Start¶
In this guide, we’ll show how you can use EvalML to automatically find the best pipeline for predicting whether or not a credit card transaction is fradulent. Along the way, we’ll highlight EvalML’s built-in tools and features for understanding and interacting with the search process.
[1]:
import evalml
from evalml import AutoMLSearch
from evalml.utils import infer_feature_types
First, we load in the features and outcomes we want to use to train our model.
[2]:
X, y = evalml.demos.load_fraud(n_rows=250)
Number of Features
Boolean 1
Categorical 6
Numeric 5
Number of training examples: 250
Targets
False 88.40%
True 11.60%
Name: fraud, dtype: object
First, we will clean the data. Since EvalML accepts a pandas input, it can run type inference on this data directly. Since we’d like to change the types inferred by EvalML, we can use the infer_feature_types
utility method. Here’s what we’re going to do with the following dataset:
Reformat the
expiration_date
column so it reflects a more familiar date format.Cast the
lat
andlng
columns from float to str.Use
infer_feature_types
to specify what types certain columns should be. For example, to avoid having theprovider
column be inferred as natural language text, we have specified it as a categorical column instead.
The infer_feature_types
utility method takes a pandas or numpy input and converts it to a pandas dataframe with a Woodwork accessor, providing us with flexibility to cast the data as necessary.
[3]:
X.ww['expiration_date'] = X['expiration_date'].apply(lambda x: '20{}-01-{}'.format(x.split("/")[1], x.split("/")[0]))
X = infer_feature_types(X, feature_types= {'store_id': 'categorical',
'expiration_date': 'datetime',
'lat': 'categorical',
'lng': 'categorical',
'provider': 'categorical'})
X.ww
[3]:
Physical Type | Logical Type | Semantic Tag(s) | |
---|---|---|---|
Column | |||
card_id | int64 | Integer | ['numeric'] |
store_id | int64 | Integer | ['numeric'] |
datetime | datetime64[ns] | Datetime | [] |
amount | int64 | Integer | ['numeric'] |
currency | string | Unknown | [] |
customer_present | bool | Boolean | [] |
expiration_date | datetime64[ns] | Datetime | [] |
provider | category | Categorical | ['category'] |
lat | float64 | Double | ['numeric'] |
lng | float64 | Double | ['numeric'] |
region | category | Categorical | ['category'] |
country | category | Categorical | ['category'] |
In order to validate the results of the pipeline creation and optimization process, we will save some of our data as a holdout set.
[4]:
X_train, X_holdout, y_train, y_holdout = evalml.preprocessing.split_data(X, y, problem_type='binary', test_size=.2)
Note: To provide data to EvalML, it is recommended that you initialize a woodwork accessor so that you control how EvalML will treat each feature, such as as a numeric feature, a categorical feature, a text feature or other type of feature. Consult the the Woodwork project for help on how to do this. Here, split_data()
returns dataframes with woodwork accessors.
EvalML has many options to configure the pipeline search. At the minimum, we need to define an objective function. For simplicity, we will use the F1 score in this example. However, the real power of EvalML is in using domain-specific objective functions or building your own.
Below EvalML utilizes Bayesian optimization (EvalML’s default optimizer) to search and find the best pipeline defined by the given objective.
EvalML provides a number of parameters to control the search process. max_batches
is one of the parameters which controls the stopping criterion for the AutoML search. It indicates the maximum number of rounds of AutoML to evaluate, where each round may train and score a variable number of pipelines. In this example, max_batches
is set to 1.
** Graphing methods, like AutoMLSearch, on Jupyter Notebook and Jupyter Lab require ipywidgets to be installed.
** If graphing on Jupyter Lab, jupyterlab-plotly required. To download this, make sure you have npm installed.
[5]:
automl = AutoMLSearch(X_train=X_train, y_train=y_train,
problem_type='binary', objective='f1',
max_batches=1, verbose=True)
Generating pipelines to search over...
Removing columns ['currency'] because they are of 'Unknown' type
8 pipelines ready for search.
When we call search()
, the search for the best pipeline will begin. There is no need to wrangle with missing data or categorical variables as EvalML includes various preprocessing steps (like imputation, one-hot encoding, feature selection) to ensure you’re getting the best results. As long as your data is in a single table, EvalML can handle it. If not, you can reduce your data to a single table by utilizing Featuretools and its Entity Sets.
You can find more information on pipeline components and how to integrate your own custom pipelines into EvalML here.
