Solution¶

<Model>Solution allows access to the solution of an sklearn.<Model>. The fitted model is considered a "solution". You may create a solution object from a model and export it to a file. At any moment, the solution (fitted model) can be loaded from a file that holds the correct information.

Consider the same scripts shown in the model section and statistics section, which fit, predict and obtain statistics with the diabetes dataset. You may use the <Model>Solution to obtain and write the solution of the model. After it has been exported, you can load the solution, create a fitted model from it, and make a basic prediction.

Dummy¶

Reference

Learn more by going to the reference for the dummy.solution module.

Write the solution.

main.py

from nextmv_sklearn import dummy

import nextmv

# Model and statistics code here.

# Create the solution object from the model and construct an output.
solution = dummy.DummyRegressorSolution.from_model(fit)
output = nextmv.Output(
    options=options,
    solution=solution.to_dict(),
    statistics=statistics,
)

# Export the output, with the solution, to a file.
file_path = "regressor.json"
nextmv.write(output, file_path)

BashOutput

cat regressor.json

{
  "options": {
    "strategy": null,
    "constant": null,
    "quantile": null
  },
  "solution": {
    "class": {
      "module": "nextmv_sklearn.dummy.solution",
      "name": "DummyRegressorSolution"
    },
    "attributes": {
      "constant_": [
        [
          152.13348416289594
        ]
      ],
      "n_features_in_": 10,
      "n_outputs_": 1
    }
  },
  "statistics": {
    "run": {
      "duration": 0.0009088516235351562
    },
    "result": {
      "custom": {
        "score": 0.0
      }
    },
    "series_data": {},
    "schema": "v1"
  },
  "assets": []
}

Load the solution and make a prediction.

main.py

import nextmv


# Load the solution from the file.
input = nextmv.load(path=file_path)
loaded_solution = nextmv.from_dict(input.data["solution"])

# Convert the solution to a model and make a prediction.
loaded_fit = loaded_solution.to_model()
print(loaded_fit.predict(X[:1]))

BashOutput

uv run main.py

[152.13348416]

Ensemble¶

Reference

Learn more by going to the reference for the ensemble.solution module.

Gradient boosting¶

Write the solution.

main.py

from nextmv_sklearn import ensemble

import nextmv

# Model and statistics code here.

# Create the solution object from the model and construct an output.
solution = ensemble.GradientBoostingRegressorSolution.from_model(fit)
output = nextmv.Output(
    options=options,
    solution=solution.to_dict(),
    statistics=statistics,
)

# Export the output, with the solution, to a file.
file_path = "regressor.json"
nextmv.write(output, file_path)

BashOutput

cat regressor.json

{
  "options": {
    "loss": null,
    "learning_rate": null,
    "n_estimators": null,
    "subsample": null,
    "criterion": null,
    "min_samples_split": null,
    "min_samples_leaf": null,
    "min_weight_fraction_leaf": null,
    "max_depth": null,
    "min_impurity_decrease": null,
    "random_state": null,
    "max_features": null,
    "alpha": null,
    "max_leaf_nodes": null,
    "warm_start": null,
    "validation_fraction": null,
    "n_iter_no_change": null,
    "tol": null,
    "ccp_alpha": null
  },
  "solution": {
    "class": {
      "module": "nextmv_sklearn.ensemble.solution",
      "name": "GradientBoostingRegressorSolution"
    },
    "attributes": {
      "n_estimators_": 100,
      "n_trees_per_iteration_": 1,
      "oob_score_": 0.0,
      "train_score_": [
        5365.788686570168,
        ...
        1195.018648214559,
        1191.6744015438958
      ],
      "init_": {
        "constant_": [
          [
            152.13348416289594
          ]
        ],
        "n_features_in_": 10,
        "n_outputs_": 1
      },
      "estimators_": [
        {
          "max_features_": 10,
          "n_features_in_": 10,
          "n_outputs_": 1,
          "tree_": "gASVqwYAAAAA....QJR0lGJ1Yi4="
        }
      ],
      "n_features_in_": 10,
      "max_features_": 10,
      "loss": "gASVYwEAAAA....AGgZiWgaiXVidWIu"
    }
  },
  "statistics": {
    "run": {
      "duration": 0.06821298599243164
    },
    "result": {
      "custom": {
        "depth": 3,
        "feature_importances_": [
          0.04868904207548975,
          0.013906683009910865,
          ...
          0.041790175253534984
        ],
        "score": 0.7990392018966865
      }
    },
    "series_data": {},
    "schema": "v1"
  },
  "assets": []
}

Load the solution and make a prediction.

main.py

import nextmv


# Load the solution from the file.
input = nextmv.load(path=file_path)
loaded_solution = nextmv.from_dict(input.data["solution"])

# Convert the solution to a model and make a prediction.
loaded_fit = loaded_solution.to_model()
print(loaded_fit.predict(X[:1]))

BashOutput

uv run main.py

[200.87337372]

Random forest¶

Write the solution.

main.py

from nextmv_sklearn import ensemble

import nextmv

# Model and statistics code here.

# Create the solution object from the model and construct an output.
solution = ensemble.RandomForestRegressorSolution.from_model(fit)
output = nextmv.Output(
    options=options,
    solution=solution.to_dict(),
    statistics=statistics,
)

# Export the output, with the solution, to a file.
file_path = "regressor.json"
nextmv.write(output, file_path)

BashOutput

cat regressor.json

{
  "options": {
    "n_estimators": null,
    "criterion": null,
    "max_depth": null,
    "min_samples_split": null,
    "min_samples_leaf": null,
    "min_weight_fraction_leaf": null,
    "max_features": null,
    "max_leaf_nodes": null,
    "min_impurity_decrease": null,
    "bootstrap": null,
    "oob_score": null,
    "n_jobs": null,
    "random_state": null,
    "verbose": null,
    "warm_start": null,
    "ccp_alpha": null,
    "max_samples": null,
    "monotonic_cst": null
  },
  "solution": {
    "class": {
      "module": "nextmv_sklearn.ensemble.solution",
      "name": "RandomForestRegressorSolution"
    },
    "attributes": {
      "estimator_": {
        "max_features_": 0,
        "n_features_in_": 0,
        "n_outputs_": 0
      },
      "estimators_": [
        {
          "max_features_": 10,
          "n_features_in_": 10,
          "n_outputs_": 1,
          "tree_": "gASVAHFAAAAAA....BtQJR0lGJ1Yi4="
        }
      ],
      "n_features_in_": 10,
      "n_outputs_": 1,
      "oob_score_": 0.0
    }
  },
  "statistics": {
    "run": {
      "duration": 0.15402698516845703
    },
    "result": {
      "custom": {
        "feature_importances_": [
          0.05862192557250181, 0.012334872517744695, 0.28574949264657873,
          0.09883856630804999, 0.047828611880482576, 0.05744602179282225,
          0.05022167910320828, 0.0236904772411412, 0.2967779353746083,
          0.06849041756286224
        ],
        "score": 0.9184641072754306
      }
    },
    "series_data": {},
    "schema": "v1"
  },
  "assets": []
}

Load the solution and make a prediction.

main.py

import nextmv


# Load the solution from the file.
input = nextmv.load(path=file_path)
loaded_solution = nextmv.from_dict(input.data["solution"])

# Convert the solution to a model and make a prediction.
loaded_fit = loaded_solution.to_model()
print(loaded_fit.predict(X[:1]))

BashOutput

uv run main.py

[176.1]

Linear model¶

Reference

Learn more by going to the reference for the linear_model.solution module.

Write the solution.

main.py

from nextmv_sklearn import linear_model

import nextmv

# Model and statistics code here.

# Create the solution object from the model and construct an output.
solution = linear_model.LinearRegressionSolution.from_model(fit)
output = nextmv.Output(
    options=options,
    solution=solution.to_dict(),
    statistics=statistics,
)

# Export the output, with the solution, to a file.
file_path = "regressor.json"
nextmv.write(output, file_path)

BashOutput

cat regressor.json

{
  "options": {
    "fit_intercept": null,
    "copy_X": null,
    "n_jobs": null,
    "positive": null
  },
  "solution": {
    "class": {
      "module": "nextmv_sklearn.linear_model.solution",
      "name": "LinearRegressionSolution"
    },
    "attributes": {
      "coef_": [
        -10.009866299810263,
        -239.81564367242262,
        519.8459200544598,
        324.38464550232345,
        -792.1756385522301,
        476.7390210052578,
        101.04326793803399,
        177.0632376713463,
        751.273699557104,
        67.62669218370486
      ],
      "rank_": 10,
      "singular_": [
        2.006043556394722,
        1.2216053690118982,
        1.0981649507815319,
        0.9774847330082032,
        0.8137452864786221,
        0.7763485529194514,
        0.7325064179373295,
        0.6585453943089912,
        0.2798571500982059,
        0.092524212112576
      ],
      "intercept_": 152.13348416289597,
      "n_features_in_": 10
    }
  },
  "statistics": {
    "run": {
      "duration": 0.0015370845794677734
    },
    "result": {
      "custom": {
        "score": 0.5177484222203499
      }
    },
    "series_data": {},
    "schema": "v1"
  },
  "assets": []
}

Load the solution and make a prediction.

main.py

import nextmv


# Load the solution from the file.
input = nextmv.load(path=file_path)
loaded_solution = nextmv.from_dict(input.data["solution"])

# Convert the solution to a model and make a prediction.
loaded_fit = loaded_solution.to_model()
print(loaded_fit.predict(X[:1]))

BashOutput

uv run main.py

[206.11667725]

Neural network¶

Reference

Learn more by going to the reference for the neural_network.solution module.

