "feature_3", "required": true}, {"type": "double", "name": "feature_4", "required": true}, {"type": "double", "name": "feature_5", "required": true}, {"type": "double", "name": "feature_6", "required": true}, {"type": "double", "name": "feature_7", "required": true}, {"type": "double", "name": "feature_8", "required": true}, {"type": "double", "name": "feature_9", "required": true}, {"type": "double", "name": "feature_10", "required": true}, {"type": "double", "name": "feature_11", "required": true}, {"type": "double", "name": "feature_12", "required": true}, {"type": "double", "name": "feature_13", "required": true}, {"type": "double", "name": "feature_14", "required": true}, {"type": "double", "name": "feature_15", "required": true}, {"type": "double", "name": "feature_16", "required": true}, {"type": "double", "name": "feature_17", "required": true}, {"type": "double", "name": "feature_18", "required": true}, {"type": "double", "name": "feature_19", "required": true}, {"type": "double", "name": "feature_20", "required": true}]', 'outputs': '[{"type": "long", "name": "label", "required": true}]', 'params': '[{"name": "model_name", "default": "model1", "shape": null, "type": "string"}]'}
c:\Users\dof07\anaconda3\envs\mlflow\lib\site-packages\mlflow\types\utils.py:452: UserWarning: Hint: Inferred schema contains integer column(s). Integer columns in Python cannot represent missing values. If your input data contains missing values at inference time, it will be encoded as floats and will cause a schema enforcement error. The best way to avoid this problem is to infer the model schema based on a realistic data sample (training dataset) that includes missing values. Alternatively, you can declare integer columns as doubles (float64) whenever these columns may have missing values. See `Handling Integers With Missing Values <https://www.mlflow.org/docs/latest/models.html#handling-integers-with-missing-values>`_ for more details.
warnings.warn(experiment_id = create_mlflow_experiment(
experiment_name="Model Signature",
artifact_location="model_signature_artifacts",
tags={"purpose": "learning"},
)Experiment Model Signature already exists.with mlflow.start_run(experiment_id=experiment_id, run_name="model_signature_run") as run:
mlflow.sklearn.log_model(
sk_model=RandomForestClassifier(),
artifact_path="model_signature",
signature=model_signature,
)
yozzum