import pandas as pd import numpy as np import seaborn as sns import matplotlib.pyplot as plt import warnings warnings.filterwarnings('ignore')

df = pd.read_csv('./inventario.csv') df.head()

df.dtypes

df.drop('Order', axis=1, inplace=True) df.head()

df.describe()

# Tipo de datos de la base df_train.dtypes

# Categoricas df_train.loc[df_train['MarketingType']=='S','MarketingType'] = 0 df_train.loc[df_train['MarketingType']=='D','MarketingType'] = 1 df_test.loc[df_test['MarketingType']=='S','MarketingType'] = 0 df_test.loc[df_test['MarketingType']=='D','MarketingType'] = 1 # Tipos a entero df_train['MarketingType'] = df_train['MarketingType'].astype('int') df_test['MarketingType'] = df_test['MarketingType'].astype('int') df_train.hist(figsize=(20,15), bins=100) plt.show()

# Correlaciones corr = df_train.corr() plt.figure(figsize=(10, 8)) sns.heatmap(corr, vmin=-1,vmax=1) plt.show()

plt.figure(figsize=(10, 8)) sns.heatmap(np.abs(corr), vmin=0,vmax=1) plt.show()

#Se separa entrada y salida X = df_train.drop(columns=['SoldFlag','SoldCount']) # sacamos los posibles target y = df_train[['SoldFlag']] y.hist() plt.show()

#!pip install imblearn from imblearn.over_sampling import RandomOverSampler ros = RandomOverSampler(random_state=0) X_resampled, y_resampled = ros.fit_resample(X, y) X = pd.DataFrame(columns=X.columns, data=X_resampled) y = pd.DataFrame(columns=y.columns, data=y_resampled) y.hist() plt.plot()

from sklearn.model_selection import train_test_split X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.2, random_state=42)

from sklearn import preprocessing scaler = preprocessing.RobustScaler().fit(X_train.values) normalized_train_X = scaler.transform(X_train.values) normalized_val_X = scaler.transform(X_val.values)

from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense, Dropout def clasificador(n_var, drop_out): model = Sequential() model.add(Dense(10, input_dim=n_var, activation='relu')) model.add(Dropout(drop_out)) model.add(Dense(10, input_dim=n_var, activation='relu')) model.add(Dropout(drop_out)) model.add(Dense(1, activation='sigmoid')) return model n_epochs = 100 n_batch = 100 n_var = normalized_train_X.shape[1] #Hay que fundamentar los hiperparámetros??

modelo_cd = clasificador(n_var, 0.1) modelo_cd.summary()

modelo_cd.compile(loss='binary_crossentropy', optimizer = 'adam', metrics=['acc', 'Precision', 'Recall']) history_cd = modelo_cd.fit(normalized_train_X, y_train, validation_split=0.2, epochs = n_epochs, verbose=1, batch_size=n_batch)

y_pred = modelo_cd.predict(normalized_val_X).round()

#!pip install mlxtend from sklearn.metrics import confusion_matrix from mlxtend.plotting import plot_confusion_matrix fig, ax = plot_confusion_matrix(conf_mat=confusion_matrix(y_val, y_pred),figsize=(6,6), show_absolute=True, show_normed=False, colorbar=False) plt.title('Matriz de confusión') plt.show()

from sklearn.metrics import classification_report print(classification_report(y_val, y_pred))

# ajustamos los datos activos X_test = df_test.drop(columns=['SoldFlag','SoldCount']) # no sirven para predecir o no están contenidos en la data activa normalized_X_test = scaler.transform(X_test.values) # predicción de clases para test y_test_pred = modelo_cd.predict(normalized_X_test).round() # probabilidades de clases para test y_test_pred_prob = modelo_cd.predict(normalized_X_test).round(3)

df_test.loc[:,['SaleProb']] = y_test_pred_prob # asociamos probabilidades a la base con datos activos rankings = df_test['SaleProb'].rank(method='min',ascending=False).values df_test.loc[:,['Rank']] = rankings # rank según el orden de estas probabilidades

# Ranking df_ranks = df_test.sort_values(by=["SaleProb"], ascending=False)[['SaleProb','Rank']] display(df_ranks)

df_train.loc[:,['RankPercentile']] = df_train['SoldCount'].rank(method='first') df_train.loc[:,['decile']] = pd.qcut(df_train['RankPercentile'].values, 10).codes X = df_train.drop(columns=['SoldFlag','SoldCount','RankPercentile','decile']) # no sirven para predecir o no están contenidos en la data activa y_cant = pd.get_dummies(df_train['decile'])

y_pred_2 = modelo_cd_2.predict(normalized_val_X_2) print('predicciones para una observación de validación:', np.round(y_pred_2[0],3))