Binary classification¶

---

This example shows how to use ATOM to solve a binary classification problem. Additonnaly, we'll perform a variety of data cleaning steps to prepare the data for modelling.

The data used is a variation on the Australian weather dataset from Kaggle. You can download it from here. The goal of this dataset is to predict whether or not it will rain tomorrow training a binary classifier on target RainTomorrow.

Load the data¶

In [1]:

# Import packages
import pandas as pd
from atom import ATOMClassifier

# Import packages import pandas as pd from atom import ATOMClassifier

In [2]:

# Load data
X = pd.read_csv("./datasets/weatherAUS.csv")

# Let's have a look
X.head()

# Load data X = pd.read_csv("./datasets/weatherAUS.csv") # Let's have a look X.head()

Out[2]:

	Location	MinTemp	MaxTemp	Rainfall	Evaporation	Sunshine	WindGustDir	WindGustSpeed	WindDir9am	WindDir3pm	...	Humidity9am	Humidity3pm	Pressure9am	Pressure3pm	Cloud9am	Cloud3pm	Temp9am	Temp3pm	RainToday	RainTomorrow
0	MelbourneAirport	18.0	26.9	21.4	7.0	8.9	SSE	41.0	W	SSE	...	95.0	54.0	1019.5	1017.0	8.0	5.0	18.5	26.0	Yes	0
1	Adelaide	17.2	23.4	0.0	NaN	NaN	S	41.0	S	WSW	...	59.0	36.0	1015.7	1015.7	NaN	NaN	17.7	21.9	No	0
2	Cairns	18.6	24.6	7.4	3.0	6.1	SSE	54.0	SSE	SE	...	78.0	57.0	1018.7	1016.6	3.0	3.0	20.8	24.1	Yes	0
3	Portland	13.6	16.8	4.2	1.2	0.0	ESE	39.0	ESE	ESE	...	76.0	74.0	1021.4	1020.5	7.0	8.0	15.6	16.0	Yes	1
4	Walpole	16.4	19.9	0.0	NaN	NaN	SE	44.0	SE	SE	...	78.0	70.0	1019.4	1018.9	NaN	NaN	17.4	18.1	No	0

5 rows × 22 columns

Run the pipeline¶

In [3]:

# Call ATOM using only 5% of the complete dataset (for explanatory purposes)
atom = ATOMClassifier(X, "RainTomorrow", n_rows=0.05, warnings=False, verbose=2)

# Call ATOM using only 5% of the complete dataset (for explanatory purposes) atom = ATOMClassifier(X, "RainTomorrow", n_rows=0.05, warnings=False, verbose=2)

<< ================== ATOM ================== >>
Algorithm task: binary classification.

Dataset stats ====================== >>
Shape: (7109, 22)
Scaled: False
Missing values: 15680 (10.0%)
Categorical features: 5 (23.8%)
---------------------------------------
Train set size: 5688
Test set size: 1421
---------------------------------------
|    | dataset    | train      | test       |
|---:|:-----------|:-----------|:-----------|
|  0 | 5538 (3.5) | 4430 (3.5) | 1108 (3.5) |
|  1 | 1571 (1.0) | 1258 (1.0) | 313 (1.0)  |

In [4]:

# Encode the target column
atom.clean()

# Encode the target column atom.clean()

Applying data cleaning...

In [5]:

# Impute missing values
atom.impute(strat_num="median", strat_cat="drop", max_nan_rows=0.8)

# Impute missing values atom.impute(strat_num="median", strat_cat="drop", max_nan_rows=0.8)

Fitting Imputer...
Imputing missing values...
 --> Dropping 780 samples for containing less than 80% non-missing values.
 --> Imputing 3 missing values with median (12.0) in feature MinTemp.
 --> Imputing 4 missing values with median (22.7) in feature MaxTemp.
 --> Imputing 30 missing values with median (0.0) in feature Rainfall.
 --> Imputing 2291 missing values with median (4.8) in feature Evaporation.
 --> Imputing 2614 missing values with median (8.4) in feature Sunshine.
 --> Dropping 199 samples due to missing values in feature WindGustDir.
 --> Dropping 298 samples due to missing values in feature WindDir9am.
 --> Dropping 20 samples due to missing values in feature WindDir3pm.
 --> Imputing 24 missing values with median (69.0) in feature Humidity9am.
 --> Imputing 65 missing values with median (52.0) in feature Humidity3pm.
 --> Imputing 52 missing values with median (1017.8) in feature Pressure9am.
 --> Imputing 50 missing values with median (1015.3) in feature Pressure3pm.
 --> Imputing 1842 missing values with median (5.0) in feature Cloud9am.
 --> Imputing 1932 missing values with median (5.0) in feature Cloud3pm.
 --> Imputing 5 missing values with median (16.7) in feature Temp9am.
 --> Imputing 39 missing values with median (21.2) in feature Temp3pm.
 --> Dropping 30 samples due to missing values in feature RainToday.

