t = train.columns.difference(test.columns)
c = train.columns.difference(t).difference(dates)
X_test = test[c]
X_train = train[train.fecha_dato.isin(['2016-05-28', '2016-04-28', '2016-03-28'])][c]
y_train = train[train.fecha_dato.isin(['2016-05-28', '2016-04-28', '2016-03-28'])][t]
clf = RandomForestClassifier(n_estimators=100, criterion='entropy', n_jobs=-1, random_state=5)
import time
t0 = time.time()
clf.fit(X_train, y_train)
time.time()-t0

last_month_target = np.array(train[train.fecha_dato == '2016-05-28'][['ncodpers']+list(t)])
last_products = {}
for i in range(last_month_target.shape[0]):
    row = last_month_target[i]
    cust_id = row[0]
    used_products = set(np.where(row[1:] == 1)[0])
    last_products[cust_id] = used_products

y_pred = np.array(clf.predict_proba(X_test))[:, :, 1].T



target_cols = np.array([
    'ind_ahor_fin_ult1',
    'ind_aval_fin_ult1',
    'ind_cco_fin_ult1',
    'ind_cder_fin_ult1',
    'ind_cno_fin_ult1',
    'ind_ctju_fin_ult1',
    'ind_ctma_fin_ult1',
    'ind_ctop_fin_ult1',
    'ind_ctpp_fin_ult1',
    'ind_deco_fin_ult1',
    'ind_deme_fin_ult1',
    'ind_dela_fin_ult1',
    'ind_ecue_fin_ult1',
    'ind_fond_fin_ult1',
    'ind_hip_fin_ult1',
    'ind_plan_fin_ult1',
    'ind_pres_fin_ult1',
    'ind_reca_fin_ult1',
    'ind_tjcr_fin_ult1',
    'ind_valo_fin_ult1',
    'ind_viv_fin_ult1',
    'ind_nomina_ult1',
    'ind_nom_pens_ult1',
    'ind_recibo_ult1'
])


ncodpers = np.array(list(set(test['ncodpers'])))

y_pred = np.argsort(y_pred, axis=1)
y_pred = np.fliplr(y_pred)

preds = []
for i in range(y_pred.shape[0]):
    cust_id = ncodpers[i]
    used_products = last_products.get(cust_id, {})

    pred_top_products = []
    for product_id in y_pred[i]:
        if product_id not in used_products:
            pred_top_products.append(product_id)
        if len(pred_top_products) == 7:
            break

    preds.append(np.array(pred_top_products))

final_preds = [' '.join(list(target_cols[pred])) for pred in preds]
out = pd.DataFrame({'ncodpers': ncodpers, 'added_products': final_preds})
out.to_csv('submission.csv', index=False)