[6]:
automl.search()
*****************************
* Beginning pipeline search *
*****************************
Optimizing for F1.
Greater score is better.
Using SequentialEngine to train and score pipelines.
Searching up to 1 batches for a total of 9 pipelines.
Allowed model families: linear_model, catboost, decision_tree, random_forest, lightgbm, xgboost, extra_trees
Evaluating Baseline Pipeline: Mode Baseline Binary Classification Pipeline
Mode Baseline Binary Classification Pipeline:
Starting cross validation
Finished cross validation - mean F1: 0.000
*****************************
* Evaluating Batch Number 1 *
*****************************
Elastic Net Classifier w/ Drop Columns Transformer + Imputer + DateTime Featurization Component + One Hot Encoder + Oversampler + Standard Scaler:
Starting cross validation
Finished cross validation - mean F1: 0.236
Decision Tree Classifier w/ Drop Columns Transformer + Imputer + DateTime Featurization Component + One Hot Encoder + Oversampler:
Starting cross validation
Finished cross validation - mean F1: 0.199
Random Forest Classifier w/ Drop Columns Transformer + Imputer + DateTime Featurization Component + One Hot Encoder + Oversampler:
Starting cross validation
Finished cross validation - mean F1: 0.403
LightGBM Classifier w/ Drop Columns Transformer + Imputer + DateTime Featurization Component + One Hot Encoder + Oversampler:
Starting cross validation
Finished cross validation - mean F1: 0.279
Logistic Regression Classifier w/ Drop Columns Transformer + Imputer + DateTime Featurization Component + One Hot Encoder + Oversampler + Standard Scaler:
Starting cross validation
Finished cross validation - mean F1: 0.197
XGBoost Classifier w/ Drop Columns Transformer + Imputer + DateTime Featurization Component + One Hot Encoder + Oversampler:
Starting cross validation
Finished cross validation - mean F1: 0.205
Extra Trees Classifier w/ Drop Columns Transformer + Imputer + DateTime Featurization Component + One Hot Encoder + Oversampler:
Starting cross validation
Finished cross validation - mean F1: 0.193
CatBoost Classifier w/ Drop Columns Transformer + Imputer + DateTime Featurization Component + Oversampler:
Starting cross validation
Finished cross validation - mean F1: 0.538
Search finished after 00:17
Best pipeline: CatBoost Classifier w/ Drop Columns Transformer + Imputer + DateTime Featurization Component + Oversampler
Best pipeline F1: 0.537892
If you would like to suppress stdout output, set verbose=False
. This is also the default behavior for AutoMLSearch
if verbose
is not specified.
[7]:
automl = AutoMLSearch(X_train=X_train, y_train=y_train,
problem_type='binary', objective='f1',
max_batches=1, verbose=False)
automl.search()
We also provide a standalone ``search` method <../generated/evalml.automl.search.html>`__ which does all of the above in a single line, and returns the AutoMLSearch
instance and data check results. If there were data check errors, AutoML will not be run and no AutoMLSearch
instance will be returned.
After the search is finished we can view all of the pipelines searched, ranked by score. Internally, EvalML performs cross validation to score the pipelines. If it notices a high variance across cross validation folds, it will warn you. EvalML also provides additional data checks to analyze your data to assist you in producing the best performing pipeline.