In [6]:

# Encode the categorical features
atom.encode(strategy="Target", max_onehot=10, frac_to_other=0.04)

# Encode the categorical features atom.encode(strategy="Target", max_onehot=10, frac_to_other=0.04)

Fitting Encoder...
Encoding categorical columns...
 --> Target-encoding feature Location. Contains 1 classes.
 --> Target-encoding feature WindGustDir. Contains 16 classes.
 --> Target-encoding feature WindDir9am. Contains 16 classes.
 --> Target-encoding feature WindDir3pm. Contains 16 classes.
 --> Ordinal-encoding feature RainToday. Contains 2 classes.

In [7]:

# Train an Extra-Trees and a Random Forest model
atom.run(models=["ET", "RF"], metric="f1", n_bootstrap=5)

# Train an Extra-Trees and a Random Forest model atom.run(models=["ET", "RF"], metric="f1", n_bootstrap=5)

Training ===================================== >>
Models: ET, RF
Metric: f1


Results for Extra-Trees:         
Fit ---------------------------------------------
Train evaluation --> f1: 1.0
Test evaluation --> f1: 0.5274
Time elapsed: 0.459s
Bootstrap ---------------------------------------
Evaluation --> f1: 0.529 ± 0.0089
Time elapsed: 1.696s
-------------------------------------------------
Total time: 2.156s


Results for Random Forest:         
Fit ---------------------------------------------
Train evaluation --> f1: 1.0
Test evaluation --> f1: 0.5728
Time elapsed: 0.686s
Bootstrap ---------------------------------------
Evaluation --> f1: 0.5525 ± 0.0191
Time elapsed: 2.923s
-------------------------------------------------
Total time: 3.609s


Final results ========================= >>
Duration: 5.765s
------------------------------------------
Extra-Trees   --> f1: 0.529 ± 0.0089 ~
Random Forest --> f1: 0.5525 ± 0.0191 ~ !

Analyze the results¶

In [8]:

# Let's have a look at the final results
atom.results

# Let's have a look at the final results atom.results

Out[8]:

	metric_train	metric_test	time_fit	mean_bootstrap	std_bootstrap	time_bootstrap	time
ET	1.0	0.527363	0.459s	0.528980	0.008948	1.696s	2.156s
RF	1.0	0.572816	0.686s	0.552495	0.019102	2.923s	3.609s

In [9]:

# Visualize the bootstrap results
atom.plot_results(title="RF vs ET performance")

# Visualize the bootstrap results atom.plot_results(title="RF vs ET performance")

In [10]:

# Print the results of some common metrics
atom.scoring()

# Print the results of some common metrics atom.scoring()

Out[10]:

	accuracy	average_precision	balanced_accuracy	f1	jaccard	matthews_corrcoef	precision	recall	roc_auc
ET	0.836489	0.692552	0.684146	0.527363	0.358108	0.466585	0.741259	0.409266	0.864461
RF	0.848537	0.709926	0.708419	0.572816	0.401361	0.513063	0.771242	0.455598	0.868884

In [11]:

# The winner attribute calls the best model (atom.winner == atom.rf)
print(f"The winner is the {atom.winner.fullname} model!!")

# The winner attribute calls the best model (atom.winner == atom.rf) print(f"The winner is the {atom.winner.fullname} model!!")

And the winner is the Random Forest model!!

In [12]:

# Visualize the distribution of predicted probabilities
atom.winner.plot_probabilities()

# Visualize the distribution of predicted probabilities atom.winner.plot_probabilities()

In [13]:

# Compare how different metrics perform for different thresholds
atom.winner.plot_threshold(metric=["f1", "accuracy", "average_precision"], steps=50)

# Compare how different metrics perform for different thresholds atom.winner.plot_threshold(metric=["f1", "accuracy", "average_precision"], steps=50)