Write the solution.

main.py

from nextmv_sklearn import neural_network

import nextmv

# Model and statistics code here.

# Create the solution object from the model and construct an output.
solution = neural_network.MLPRegressorSolution.from_model(fit)
output = nextmv.Output(
    options=options,
    solution=solution.to_dict(),
    statistics=statistics,
)

# Export the output, with the solution, to a file.
file_path = "regressor.json"
nextmv.write(output, file_path)

BashOutput

cat regressor.json

{
  "options": {
    "hidden_layer_sizes": null,
    "activation": null,
    "solver": null,
    "alpha": null,
    "batch_size": null,
    "learning_rate": null,
    "learning_rate_init": null,
    "power_t": null,
    "max_iter": 2500,
    "shuffle": null,
    "random_state": null,
    "tol": null,
    "verbose": null,
    "warm_start": null,
    "momentum": null,
    "nesterovs_momentum": null,
    "early_stopping": null,
    "validation_fraction": null,
    "beta_1": null,
    "beta_2": null,
    "epsilon": null,
    "n_iter_no_change": null,
    "max_fun": null
  },
  "solution": {
    "class": {
      "module": "nextmv_sklearn.neural_network.solution",
      "name": "MLPRegressorSolution"
    },
    "attributes": {
      "loss_": 1446.176859835485,
      "best_loss_": 1445.790195444945,
      "loss_curve_": [
        14531.186602180687,
        ...
        1446.024979302395,
        1446.176859835485
      ],
      "t_": 893724,
      "coefs_": [
        [
          [
            -0.017255350152175514,
            ...
            0.04931994332298196
          ],
          ...
        ],
        [
          [
            -0.021388414548190945
          ],
          ...
        ]
      ],
      "intercepts_": [
        [
          -0.03335657429574056,
          ...
          -0.03597971253262282
        ],
        ...
      ],
      "n_features_in_": 10,
      "n_iter_": 2022,
      "n_layers_": 3,
      "n_outputs_": 1,
      "out_activation_": "identity"
    }
  },
  "statistics": {
    "run": {
      "duration": 0.9377310276031494
    },
    "result": {
      "custom": {
        "score": 0.5123771786534843
      }
    },
    "series_data": {},
    "schema": "v1"
  },
  "assets": []
}

Load the solution and make a prediction.

main.py

import nextmv


# Load the solution from the file.
input = nextmv.load(path=file_path)
loaded_solution = nextmv.from_dict(input.data["solution"])

# Convert the solution to a model and make a prediction.
loaded_fit = loaded_solution.to_model()
print(loaded_fit.predict(X[:1]))

BashOutput

uv run main.py -max_iter 2500

[199.38432603]

Tree¶

Reference

Learn more by going to the reference for the tree.solution module.

Write the solution.

main.py

from nextmv_sklearn import tree

import nextmv

# Model and statistics code here.

# Create the solution object from the model and construct an output.
solution = tree.DecisionTreeRegressorSolution.from_model(fit)
output = nextmv.Output(
    options=options,
    solution=solution.to_dict(),
    statistics=statistics,
)

# Export the output, with the solution, to a file.
file_path = "regressor.json"
nextmv.write(output, file_path)

BashOutput

cat regressor.json

{
  "options": {
    "criterion": "squared_error",
    "splitter": "best",
    "max_depth": null,
    "min_samples_split": null,
    "min_samples_leaf": null,
    "min_weight_fraction_leaf": null,
    "max_features": null,
    "random_state": null,
    "max_leaf_nodes": null,
    "min_impurity_decrease": null,
    "ccp_alpha": null
  },
  "solution": {
    "class": {
      "module": "nextmv_sklearn.tree.solution",
      "name": "DecisionTreeRegressorSolution"
    },
    "attributes": {
      "max_features_": 10,
      "n_features_in_": 10,
      "n_outputs_": 1,
      "tree_": "gASVMfUAAAAAAACMEnNrbGVhcm4udH...R0lGJ1Yi4="
    }
  },
  "statistics": {
    "run": {
      "duration": 0.0044820308685302734
    },
    "result": {
      "custom": {
        "depth": 20,
        "feature_importances_": [
          0.03869204610835461,
          0.010414593479079177,
          0.23623009367609873,
          0.08523716847272961,
          0.06178429111080888,
          0.07221426644525238,
          0.07449166752232912,
          0.012499198167788545,
          0.3394458949538394,
          0.06899078006371957
        ],
        "n_leaves": 432,
        "score": 1.0
      }
    },
    "series_data": {},
    "schema": "v1"
  },
  "assets": []
}

Load the solution and make a prediction.

main.py

import nextmv


# Load the solution from the file.
input = nextmv.load(path=file_path)
loaded_solution = nextmv.from_dict(input.data["solution"])

# Convert the solution to a model and make a prediction.
loaded_fit = loaded_solution.to_model()
print(loaded_fit.predict(X[:1]))

BashOutput

uv run main.py

[151.]