import copy
import math
import random
import multiprocessing
import load_and_clean_data as lcd
import linearsvm_sqhingeloss as lsvm
import linearsvmmulticlass_sqhingeloss as lsvmmc
from sklearn import cross_validation
import utilityfunctions as util
dirname = 'linearsvmmulticlass'
path = util.makeresultsdir(dirname)
seed = 0
image_features_dir = r'https://s3.amazonaws.com/data558filessuman/DataCompetitionfiles/data'
train_features, train_labels, test_features, test_labels = lcd.load_image_data(
image_features_dir)
lambduh = 0.5
x_train, x_test, y_train, y_test = cross_validation.train_test_split(
train_features, train_labels, random_state=0, test_size=0.25)
#cross validation (Note: I am only doing the 2-class case for the CV here
#and also plot confusion matrix, log accuracy score
threads = multiprocessing.cpu_count()
folds = 10
fitclasslabel = y_train[0]
uniqueclasses = np.unique(y_train).astype(int)
classindex = [x[0] for x in np.where(uniqueclasses == fitclasslabel)]
ysub = y_train - fitclasslabel
classidxs = (ysub == 0)
from IPython.core.display import display
from datetime import datetime
import sklearn
from sklearn import linear_model as lm
from sklearn import cross_validation
from sklearn.metrics import accuracy_score, confusion_matrix
import matplotlib.pyplot as plt
import load_and_clean_data as lcd
import utilityfunctions as util
seed = 0
image_features_dir = r'https://s3.amazonaws.com/data558filessuman/DataCompetitionfiles/data'
train_features, train_labels, test_features, test_labels = lcd.load_image_data(
image_features_dir, standardize=1)
#display(labelnames)
#scikitlearn example from kernix page:
# https://www.kernix.com/blog/image-classification-with-a-pre-trained-deep-neural-network_p11
lambda_start = -10
lambda_end = 2
lambdas = np.logspace(lambda_start, lambda_end, 50, base=10)
C = 2 / (train_features.shape[0] * lambdas)
print('trying with the following hyperparameters for tuning...')
display(C)