Para este caso usaremos el DataSet de UCI default of credit card clients Data Set
Descripcion de las columnas:
X1: Monto del crédito otorgado ($USD)
X2: Género (1=hombre, 2=mujer)
X3: Educación (1=secundaria incompleta,2=universitario,3=secundaria completa,4=otros)
X4: Estado civil (1=casado,2=soltero,3=otros)
X5: Edad (años)
X6-X11: Historial de pagos (6 últimos meses|-1 = pago en fecha; 1 = retraso de un mes; 2 = retraso de dos meses; . . . 8 = retraso en ocho meses; 9 = retraso en nueve meses o más.)
X12-X17: Monto de deuda mensual (6 últimos meses)
X18-X23: Monto de pago mensual (6 últimos meses)
Y: Sí/No (1 / 0)
Instalamos csvkit para convertir cd xls a csv
!pip install csvkit
Collecting csvkit
Downloading https://files.pythonhosted.org/packages/66/d8/206e4da52bcf9cc29dfa3a93837b14b37ba42f58ccbd22a42a3b3ae0381a/csvkit-1.0.4.tar.gz (3.8MB)
|████████████████████████████████| 3.8MB 9.8MB/s
Collecting agate>=1.6.1
Downloading https://files.pythonhosted.org/packages/92/77/ef675f16486884ff7f77f3cb87aafa3429c6bb869d4d73ee23bf4675e384/agate-1.6.1-py2.py3-none-any.whl (98kB)
|████████████████████████████████| 102kB 13.3MB/s
Collecting agate-excel>=0.2.2
Downloading https://files.pythonhosted.org/packages/a9/cd/ba7ce638900a91f00e6ebaa72c46fc90bfc13cb595071cee82c96057d5d6/agate-excel-0.2.3.tar.gz (153kB)
|████████████████████████████████| 163kB 68.1MB/s
Collecting agate-dbf>=0.2.0
Downloading https://files.pythonhosted.org/packages/cb/05/8bd93fd8f47354e5a31b1ba5876a9498a59fa32166b2e3315da43774adb8/agate-dbf-0.2.1.tar.gz
Collecting agate-sql>=0.5.3
Downloading https://files.pythonhosted.org/packages/4a/fb/796c6e7b625fde74274786da69f08aca5c5eefb891db77344f95ad7b75db/agate-sql-0.5.4.tar.gz
Requirement already satisfied: six>=1.6.1 in /usr/local/lib/python3.6/dist-packages (from csvkit) (1.12.0)
Collecting leather>=0.3.2
Downloading https://files.pythonhosted.org/packages/45/f4/692a53df6708caca1c6d088c6d9003940f164f98bd9df2bdc86233641e9c/leather-0.3.3-py3-none-any.whl
Collecting isodate>=0.5.4
Downloading https://files.pythonhosted.org/packages/9b/9f/b36f7774ff5ea8e428fdcfc4bb332c39ee5b9362ddd3d40d9516a55221b2/isodate-0.6.0-py2.py3-none-any.whl (45kB)
|████████████████████████████████| 51kB 7.8MB/s
Requirement already satisfied: Babel>=2.0 in /usr/local/lib/python3.6/dist-packages (from agate>=1.6.1->csvkit) (2.7.0)
Requirement already satisfied: python-slugify>=1.2.1 in /usr/local/lib/python3.6/dist-packages (from agate>=1.6.1->csvkit) (4.0.0)
Collecting parsedatetime>=2.1
Downloading https://files.pythonhosted.org/packages/e3/b3/02385db13f1f25f04ad7895f35e9fe3960a4b9d53112775a6f7d63f264b6/parsedatetime-2.4.tar.gz (58kB)
|████████████████████████████████| 61kB 9.3MB/s
Collecting pytimeparse>=1.1.5
Downloading https://files.pythonhosted.org/packages/1b/b4/afd75551a3b910abd1d922dbd45e49e5deeb4d47dc50209ce489ba9844dd/pytimeparse-1.1.8-py2.py3-none-any.whl
Requirement already satisfied: xlrd>=0.9.4 in /usr/local/lib/python3.6/dist-packages (from agate-excel>=0.2.2->csvkit) (1.1.0)
Requirement already satisfied: openpyxl>=2.3.0 in /usr/local/lib/python3.6/dist-packages (from agate-excel>=0.2.2->csvkit) (2.5.9)
Collecting dbfread>=2.0.5
Downloading https://files.pythonhosted.org/packages/4c/94/51349e43503e30ed7b4ecfe68a8809cdb58f722c0feb79d18b1f1e36fe74/dbfread-2.0.7-py2.py3-none-any.whl
Requirement already satisfied: sqlalchemy>=1.0.8 in /usr/local/lib/python3.6/dist-packages (from agate-sql>=0.5.3->csvkit) (1.3.10)
Requirement already satisfied: pytz>=2015.7 in /usr/local/lib/python3.6/dist-packages (from Babel>=2.0->agate>=1.6.1->csvkit) (2018.9)
Requirement already satisfied: text-unidecode>=1.3 in /usr/local/lib/python3.6/dist-packages (from python-slugify>=1.2.1->agate>=1.6.1->csvkit) (1.3)
Requirement already satisfied: future in /usr/local/lib/python3.6/dist-packages (from parsedatetime>=2.1->agate>=1.6.1->csvkit) (0.16.0)
Requirement already satisfied: jdcal in /usr/local/lib/python3.6/dist-packages (from openpyxl>=2.3.0->agate-excel>=0.2.2->csvkit) (1.4.1)
Requirement already satisfied: et-xmlfile in /usr/local/lib/python3.6/dist-packages (from openpyxl>=2.3.0->agate-excel>=0.2.2->csvkit) (1.0.1)