[8]:
automl.rankings
[8]:
id | pipeline_name | search_order | mean_cv_score | standard_deviation_cv_score | validation_score | percent_better_than_baseline | high_variance_cv | parameters | |
---|---|---|---|---|---|---|---|---|---|
0 | 8 | CatBoost Classifier w/ Drop Columns Transforme... | 8 | 0.537892 | 0.201574 | 0.769231 | 53.789174 | False | {'Drop Columns Transformer': {'columns': ['cur... |
1 | 3 | Random Forest Classifier w/ Drop Columns Trans... | 3 | 0.402694 | 0.040471 | 0.400000 | 40.269360 | False | {'Drop Columns Transformer': {'columns': ['cur... |
2 | 4 | LightGBM Classifier w/ Drop Columns Transforme... | 4 | 0.278620 | 0.074925 | 0.222222 | 27.861953 | False | {'Drop Columns Transformer': {'columns': ['cur... |
3 | 1 | Elastic Net Classifier w/ Drop Columns Transfo... | 1 | 0.236111 | 0.122615 | 0.142857 | 23.611111 | False | {'Drop Columns Transformer': {'columns': ['cur... |
4 | 6 | XGBoost Classifier w/ Drop Columns Transformer... | 6 | 0.204545 | 0.186031 | 0.000000 | 20.454545 | False | {'Drop Columns Transformer': {'columns': ['cur... |
5 | 2 | Decision Tree Classifier w/ Drop Columns Trans... | 2 | 0.199216 | 0.075506 | 0.213333 | 19.921569 | False | {'Drop Columns Transformer': {'columns': ['cur... |
6 | 5 | Logistic Regression Classifier w/ Drop Columns... | 5 | 0.197114 | 0.063306 | 0.142857 | 19.711400 | False | {'Drop Columns Transformer': {'columns': ['cur... |
7 | 7 | Extra Trees Classifier w/ Drop Columns Transfo... | 7 | 0.193023 | 0.167490 | 0.279070 | 19.302326 | False | {'Drop Columns Transformer': {'columns': ['cur... |
8 | 0 | Mode Baseline Binary Classification Pipeline | 0 | 0.000000 | 0.000000 | 0.000000 | 0.000000 | False | {'Baseline Classifier': {'strategy': 'mode'}} |
If we are interested in see more details about the pipeline, we can view a summary description using the id
from the rankings table:
[9]:
automl.describe_pipeline(3)
*************************************************************************************************************************************
* Random Forest Classifier w/ Drop Columns Transformer + Imputer + DateTime Featurization Component + One Hot Encoder + Oversampler *
*************************************************************************************************************************************
Problem Type: binary
Model Family: Random Forest
Pipeline Steps
==============
1. Drop Columns Transformer
* columns : ['currency']
2. Imputer
* categorical_impute_strategy : most_frequent
* numeric_impute_strategy : mean
* categorical_fill_value : None
* numeric_fill_value : None
3. DateTime Featurization Component
* features_to_extract : ['year', 'month', 'day_of_week', 'hour']
* encode_as_categories : False
* date_index : None
4. One Hot Encoder
* top_n : 10
* features_to_encode : None
* categories : None
* drop : if_binary
* handle_unknown : ignore
* handle_missing : error
5. Oversampler
* sampling_ratio : 0.25
* k_neighbors_default : 5
* n_jobs : -1
* sampling_ratio_dict : None
* k_neighbors : 5
6. Random Forest Classifier
* n_estimators : 100
* max_depth : 6
* n_jobs : -1
Training
========
Training for binary problems.
Objective to optimize binary classification pipeline thresholds for: <evalml.objectives.standard_metrics.F1 object at 0x7f72f4a229a0>
Total training time (including CV): 2.1 seconds
Cross Validation
----------------
F1 MCC Binary Log Loss Binary Gini AUC Precision Balanced Accuracy Binary Accuracy Binary # Training # Validation
0 0.400 0.476 0.277 0.593 0.797 1.000 0.625 0.910 133 67
1 0.364 0.365 0.340 0.318 0.659 0.667 0.617 0.896 133 67
2 0.444 0.513 0.253 0.724 0.862 1.000 0.643 0.924 134 66
mean 0.403 0.452 0.290 0.545 0.772 0.889 0.628 0.910 - -
std 0.040 0.077 0.045 0.207 0.104 0.192 0.013 0.014 - -
coef of var 0.101 0.170 0.156 0.380 0.134 0.217 0.021 0.016 - -
We can also view the pipeline parameters directly:
[10]:
pipeline = automl.get_pipeline(3)
print(pipeline.parameters)
{'Drop Columns Transformer': {'columns': ['currency']}, 'Imputer': {'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, 'DateTime Featurization Component': {'features_to_extract': ['year', 'month', 'day_of_week', 'hour'], 'encode_as_categories': False, 'date_index': None}, 'One Hot Encoder': {'top_n': 10, 'features_to_encode': None, 'categories': None, 'drop': 'if_binary', 'handle_unknown': 'ignore', 'handle_missing': 'error'}, 'Oversampler': {'sampling_ratio': 0.25, 'k_neighbors_default': 5, 'n_jobs': -1, 'sampling_ratio_dict': None, 'k_neighbors': 5}, 'Random Forest Classifier': {'n_estimators': 100, 'max_depth': 6, 'n_jobs': -1}}
We can now select the best pipeline and score it on our holdout data:
[11]:
pipeline = automl.best_pipeline
pipeline.score(X_holdout, y_holdout, ["f1"])
[11]:
OrderedDict([('F1', 0.6666666666666666)])
We can also visualize the structure of the components contained by the pipeline:
[12]:
pipeline.graph()
[12]: