German tuning test hyperparameters

>>> param
{'max_depth': 4, 'eta': 0.05, 'objective': 'binary:logistic', 'min_child_weight': 0.000625, 'lambda': 0.1, 'scale_pos_weight': 2.3333333333333335}

It actually works here

scons .96

acc ~.713 small gaps .0616 .0866
80 steps boosting
eps 2.0
{'max_depth': 4, 'eta': 0.0001, 'objective': 'binary:logistic', 'min_child_weight': 10.0, 'lambda': 0.001, 'scale_pos_weight': 2.3333333333333335}

acc ~.7155 small gaps .06489 .0825
85 steps boosting
eps 2.0
{'max_depth': 4, 'eta': 0.0001, 'objective': 'binary:logistic', 'min_child_weight': 10.0, 'lambda': 0.001, 'scale_pos_weight': 2.3333333333333335}

(well that's nice)

.97

acc .0701 gaps .0642 .07827
125 steps boosting
eps 1.0
{'max_depth': 7, 'eta': 0.005, 'objective': 'binary:logistic', 'min_child_weight': 0.01, 'lambda': 1.0, 'scale_pos_weight': 2.3333333333333335}

acc .7131
15 steps boosting...
eps 1.0
{'max_depth': 4, 'eta': 0.001, 'objective': 'binary:logistic' ,'min_child_weight': 0.01, 'lambda': 0.1, 'scale_pos_weight': 2.3333333333333335}


0.7143 gaps .0986 .1298
165 steps boosting
eps 1.0
{'max_depth': 10, 'eta': 0.01, 'objective': 'binary:logistic', 'min_child_weight': 0.01, 'lambda': 1.0, 'scale_pos_weight': 2.3333333333333335}

.98 is harder it seems (Gaps pretty consistent at around .1 - .14

Here's acc .714
eps 1.0
{'max_depth': 10, 'eta': 0.005, 'objective': 'binary:logistic', 'min_child_weight': 0.1, 'lambda': 1.0, 'scale_pos_weight': 2.3333333333333335}

Options
Eps 2, small max_depth, high min_child_weight, low lambda

Eps ~1, max_depth 4/7/10, low min_child_weight, lambda 1 	