Building wheels for collected packages: csvkit, agate-excel, agate-dbf, agate-sql, parsedatetime
Building wheel for csvkit (setup.py) ... done
Created wheel for csvkit: filename=csvkit-1.0.4-cp36-none-any.whl size=41398 sha256=df4d7bd53b3e5e93edb062b7152e9f60c658864fa01c982ca90a3ca8323ab4cc
Stored in directory: /root/.cache/pip/wheels/5f/be/3f/d151aff6c6cce1aa1d56233d68c4b9d38b045bbe5fda018d45
Building wheel for agate-excel (setup.py) ... done
Created wheel for agate-excel: filename=agate_excel-0.2.3-py2.py3-none-any.whl size=6271 sha256=52b17db71ae31e776b1796be5a7638cd3db9dfd9e5c28d7fdb9b6bcc2563fe4f
Stored in directory: /root/.cache/pip/wheels/8a/2f/99/dbf1c6af14192030927240678c0d2176b479dcc44b51a3a6d0
Building wheel for agate-dbf (setup.py) ... done
Created wheel for agate-dbf: filename=agate_dbf-0.2.1-py2.py3-none-any.whl size=3520 sha256=2b992cda1ef453edbaa97f1606c4a7dba93a8ae01d696a9a1178a78528dac04f
Stored in directory: /root/.cache/pip/wheels/06/21/5a/84eb0a6b77a3b3d3254b30aaf993533b2236fd29083e120b24
Building wheel for agate-sql (setup.py) ... done
Created wheel for agate-sql: filename=agate_sql-0.5.4-py2.py3-none-any.whl size=7015 sha256=7378becb9a656da925f1114e0d50745ed9f52648b69532a1e171044d7876c06e
Stored in directory: /root/.cache/pip/wheels/4c/a3/09/e002ac6fb3921eb0aefd511f7aeca7d0b817b4fdb4b441cc7e
Building wheel for parsedatetime (setup.py) ... done
Created wheel for parsedatetime: filename=parsedatetime-2.4-cp36-none-any.whl size=42748 sha256=24c5f79302f665c6bd9d8d3c73b8df42b4efec55ae575c2d1e3cc6973750ba8a
Stored in directory: /root/.cache/pip/wheels/e9/d0/db/aa6af26d9762852afc0c982d96f9b4f29a373205889453555b
Successfully built csvkit agate-excel agate-dbf agate-sql parsedatetime
Installing collected packages: leather, isodate, parsedatetime, pytimeparse, agate, agate-excel, dbfread, agate-dbf, agate-sql, csvkit
Successfully installed agate-1.6.1 agate-dbf-0.2.1 agate-excel-0.2.3 agate-sql-0.5.4 csvkit-1.0.4 dbfread-2.0.7 isodate-0.6.0 leather-0.3.3 parsedatetime-2.4 pytimeparse-1.1.8Validamos que el archivo no exista para descargarlo:
%%bash
if [ ! -f "default_credit_card.csv" ]; then
wget archive.ics.uci.edu/ml/machine-learning-databases/00350/default%20of%20credit%20card%20clients.xls
in2csv "default of credit card clients.xls" > default_credit_card.csv
fi
ls -l
total 9684
-rw-r--r-- 1 root root 4367295 Nov 9 20:19 default_credit_card.csv
-rw-r--r-- 1 root root 5539328 Jan 26 2016 default of credit card clients.xls
drwxr-xr-x 1 root root 4096 Nov 6 16:17 sample_data
--2019-11-09 20:19:21-- http://archive.ics.uci.edu/ml/machine-learning-databases/00350/default%20of%20credit%20card%20clients.xls
Resolving archive.ics.uci.edu (archive.ics.uci.edu)... 128.195.10.252
Connecting to archive.ics.uci.edu (archive.ics.uci.edu)|128.195.10.252|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 5539328 (5.3M) [application/x-httpd-php]
Saving to: ‘default of credit card clients.xls’
0K .......... .......... .......... .......... .......... 0% 167K 32s
50K .......... .......... .......... .......... .......... 1% 353K 23s
100K .......... .......... .......... .......... .......... 2% 191M 15s
150K .......... .......... .......... .......... .......... 3% 261M 12s
200K .......... .......... .......... .......... .......... 4% 357K 12s
250K .......... .......... .......... .......... .......... 5% 40.6M 10s
300K .......... .......... .......... .......... .......... 6% 362K 10s
350K .......... .......... .......... .......... .......... 7% 40.3M 9s
400K .......... .......... .......... .......... .......... 8% 38.1M 8s
450K .......... .......... .......... .......... .......... 9% 37.0M 7s
500K .......... .......... .......... .......... .......... 10% 374K 8s
550K .......... .......... .......... .......... .......... 11% 35.9M 7s
600K .......... .......... .......... .......... .......... 12% 64.5M 6s
650K .......... .......... .......... .......... .......... 12% 63.4M 6s
700K .......... .......... .......... .......... .......... 13% 64.3M 5s
750K .......... .......... .......... .......... .......... 14% 66.8M 5s
800K .......... .......... .......... .......... .......... 15% 77.3M 5s
850K .......... .......... .......... .......... .......... 16% 95.4M 4s
900K .......... .......... .......... .......... .......... 17% 77.0M 4s
950K .......... .......... .......... .......... .......... 18% 377K 4s
1000K .......... .......... .......... .......... .......... 19% 83.8M 4s
1050K .......... .......... .......... .......... .......... 20% 85.5M 4s
1100K .......... .......... .......... .......... .......... 21% 82.7M 4s
1150K .......... .......... .......... .......... .......... 22% 91.6M 4s
1200K .......... .......... .......... .......... .......... 23% 98.0M 3s
1250K .......... .......... .......... .......... .......... 24% 73.5M 3s
1300K .......... .......... .......... .......... .......... 24% 85.6M 3s
1350K .......... .......... .......... .......... .......... 25% 79.0M 3s
1400K .......... .......... .......... .......... .......... 26% 465M 3s
1450K .......... .......... .......... .......... .......... 27% 114M 3s
1500K .......... .......... .......... .......... .......... 28% 126M 2s
1550K .......... .......... .......... .......... .......... 29% 101M 2s
1600K .......... .......... .......... .......... .......... 30% 76.2M 2s
1650K .......... .......... .......... .......... .......... 31% 51.6M 2s
1700K .......... .......... .......... .......... .......... 32% 89.7M 2s
1750K .......... .......... .......... .......... .......... 33% 383K 2s
1800K .......... .......... .......... .......... .......... 34% 64.3M 2s
1850K .......... .......... .......... .......... .......... 35% 102M 2s
1900K .......... .......... .......... .......... .......... 36% 111M 2s
1950K .......... .......... .......... .......... .......... 36% 123M 2s
2000K .......... .......... .......... .......... .......... 37% 116M 2s
2050K .......... .......... .......... .......... .......... 38% 105M 2s
2100K .......... .......... .......... .......... .......... 39% 112M 2s
2150K .......... .......... .......... .......... .......... 40% 110M 2s
2200K .......... .......... .......... .......... .......... 41% 123M 2s
2250K .......... .......... .......... .......... .......... 42% 115M 2s
2300K .......... .......... .......... .......... .......... 43% 47.6M 1s
2350K .......... .......... .......... .......... .......... 44% 135M 1s
2400K .......... .......... .......... .......... .......... 45% 110M 1s
2450K .......... .......... .......... .......... .......... 46% 113M 1s
2500K .......... .......... .......... .......... .......... 47% 114M 1s
2550K .......... .......... .......... .......... .......... 48% 109M 1s
2600K .......... .......... .......... .......... .......... 48% 68.8M 1s
2650K .......... .......... .......... .......... .......... 49% 86.0M 1s
2700K .......... .......... .......... .......... .......... 50% 127M 1s
2750K .......... .......... .......... .......... .......... 51% 88.9M 1s
2800K .......... .......... .......... .......... .......... 52% 146M 1s
2850K .......... .......... .......... .......... .......... 53% 112M 1s
2900K .......... .......... .......... .......... .......... 54% 105M 1s
2950K .......... .......... .......... .......... .......... 55% 142M 1s
3000K .......... .......... .......... .......... .......... 56% 89.8M 1s
3050K .......... .......... .......... .......... .......... 57% 117M 1s
3100K .......... .......... .......... .......... .......... 58% 133M 1s
3150K .......... .......... .......... .......... .......... 59% 101M 1s
3200K .......... .......... .......... .......... .......... 60% 130M 1s
3250K .......... .......... .......... .......... .......... 61% 98.8M 1s
3300K .......... .......... .......... .......... .......... 61% 55.4M 1s
3350K .......... .......... .......... .......... .......... 62% 135M 1s
3400K .......... .......... .......... .......... .......... 63% 104M 1s
3450K .......... .......... .......... .......... .......... 64% 407K 1s
3500K .......... .......... .......... .......... .......... 65% 168M 1s
3550K .......... .......... .......... .......... .......... 66% 84.5M 1s
3600K .......... .......... .......... .......... .......... 67% 116M 1s
3650K .......... .......... .......... .......... .......... 68% 118M 1s
3700K .......... .......... .......... .......... .......... 69% 101M 1s
3750K .......... .......... .......... .......... .......... 70% 124M 1s
3800K .......... .......... .......... .......... .......... 71% 101M 1s
3850K .......... .......... .......... .......... .......... 72% 107M 0s
3900K .......... .......... .......... .......... .......... 73% 126M 0s
3950K .......... .......... .......... .......... .......... 73% 110M 0s
4000K .......... .......... .......... .......... .......... 74% 113M 0s
4050K .......... .......... .......... .......... .......... 75% 135M 0s
4100K .......... .......... .......... .......... .......... 76% 76.4M 0s
4150K .......... .......... .......... .......... .......... 77% 122M 0s
4200K .......... .......... .......... .......... .......... 78% 161M 0s
4250K .......... .......... .......... .......... .......... 79% 95.6M 0s
4300K .......... .......... .......... .......... .......... 80% 139M 0s
4350K .......... .......... .......... .......... .......... 81% 101M 0s
4400K .......... .......... .......... .......... .......... 82% 114M 0s
4450K .......... .......... .......... .......... .......... 83% 113M 0s
4500K .......... .......... .......... .......... .......... 84% 117M 0s
4550K .......... .......... .......... .......... .......... 85% 109M 0s
4600K .......... .......... .......... .......... .......... 85% 108M 0s
4650K .......... .......... .......... .......... .......... 86% 52.8M 0s
4700K .......... .......... .......... .......... .......... 87% 62.1M 0s
4750K .......... .......... .......... .......... .......... 88% 67.7M 0s
4800K .......... .......... .......... .......... .......... 89% 98.2M 0s
4850K .......... .......... .......... .......... .......... 90% 107M 0s
4900K .......... .......... .......... .......... .......... 91% 103M 0s
4950K .......... .......... .......... .......... .......... 92% 112M 0s
5000K .......... .......... .......... .......... .......... 93% 103M 0s
5050K .......... .......... .......... .......... .......... 94% 136M 0s
5100K .......... .......... .......... .......... .......... 95% 85.9M 0s
5150K .......... .......... .......... .......... .......... 96% 158M 0s
5200K .......... .......... .......... .......... .......... 97% 54.8M 0s
5250K .......... .......... .......... .......... .......... 97% 122M 0s
5300K .......... .......... .......... .......... .......... 98% 96.4M 0s
5350K .......... .......... .......... .......... .......... 99% 153M 0s
5400K ......... 100% 83.4M=1.3s
2019-11-09 20:19:23 (4.09 MB/s) - ‘default of credit card clients.xls’ saved [5539328/5539328]
/usr/local/lib/python3.6/dist-packages/agate/utils.py:276: UnnamedColumnWarning: Column 0 has no name. Using "a".Creamos funciones necesarias que usaremos más adelante:
# Funciones
import numpy as np
from sklearn.metrics import confusion_matrix
from sklearn.utils.multiclass import unique_labels
from sklearn.externals import joblib
import seaborn as sns
import matplotlib.pyplot as plt
def plot_confusion_matrix(y_true, y_pred,
normalize=False,
title=None):
"""
Esta función imprime y traza la matriz de confusión.
La normalización se puede aplicar configurando `normalize=True`.
"""
if not title:
if normalize:
title = 'Matriz de Confusión Normalizada'
else:
title = 'Matriz de Confusión sin Normalizar'
# Calculando la Matriz de Confusion
cm = confusion_matrix(y_true, y_pred)
# solo usar las etiquetas que se tienen en la data
classes = unique_labels(y_true, y_pred)
if normalize:
cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]
print("Matriz de Confusión Normalizada")
else:
print('Matriz de Confusión sin Normalizar')
print(cm)
fig, ax = plt.subplots()
im = ax.imshow(cm, interpolation='nearest', cmap=plt.cm.Blues)
ax.figure.colorbar(im, ax=ax)
ax.grid(linewidth=.0)
# Queremos mostrar todos los puntos...
ax.set(xticks=np.arange(cm.shape[1]),
yticks=np.arange(cm.shape[0]),
# ... etiquetando la lista de datos
xticklabels=classes, yticklabels=classes,
title=title,
ylabel='True label',
xlabel='Predicted label')
# rotando las etiquedas de los puntos.
plt.setp(ax.get_xticklabels(), rotation=45, ha="right",rotation_mode="anchor")
# Loop over data dimensions and create text annotations.
fmt = '.2f' if normalize else 'd'
thresh = cm.max() / 2.
for i in range(cm.shape[0]):
for j in range(cm.shape[1]):
ax.text(j, i, format(cm[i, j], fmt),
ha="center", va="center",
color="white" if cm[i, j] > thresh else "black")
fig.tight_layout()
plt.show()
return ax
def saveFile(object_to_save, scaler_filename):
joblib.dump(object_to_save, scaler_filename)
def loadFile(scaler_filename):
return joblib.load(scaler_filename)
def plotHistogram(dataset_final):
dataset_final.hist(figsize=(20,14), edgecolor="black", bins=40)
plt.show()
def plotCorrelations(dataset_final):
fig, ax = plt.subplots(figsize=(10,8)) # size in inches
g = sns.heatmap(dataset_final.corr(), annot=True, cmap="YlGnBu", ax=ax)
g.set_yticklabels(g.get_yticklabels(), rotation = 0)
g.set_xticklabels(g.get_xticklabels(), rotation = 45)
fig.tight_layout()
plt.show()
Cargamos y limpiamos la data quitamos la segunda fila innecesaria y volvemos a castear los tipos de datos de las columnas:
# Importando librerías
import pandas as pd
import numpy as np
from sklearn.preprocessing import LabelEncoder
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
# Importando Datasets
dataset_csv = pd.read_csv('default_credit_card.csv')
# Columnas de la data
print ("\nColumnas del DataSet: ")
print (dataset_csv.columns)
print ("\nDataset Total: ")
print("\n",dataset_csv.head())
# Delete the first five rows using iloc selector
dataset = dataset_csv.iloc[2:,]
dataset = dataset.iloc[:,1:25]
dataset_columns = dataset.columns
dataset_values = dataset.values
print ("\nDataset reducido: ")
print("\n",dataset.head())
# Describir la data original
print ("\nDataset original:\n", dataset.describe(include='all'))
#Casteando las columnas
dataset.X1 = dataset.X1.astype(np.number)
dataset.X2 = dataset.X2.astype(np.number)
dataset.X3 = dataset.X3.astype(np.number)
dataset.X4 = dataset.X4.astype(np.number)
dataset.X5 = dataset.X5.astype(np.number)
dataset.X6 = dataset.X6.astype(np.number)
dataset.X7 = dataset.X7.astype(np.number)
dataset.X8 = dataset.X8.astype(np.number)
dataset.X9 = dataset.X9.astype(np.number)
dataset.X10 = dataset.X10.astype(np.number)
dataset.X11 = dataset.X11.astype(np.number)
dataset.X12 = dataset.X12.astype(np.number)
dataset.X13 = dataset.X13.astype(np.number)
dataset.X14 = dataset.X14.astype(np.number)
dataset.X15 = dataset.X15.astype(np.number)
dataset.X16 = dataset.X16.astype(np.number)
dataset.X17 = dataset.X17.astype(np.number)
dataset.X18 = dataset.X18.astype(np.number)
dataset.X19 = dataset.X19.astype(np.number)
dataset.X20 = dataset.X20.astype(np.number)
dataset.X21 = dataset.X21.astype(np.number)
dataset.X22 = dataset.X22.astype(np.number)
dataset.X23 = dataset.X23.astype(np.number)
dataset.Y = dataset.Y.str.replace('.0', '').astype(int)
# Revisamos los tipos de datos de las Columnas
print ("\nTipos de Columnas del Dataset: ")
print(dataset.dtypes)
Columnas del DataSet:
Index(['a', 'X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10', 'X11',
'X12', 'X13', 'X14', 'X15', 'X16', 'X17', 'X18', 'X19', 'X20', 'X21',
'X22', 'X23', 'Y'],
dtype='object')
Dataset Total:
a X1 X2 ... X22 X23 Y
0 ID LIMIT_BAL SEX ... PAY_AMT5 PAY_AMT6 default payment next month
1 1.0 20000.0 2.0 ... 0.0 0.0 1.0
2 2.0 120000.0 2.0 ... 0.0 2000.0 1.0
3 3.0 90000.0 2.0 ... 1000.0 5000.0 0.0
4 4.0 50000.0 2.0 ... 1069.0 1000.0 0.0
[5 rows x 25 columns]
Dataset reducido:
X1 X2 X3 X4 X5 ... X20 X21 X22 X23 Y
2 120000.0 2.0 2.0 2.0 26.0 ... 1000.0 1000.0 0.0 2000.0 1.0
3 90000.0 2.0 2.0 2.0 34.0 ... 1000.0 1000.0 1000.0 5000.0 0.0
4 50000.0 2.0 2.0 1.0 37.0 ... 1200.0 1100.0 1069.0 1000.0 0.0
5 50000.0 1.0 2.0 1.0 57.0 ... 10000.0 9000.0 689.0 679.0 0.0
6 50000.0 1.0 1.0 2.0 37.0 ... 657.0 1000.0 1000.0 800.0 0.0
[5 rows x 24 columns]
Dataset original:
X1 X2 X3 X4 X5 ... X20 X21 X22 X23 Y
count 29999 29999 29999 29999 29999 ... 29999 29999 29999 29999 29999
unique 81 2 7 4 56 ... 7518 6937 6897 6939 2
top 50000.0 2.0 2.0 2.0 29.0 ... 0.0 0.0 0.0 0.0 0.0
freq 3365 18111 14029 15964 1605 ... 5967 6407 6702 7172 23364
[4 rows x 24 columns]
Tipos de Columnas del Dataset:
X1 float64
X2 float64
X3 float64
X4 float64
X5 float64
X6 float64
X7 float64
X8 float64
X9 float64
X10 float64
X11 float64
X12 float64
X13 float64
X14 float64
X15 float64
X16 float64
X17 float64
X18 float64
X19 float64
X20 float64
X21 float64
X22 float64
X23 float64
Y int64
dtype: objectEscalamos y normalizamos los valores:
# Escalamiento/Normalización de Features (StandardScaler: (x-u)/s)
stdScaler = StandardScaler()
dataset_values[:,0:23] = stdScaler.fit_transform(dataset_values[:,0:23])
# Dataset final normalizado
dataset_final = pd.DataFrame(dataset_values,columns=dataset_columns, dtype=np.float64)
print ("\nDataset Final:")
print(dataset_final.describe(include='all'))
print("\n", dataset_final.head())
Dataset Final:
X1 X2 ... X23 Y
count 2.999900e+04 2.999900e+04 ... 2.999900e+04 29999.000000
mean 9.979387e-16 4.700634e-15 ... 6.060447e-17 0.221174
std 1.000017e+00 1.000017e+00 ... 1.000017e+00 0.415044
min -1.213838e+00 -1.234289e+00 ... -2.933874e-01 0.000000
25% -9.055406e-01 -1.234289e+00 ... -2.867497e-01 0.000000
50% -2.118715e-01 8.101831e-01 ... -2.090108e-01 0.000000
75% 5.588719e-01 8.101831e-01 ... -6.838310e-02 0.000000
max 6.416522e+00 8.101831e-01 ... 2.944464e+01 1.000000
[8 rows x 24 columns]
X1 X2 X3 X4 ... X21 X22 X23 Y
0 -0.366020 0.810183 0.185831 0.858524 ... -0.244236 -0.314142 -0.180885 1.0
1 -0.597243 0.810183 0.185831 0.858524 ... -0.244236 -0.248689 -0.012132 0.0
2 -0.905541 0.810183 0.185831 -1.057332 ... -0.237853 -0.244173 -0.237136 0.0
3 -0.905541 -1.234289 0.185831 -1.057332 ... 0.266419 -0.269045 -0.255193 0.0
4 -0.905541 -1.234289 -1.079434 0.858524 ... -0.244236 -0.248689 -0.248387 0.0
[5 rows x 24 columns]Graficando datos:
# Distribuciones de la data y Correlaciones
print("\n Histogramas:")
plotHistogram(dataset_final)
print("\n Correlaciones:")
plotCorrelations(dataset_final)
Separamos los predictores del objetivo y partimos la data en 80% / 20%
# Obteniendo valores a procesar
X = dataset_final.iloc[:, 0:23].values
y = dataset_final.iloc[:, 23].values
# Dividiendo el Dataset en sets de Training y Test
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 0)
Creamos una arquitetura:
Entrada => 23
Oculta => 20 / 10 / 5
Salida => 1
# Importando Keras y Tensorflow
from keras.models import Sequential
from keras.layers import Dense
from keras.initializers import RandomUniform
# Inicializando la Red Neuronal
neural_network = Sequential()
# kernel_initializer Define la forma como se asignará los Pesos iniciales Wi
initial_weights = RandomUniform(minval = -0.5, maxval = 0.5)
# Agregado la Capa de entrada y la primera capa oculta
# 10 Neuronas en la capa de entrada y 8 Neuronas en la primera capa oculta
neural_network.add(Dense(units = 20, kernel_initializer = initial_weights, activation = 'relu', input_dim = 23))
# Agregando capa oculta
neural_network.add(Dense(units = 10, kernel_initializer = initial_weights, activation = 'relu'))
# Agregando capa oculta
neural_network.add(Dense(units = 5, kernel_initializer = initial_weights, activation = 'relu'))
# Agregando capa de salida
neural_network.add(Dense(units = 1, kernel_initializer = initial_weights, activation = 'sigmoid'))
Imprimimos la aquitectura de la Red:
neural_network.summary()
Model: "sequential_1"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
dense_1 (Dense) (None, 20) 480
_________________________________________________________________
dense_2 (Dense) (None, 10) 210
_________________________________________________________________
dense_3 (Dense) (None, 5) 55
_________________________________________________________________
dense_4 (Dense) (None, 1) 6
=================================================================
Total params: 751
Trainable params: 751
Non-trainable params: 0
_________________________________________________________________
Entrenamos el modelo en 100 épocas:
# Compilando la Red Neuronal
# optimizer: Algoritmo de optimización | binary_crossentropy = 2 Classes
# loss: error
neural_network.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])
# Entrenamiento
neural_network.fit(X_train, y_train, batch_size = 32, epochs = 100)
Epoch 1/100
23999/23999 [==============================] - 3s 133us/step - loss: 0.4300 - acc: 0.8207
Epoch 2/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4291 - acc: 0.8207
Epoch 3/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4281 - acc: 0.8207
Epoch 4/100
23999/23999 [==============================] - 3s 118us/step - loss: 0.4276 - acc: 0.8212
Epoch 5/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4271 - acc: 0.8210
Epoch 6/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4265 - acc: 0.8208
Epoch 7/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4254 - acc: 0.8213
Epoch 8/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4251 - acc: 0.8212
Epoch 9/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4251 - acc: 0.8216
Epoch 10/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4241 - acc: 0.8215
Epoch 11/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4241 - acc: 0.8213
Epoch 12/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4237 - acc: 0.8228
Epoch 13/100
23999/23999 [==============================] - 3s 117us/step - loss: 0.4235 - acc: 0.8221
Epoch 14/100
23999/23999 [==============================] - 3s 119us/step - loss: 0.4229 - acc: 0.8236
Epoch 15/100
23999/23999 [==============================] - 3s 119us/step - loss: 0.4226 - acc: 0.8220
Epoch 16/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4222 - acc: 0.8229
Epoch 17/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4221 - acc: 0.8220
Epoch 18/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4218 - acc: 0.8229
Epoch 19/100
23999/23999 [==============================] - 3s 119us/step - loss: 0.4215 - acc: 0.8231
Epoch 20/100
23999/23999 [==============================] - 3s 118us/step - loss: 0.4212 - acc: 0.8217
Epoch 21/100
23999/23999 [==============================] - 3s 117us/step - loss: 0.4211 - acc: 0.8229
Epoch 22/100
23999/23999 [==============================] - 3s 117us/step - loss: 0.4209 - acc: 0.8231
Epoch 23/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4205 - acc: 0.8234
Epoch 24/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4206 - acc: 0.8231
Epoch 25/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4203 - acc: 0.8229
Epoch 26/100
23999/23999 [==============================] - 3s 118us/step - loss: 0.4199 - acc: 0.8224
Epoch 27/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4197 - acc: 0.8225
Epoch 28/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4197 - acc: 0.8222
Epoch 29/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4199 - acc: 0.8237
Epoch 30/100
23999/23999 [==============================] - 3s 117us/step - loss: 0.4192 - acc: 0.8234
Epoch 31/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4190 - acc: 0.8231
Epoch 32/100
23999/23999 [==============================] - 3s 118us/step - loss: 0.4187 - acc: 0.8233
Epoch 33/100
23999/23999 [==============================] - 3s 113us/step - loss: 0.4185 - acc: 0.8227
Epoch 34/100
23999/23999 [==============================] - 3s 113us/step - loss: 0.4182 - acc: 0.8234
Epoch 35/100
23999/23999 [==============================] - 3s 120us/step - loss: 0.4181 - acc: 0.8226
Epoch 36/100
23999/23999 [==============================] - 3s 121us/step - loss: 0.4180 - acc: 0.8228
Epoch 37/100
23999/23999 [==============================] - 3s 121us/step - loss: 0.4180 - acc: 0.8229
Epoch 38/100
23999/23999 [==============================] - 3s 121us/step - loss: 0.4174 - acc: 0.8234
Epoch 39/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4176 - acc: 0.8230
Epoch 40/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4173 - acc: 0.8232
Epoch 41/100
23999/23999 [==============================] - 3s 119us/step - loss: 0.4170 - acc: 0.8241
Epoch 42/100
23999/23999 [==============================] - 3s 122us/step - loss: 0.4172 - acc: 0.8236
Epoch 43/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4167 - acc: 0.8235
Epoch 44/100
23999/23999 [==============================] - 3s 111us/step - loss: 0.4165 - acc: 0.8230
Epoch 45/100
23999/23999 [==============================] - 3s 111us/step - loss: 0.4163 - acc: 0.8245
Epoch 46/100
23999/23999 [==============================] - 3s 117us/step - loss: 0.4161 - acc: 0.8245
Epoch 47/100
23999/23999 [==============================] - 3s 117us/step - loss: 0.4159 - acc: 0.8244
Epoch 48/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4161 - acc: 0.8236
Epoch 49/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4157 - acc: 0.8250
Epoch 50/100
23999/23999 [==============================] - 3s 117us/step - loss: 0.4161 - acc: 0.8237
Epoch 51/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4158 - acc: 0.8232
Epoch 52/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4154 - acc: 0.8247
Epoch 53/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4159 - acc: 0.8237
Epoch 54/100
23999/23999 [==============================] - 3s 113us/step - loss: 0.4154 - acc: 0.8249
Epoch 55/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4150 - acc: 0.8251
Epoch 56/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4153 - acc: 0.8240
Epoch 57/100
23999/23999 [==============================] - 3s 113us/step - loss: 0.4153 - acc: 0.8250
Epoch 58/100
23999/23999 [==============================] - 3s 113us/step - loss: 0.4151 - acc: 0.8252
