In [1]:

    

# Imports
%load_ext autoreload
%autoreload 2
%matplotlib inline
from matplotlib import pyplot as plt
import csv
import importlib
from tqdm import tqdm
from scripts import proj1_helpers, helpers
from scripts import implementation, feature_processing, k_fold, model_linear, model_logistic
import numpy as np


    

In [2]:

    

train_path = '../data/train.csv'
test_path  = '../data/test.csv'


    

In [3]:

    

# loading data
y, X, idx = proj1_helpers.load_csv_data(train_path)
y_t, X_t, ids_t = proj1_helpers.load_csv_data(test_path)


    

In [4]:

    

X_p, (x_mean, x_std) = feature_processing.process_X(X, 3)
X_t_p, _ = feature_processing.process_X(X_t, 3, (x_mean, x_std))


    

    




100%|██████████| 5/5 [00:22<00:00,  4.02s/it]
100%|██████████| 5/5 [00:51<00:00,  8.78s/it]

In [5]:

    

#Linear
model = implementation.ridge_regression
loss = model_linear.compute_accuracy_loss

lambdas = np.logspace(-13, -4, 3)

idx_min_l, rmse_all_l, lambdas_l = k_fold.cross_validation_select(X_p, y, model, loss, seed = 1,
                                                            k_fold = 3, lambdas = lambdas, do_plot = True, do_tqdm = True)


    

    




100%|██████████| 3/3 [00:31<00:00, 10.34s/it]






    

In [7]:

    

lambda_ = lambdas_l[idx_min_l]
print(lambda_)
w, l = implementation.ridge_regression(y, X_p, lambda_ = 0)


    

    




1e-13

In [8]:

    

#Logistic
y_01 = helpers.y_to_01(np.array([y]).T)
np.random.seed(1)
w0 = np.zeros((X_p.shape[1], 1))


    

In [10]:

    

model = implementation.reg_logistic_regression_newton_batch
model_args = {'initial_w': w0, 'max_iters': 10, 'gamma': 1e-5, 'debug': False,
              'batch_size': 50000}
loss = model_linear.compute_accuracy_loss

lambdas = np.logspace(-8, -4, 2)

idx_min, rmse_all, lambdas = k_fold.cross_validation_select(X_p, y_01, model, loss, kw_model = model_args, seed = 1,
                                                            k_fold = 5, lambdas = lambdas, do_plot = True, do_tqdm = False)


    

    




 80%|████████  | 8/10 [00:04<00:01,  1.92epoch/s, acc=-0.79, diff=1.67e+05, grad=9.83e+03, loss=1.39e+05, w=0.29]
 80%|████████  | 8/10 [00:03<00:00,  2.13epoch/s, acc=-0.79, diff=1.65e+05, grad=1.14e+04, loss=1.39e+05, w=0.34]
 80%|████████  | 8/10 [00:04<00:01,  1.67epoch/s, acc=-0.79, diff=1.66e+05, grad=1.15e+04, loss=1.39e+05, w=0.34]
 80%|████████  | 8/10 [00:04<00:01,  1.81epoch/s, acc=-0.79, diff=1.75e+05, grad=1.09e+04, loss=1.39e+05, w=0.31]
 80%|████████  | 8/10 [00:06<00:01,  1.27epoch/s, acc=-0.79, diff=1.7e+05, grad=1.1e+04, loss=1.39e+05, w=0.32]
 80%|████████  | 8/10 [00:06<00:01,  1.32epoch/s, acc=-0.79, diff=1.64e+05, grad=90.5, loss=1.39e+05, w=0]
 80%|████████  | 8/10 [00:06<00:01,  1.27epoch/s, acc=-0.79, diff=1.67e+05, grad=98.1, loss=1.39e+05, w=0]
 80%|████████  | 8/10 [00:05<00:01,  1.48epoch/s, acc=-0.79, diff=1.67e+05, grad=101, loss=1.39e+05, w=0]
 80%|████████  | 8/10 [00:05<00:01,  1.44epoch/s, acc=-0.79, diff=1.71e+05, grad=86.4, loss=1.39e+05, w=0]
 80%|████████  | 8/10 [00:04<00:01,  1.90epoch/s, acc=-0.79, diff=1.65e+05, grad=93.5, loss=1.39e+05, w=0]






    

In [13]:

    

lambda_ = lambdas[idx_min]
w, l = model(y_01, X_p, lambda_, **model_args)
print(lambda_)


    

    




 90%|█████████ | 9/10 [00:04<00:00,  1.86epoch/s, acc=-0.79, diff=1.72e+05, grad=1.12e+04, loss=1.73e+05, w=0.39]





    




1e-08






    

In [14]:

    

y_pred = proj1_helpers.predict_labels(w, X_t_p)


    

In [15]:

    

output_path = '../data/example_submission.csv'
proj1_helpers.create_csv_submission(ids_t, y_pred, output_path)


    

In [16]:

    

np.mean(y_pred>0)
tqdm.get_lock().locks = []


    

In [ ]: