Performance-over-time plot

This example shows, how to use the performance_over_time_ attribute to plot the performance over train time. performance_over_time_ can contain multiple metrics within a pandas dataframe, namely:

• ensemble_optimization_score

• ensemble_test_score

• single_best_optimization_score

• single_best_test_score

• single_best_train_score

auto-sklearn can automatically encode categorical columns using a label/ordinal encoder. This example highlights how to properly set the dtype in a DataFrame for this to happen, and showcase how to input also testing data to autosklearn.

The X_train/y_train arguments to the fit function will be used to fit the scikit-learn model, whereas the X_test/y_test will be used to evaluate how good this scikit-learn model generalizes to unseen data (i.e. data not in X_train/y_train). Using test data is a good mechanism to measure if the trained model suffers from overfit, and more details can be found on evaluating estimator performance.

In order to provide *_test_score metrics, X_test and y_test must be provided to the AutoML-Model, as shown in this example.

There is also support to manually indicate the feature types (whether a column is categorical or numerical) via the argument feat_types from fit(). This is important when working with list or numpy arrays as there is no per-column dtype (further details in the example Feature Types).

import time

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import sklearn.model_selection
import sklearn.datasets
import sklearn.metrics

from smac.tae import StatusType

import autosklearn.classification

Data Loading

# Using Australian dataset https://www.openml.org/d/40981.
# This example will use the command fetch_openml, which will
# download a properly formatted dataframe if you use as_frame=True.
# For demonstration purposes, we will download a numpy array using
# as_frame=False, and manually creating the pandas DataFrame
X, y = sklearn.datasets.fetch_openml(data_id=40981, return_X_y=True, as_frame=False)

# bool and category will be automatically encoded.
# Targets for classification are also automatically encoded
# If using fetch_openml, data is already properly encoded, below
# is an example for user reference
X = pd.DataFrame(data=X, columns=["A" + str(i) for i in range(1, 15)])
desired_boolean_columns = ["A1"]
desired_categorical_columns = ["A4", "A5", "A6", "A8", "A9", "A11", "A12"]
desired_numerical_columns = ["A2", "A3", "A7", "A10", "A13", "A14"]
for column in X.columns:
    if column in desired_boolean_columns:
        X[column] = X[column].astype("bool")
    elif column in desired_categorical_columns:
        X[column] = X[column].astype("category")
    else:
        X[column] = pd.to_numeric(X[column])

y = pd.DataFrame(y, dtype="category")

X_train, X_test, y_train, y_test = sklearn.model_selection.train_test_split(
    X, y, test_size=0.5, random_state=3
)
print(X.dtypes)

A1         bool
A2      float64
A3      float64
A4     category
A5     category
A6     category
A7      float64
A8     category
A9     category
A10     float64
A11    category
A12    category
A13     float64
A14     float64
dtype: object

Build and fit a classifier

cls = autosklearn.classification.AutoSklearnClassifier(
    time_left_for_this_task=120,
    per_run_time_limit=30,
)
cls.fit(X_train, y_train, X_test, y_test)