Epoch 59/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4145 - acc: 0.8245
Epoch 60/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4147 - acc: 0.8245
Epoch 61/100
23999/23999 [==============================] - 3s 113us/step - loss: 0.4146 - acc: 0.8253
Epoch 62/100
23999/23999 [==============================] - 3s 117us/step - loss: 0.4146 - acc: 0.8251
Epoch 63/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4140 - acc: 0.8251
Epoch 64/100
23999/23999 [==============================] - 3s 112us/step - loss: 0.4142 - acc: 0.8242
Epoch 65/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4142 - acc: 0.8246
Epoch 66/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4137 - acc: 0.8247
Epoch 67/100
23999/23999 [==============================] - 3s 112us/step - loss: 0.4141 - acc: 0.8249
Epoch 68/100
23999/23999 [==============================] - 3s 117us/step - loss: 0.4140 - acc: 0.8260
Epoch 69/100
23999/23999 [==============================] - 3s 112us/step - loss: 0.4137 - acc: 0.8248
Epoch 70/100
23999/23999 [==============================] - 3s 112us/step - loss: 0.4139 - acc: 0.8252
Epoch 71/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4131 - acc: 0.8251
Epoch 72/100
23999/23999 [==============================] - 3s 112us/step - loss: 0.4133 - acc: 0.8253
Epoch 73/100
23999/23999 [==============================] - 3s 112us/step - loss: 0.4135 - acc: 0.8254
Epoch 74/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4128 - acc: 0.8245
Epoch 75/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4134 - acc: 0.8254
Epoch 76/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4132 - acc: 0.8245
Epoch 77/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4129 - acc: 0.8249
Epoch 78/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4129 - acc: 0.8255
Epoch 79/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4130 - acc: 0.8252
Epoch 80/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4130 - acc: 0.8253
Epoch 81/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4125 - acc: 0.8259
Epoch 82/100
23999/23999 [==============================] - 3s 116us/step - loss: 0.4123 - acc: 0.8257
Epoch 83/100
23999/23999 [==============================] - 3s 115us/step - loss: 0.4126 - acc: 0.8252
Epoch 84/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4124 - acc: 0.8255
Epoch 85/100
23999/23999 [==============================] - 3s 111us/step - loss: 0.4120 - acc: 0.8254
Epoch 86/100
23999/23999 [==============================] - 3s 113us/step - loss: 0.4123 - acc: 0.8256
Epoch 87/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4124 - acc: 0.8254
Epoch 88/100
23999/23999 [==============================] - 3s 112us/step - loss: 0.4117 - acc: 0.8263
Epoch 89/100
23999/23999 [==============================] - 3s 114us/step - loss: 0.4118 - acc: 0.8252
Epoch 90/100
23999/23999 [==============================] - 3s 118us/step - loss: 0.4121 - acc: 0.8253
Epoch 91/100
23999/23999 [==============================] - 3s 112us/step - loss: 0.4118 - acc: 0.8256
Epoch 92/100
23999/23999 [==============================] - 3s 113us/step - loss: 0.4115 - acc: 0.8264
Epoch 93/100
23999/23999 [==============================] - 3s 112us/step - loss: 0.4113 - acc: 0.8255
Epoch 94/100
23999/23999 [==============================] - 3s 113us/step - loss: 0.4120 - acc: 0.8256
Epoch 95/100
23999/23999 [==============================] - 3s 112us/step - loss: 0.4113 - acc: 0.8260
Epoch 96/100
23999/23999 [==============================] - 3s 111us/step - loss: 0.4114 - acc: 0.8261
Epoch 97/100
23999/23999 [==============================] - 3s 111us/step - loss: 0.4114 - acc: 0.8255
Epoch 98/100
23999/23999 [==============================] - 3s 113us/step - loss: 0.4113 - acc: 0.8267
Epoch 99/100
23999/23999 [==============================] - 3s 112us/step - loss: 0.4112 - acc: 0.8254
Epoch 100/100
23999/23999 [==============================] - 3s 111us/step - loss: 0.4112 - acc: 0.8253
<keras.callbacks.History at 0x7fd8588fa2e8>Obtenemos un accuracy de 82.5%
Realizamos las predicciones con los datos de Test y generamos la matriz de confusión:
# Haciendo predicción de los resultados del Test
y_pred = neural_network.predict(X_test)
y_pred_norm = (y_pred > 0.5)
y_pred_norm = y_pred_norm.astype(int)
y_test = y_test.astype(int)
plot_confusion_matrix(y_test, y_pred_norm, normalize=False,title="Matriz de Confusión: Incumplimiento de Pagos de Tarjetas de Credito")
Matriz de Confusión sin Normalizar
[[4458 235]
[ 867 440]]
ES 
EN