Fitting to the training data:   0%|          | 0/120 [00:00<?, ?it/s, The total time budget for this task is 0:02:00]
Fitting to the training data:   1%|          | 1/120 [00:01<01:59,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:   2%|1         | 2/120 [00:02<01:58,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:   2%|2         | 3/120 [00:03<01:57,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:   3%|3         | 4/120 [00:04<01:56,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:   4%|4         | 5/120 [00:05<01:55,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:   5%|5         | 6/120 [00:06<01:54,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:   6%|5         | 7/120 [00:07<01:53,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:   7%|6         | 8/120 [00:08<01:52,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:   8%|7         | 9/120 [00:09<01:51,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:   8%|8         | 10/120 [00:10<01:50,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:   9%|9         | 11/120 [00:11<01:49,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  10%|#         | 12/120 [00:12<01:48,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  11%|#         | 13/120 [00:13<01:47,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  12%|#1        | 14/120 [00:14<01:46,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  12%|#2        | 15/120 [00:15<01:45,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  13%|#3        | 16/120 [00:16<01:44,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  14%|#4        | 17/120 [00:17<01:43,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  15%|#5        | 18/120 [00:18<01:42,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  16%|#5        | 19/120 [00:19<01:41,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  17%|#6        | 20/120 [00:20<01:40,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  18%|#7        | 21/120 [00:21<01:39,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  18%|#8        | 22/120 [00:22<01:38,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  19%|#9        | 23/120 [00:23<01:37,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  20%|##        | 24/120 [00:24<01:36,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  21%|##        | 25/120 [00:25<01:35,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  22%|##1       | 26/120 [00:26<01:34,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  22%|##2       | 27/120 [00:27<01:33,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  23%|##3       | 28/120 [00:28<01:32,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  24%|##4       | 29/120 [00:29<01:31,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  25%|##5       | 30/120 [00:30<01:30,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  26%|##5       | 31/120 [00:31<01:29,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  27%|##6       | 32/120 [00:32<01:28,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  28%|##7       | 33/120 [00:33<01:27,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  28%|##8       | 34/120 [00:34<01:26,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  29%|##9       | 35/120 [00:35<01:25,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  30%|###       | 36/120 [00:36<01:24,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  31%|###       | 37/120 [00:37<01:23,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  32%|###1      | 38/120 [00:38<01:22,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  32%|###2      | 39/120 [00:39<01:21,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  33%|###3      | 40/120 [00:40<01:20,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  34%|###4      | 41/120 [00:41<01:19,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  35%|###5      | 42/120 [00:42<01:18,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  36%|###5      | 43/120 [00:43<01:17,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  37%|###6      | 44/120 [00:44<01:16,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  38%|###7      | 45/120 [00:45<01:15,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  38%|###8      | 46/120 [00:46<01:14,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  39%|###9      | 47/120 [00:47<01:13,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  40%|####      | 48/120 [00:48<01:12,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  41%|####      | 49/120 [00:49<01:11,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  42%|####1     | 50/120 [00:50<01:10,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  42%|####2     | 51/120 [00:51<01:09,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  43%|####3     | 52/120 [00:52<01:08,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  44%|####4     | 53/120 [00:53<01:07,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  45%|####5     | 54/120 [00:54<01:06,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  46%|####5     | 55/120 [00:55<01:05,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  47%|####6     | 56/120 [00:56<01:04,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  48%|####7     | 57/120 [00:57<01:03,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  48%|####8     | 58/120 [00:58<01:02,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  49%|####9     | 59/120 [00:59<01:01,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  50%|#####     | 60/120 [01:00<01:00,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  51%|#####     | 61/120 [01:01<00:59,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  52%|#####1    | 62/120 [01:02<00:58,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  52%|#####2    | 63/120 [01:03<00:57,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  53%|#####3    | 64/120 [01:04<00:56,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  54%|#####4    | 65/120 [01:05<00:55,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  55%|#####5    | 66/120 [01:06<00:54,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  56%|#####5    | 67/120 [01:07<00:53,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  57%|#####6    | 68/120 [01:08<00:52,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  57%|#####7    | 69/120 [01:09<00:51,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  58%|#####8    | 70/120 [01:10<00:50,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  59%|#####9    | 71/120 [01:11<00:49,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  60%|######    | 72/120 [01:12<00:48,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  61%|######    | 73/120 [01:13<00:47,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  62%|######1   | 74/120 [01:14<00:46,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  62%|######2   | 75/120 [01:15<00:45,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  63%|######3   | 76/120 [01:16<00:44,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  64%|######4   | 77/120 [01:17<00:43,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  65%|######5   | 78/120 [01:18<00:42,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  66%|######5   | 79/120 [01:19<00:41,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  67%|######6   | 80/120 [01:20<00:40,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  68%|######7   | 81/120 [01:21<00:39,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  68%|######8   | 82/120 [01:22<00:38,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  69%|######9   | 83/120 [01:23<00:37,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  70%|#######   | 84/120 [01:24<00:36,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  71%|#######   | 85/120 [01:25<00:35,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  72%|#######1  | 86/120 [01:26<00:34,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  72%|#######2  | 87/120 [01:27<00:33,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  73%|#######3  | 88/120 [01:28<00:32,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  74%|#######4  | 89/120 [01:29<00:31,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  75%|#######5  | 90/120 [01:30<00:30,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  76%|#######5  | 91/120 [01:31<00:29,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  77%|#######6  | 92/120 [01:32<00:28,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  78%|#######7  | 93/120 [01:33<00:27,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  78%|#######8  | 94/120 [01:34<00:26,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  79%|#######9  | 95/120 [01:35<00:25,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  80%|########  | 96/120 [01:36<00:24,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  81%|########  | 97/120 [01:37<00:23,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  82%|########1 | 98/120 [01:38<00:22,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  82%|########2 | 99/120 [01:39<00:21,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  83%|########3 | 100/120 [01:40<00:20,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  84%|########4 | 101/120 [01:41<00:19,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  85%|########5 | 102/120 [01:42<00:18,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  86%|########5 | 103/120 [01:43<00:17,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  87%|########6 | 104/120 [01:44<00:16,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  88%|########7 | 105/120 [01:45<00:15,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  88%|########8 | 106/120 [01:46<00:14,  1.00s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  89%|########9 | 107/120 [01:47<00:13,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  90%|######### | 108/120 [01:48<00:12,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  91%|######### | 109/120 [01:49<00:11,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  92%|#########1| 110/120 [01:50<00:10,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  92%|#########2| 111/120 [01:51<00:09,  1.02s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  93%|#########3| 112/120 [01:52<00:08,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data:  94%|#########4| 113/120 [01:53<00:07,  1.01s/it, The total time budget for this task is 0:02:00]
Fitting to the training data: 100%|##########| 120/120 [01:53<00:00,  1.06it/s, The total time budget for this task is 0:02:00]

AutoSklearnClassifier(ensemble_class=<class 'autosklearn.ensembles.ensemble_selection.EnsembleSelection'>,
                      per_run_time_limit=30, time_left_for_this_task=120)

Get the Score of the final ensemble

predictions = cls.predict(X_test)
print("Accuracy score", sklearn.metrics.accuracy_score(y_test, predictions))

Accuracy score 0.8608695652173913

Plot the ensemble performance

The performance_over_time_ attribute returns a pandas dataframe, which can be directly used for plotting

poT = cls.performance_over_time_
poT.plot(
    x="Timestamp",
    kind="line",
    legend=True,
    title="Auto-sklearn accuracy over time",
    grid=True,
)
plt.show()