def main():
data = load_data()
data = data['Baseline']
t = max(data.index)
state = interp(data, t)
g = get_graph()
show(g, state)
def mrmr(K=908,log=True,dataset='SHL_ext'):
import pandas as pd
from sklearn.feature_selection import f_regression
from sliding_window import get_FX_names
P = Params(name='mRMR_full_user1',dataset=dataset,FX_sel='all',cross_val='combined',undersampling=True)
V = ds.load_data(P)
F = pp.perform_preprocessing(P, V)
# X = np.concatenate([X0 for X0,_ in F])
# Y = np.concatenate([Y0 for _,Y0 in F])
X, Y = F[0]
P.log(f"X: {X.shape}, Y: {Y.shape}")
P.log(', '.join([ f"{int(val)}: {count}" for val,count in zip(*np.unique(Y,return_counts=True))]))
X = pd.DataFrame(X, columns = get_FX_names())
Y = pd.Series(Y.ravel())
F = pd.Series(f_regression(X, Y)[0], index = X.columns)
corr = pd.DataFrame(.00001, index = X.columns, columns = X.columns)
# initialize list of selected features and list of excluded features
selected = []
not_selected = X.columns.to_list()
# repeat K times
for i in range(K):
# compute (absolute) correlations between the last selected feature and all the (currently) excluded features
if i > 0:
last_selected = selected[-1]
corr.loc[not_selected, last_selected] = X[not_selected].corrwith(X[last_selected]).abs().clip(.00001)
# compute FCQ score for all the (currently) excluded features (this is Formula 2)
score = F.loc[not_selected] / corr.loc[not_selected, selected].mean(axis = 1).fillna(.00001)
# find best feature, add it to selected and remove it from not_selected
best = score.index[score.argmax()]
if log:
P.log(str(i+1).rjust(3,' ')+f': {best} (Score: {score[best]:.4f})')
selected.append(best)
not_selected.remove(best)
indeces = [X.columns.get_loc(c) for c in selected]
if log:
P.log(str(selected))
P.log(str(indeces))
return selected, indeces
seq_length = 20
# iterate through this many frames before training
washout = 10
# fraction of frames to reserve for testing
test = 0.1
# iterate over training data this many times
epochs = 100
print('Building model')
network = RadarNet()
print('Loading input data')
train_inputs, test_inputs = load_data('data', test_rate=test)
print('Generating target outputs')
train_outputs = generate_targets(train_inputs, pred_distance, target_location,
target_size)
test_outputs = generate_targets(test_inputs, pred_distance, target_location,
target_size)
train_inputs = train_inputs[0:len(train_outputs)]
test_inputs = test_inputs[0:len(test_outputs)]
print('Data length: train={}, test={}'.format(len(train_outputs),
len(test_outputs)))
with tf.Session() as session:
session.run(tf.global_variables_initializer())
err_plot = []
def main():
report(load_data()['Baseline'].iloc[-1])
def load_data(params,dataset):
X, _ = ds.load_data(params,dataset)
return X
def get_data(params, dataset):
X, Y = ds.load_data(params, dataset)
if Y is not None:
Y = labels_to_one_hot(params, Y)
return X, Y
def main():
import argparse
from params import DEFAULT_PARAMS as default
parser = argparse.ArgumentParser()
parser.add_argument('-clear', dest='CLEAR', action='store_true')
parser.set_defaults(CLEAR=False)
parser.add_argument('-test', dest='TEST', action='store_true')
parser.set_defaults(TEST=False)
parser.add_argument('-run', dest='RUN', action='store_true')
parser.set_defaults(RUN=False)
parser.add_argument('-eval', dest='EVAL', action='store_true')
parser.set_defaults(EVAL=False)
parser.add_argument('-eval_r', dest='BASE', action='store_true')
parser.add_argument('-eval_c', dest='BASE', action='store_true')
parser.set_defaults(BASE=False)
parser.add_argument('-search', dest='SEARCH', action='store_true')
parser.set_defaults(SEARCH=False)
parser.add_argument('-search_r', dest='SEARCH_C', action='store_true')
parser.add_argument('-search_c', dest='SEARCH_C', action='store_true')
parser.set_defaults(SEARCH_C=False)
parser.add_argument('-search_gd', dest='SEARCH_GD', action='store_true')
parser.set_defaults(SEARCH_GD=False)
parser.add_argument('-mrmr', dest='MRMR', action='store_true')
parser.set_defaults(MRMR=False)
parser.add_argument('-sklearn', dest='SKLEARN', action='store_true')
parser.set_defaults(SKLEARN=False)
parser.add_argument('-fx_num', type=int, dest='FX_NUM')
parser.set_defaults(FX_NUM=None)
parser.add_argument('-cuda', dest='CUDA', action='store_true')
parser.add_argument('-cpu', dest='CUDA', action='store_false')
parser.set_defaults(CUDA=default['CUDA'])
parser.add_argument('-data_path', type=str, dest='data_path')
parser.add_argument('-datapath', type=str, dest='data_path')
parser.set_defaults(data_path=default['data_path'])
parser.add_argument('-dataset', type=str, dest='dataset')
parser.set_defaults(dataset=default['dataset'])
parser.add_argument('-print', dest='PRINT', action='store_true')
parser.set_defaults(PRINT=default['print_epoch'])
parser.add_argument('-basic', dest='BASIC', action='store_true')
parser.add_argument('-no_basic', dest='BASIC', action='store_false')
parser.set_defaults(BASIC=default['C_basic_train'])
parser.add_argument('-max_evals', type=int, dest='max_evals')
parser.set_defaults(max_evals=None)
args = parser.parse_args()
P_args = Params(
data_path = args.data_path,
CUDA = args.CUDA,
print_epoch = args.PRINT,
C_basic_train = args.BASIC,
dataset = args.dataset,
)
P_test = P_args.copy().set_keys(
name = 'Test',
dataset = 'Test',
epochs = 10,
GD_ratio = 0.5,
save_step = 1,
runs = 1,
FX_sel = 'all',
cross_val = 'user',
sample_no = None,
undersampling = False,
oversampling = False,
CB1 = 0.02482259369526197,
CLR = 0.00033565485364740803,
C_ac_func = 'relu',
C_hidden = 92,
C_optim = 'AdamW',
DB1 = 0.1294935579262613,
DLR = 0.010144020667237321,
D_ac_func = 'leaky',
D_hidden = 317,
D_optim = 'AdamW',
GB1 = 0.023718651003136713,
GLR = 0.0005411668775518598,
G_ac_func = 'relu',
G_hidden = 140,
G_optim = 'SGD',
batch_size = 110
)
P = P_args.copy().set_keys(
name = 'eval_user1',
epochs = 300,
save_step = 1,
runs = 10,
FX_sel = 'all',
R_active = False,
cross_val = 'user1',
sample_no = None,
undersampling = False,
oversampling = False,
User_L = 1,
User_U = 2,
User_V = 3,
batch_size = 512,
FX_num = 150,
GD_ratio = 0,
RB1 = 0.02621587421913803,
RLR = 0.03451171211996072,
R_ac_func = 'leaky20',
R_hidden = 1294,
R_hidden_no = 4,
R_optim = 'SGD',
CB1 = 0.8661148142428583,
CLR = 8.299645247840653e-05,
C_ac_func = 'leaky20',
C_hidden = 1790,
C_hidden_no = 2,
C_optim = 'AdamW',
C_tau = 2.833757972503762,
DB1 = 0.04397295845368007,
DLR = 0.0243252689035249,
D_ac_func = 'leaky',
D_hidden = 113,
D_hidden_no = 6,
GB1 = 0.6201555853224091,
GLR = 0.006959406242448824,
G_ac_func = 'relu',
G_hidden = 318,
G_hidden_no = 5,
)
if args.CLEAR:
clear_cache()
if args.TEST:
param_space={'GD_ratio': hp.uniform('GD_ratio', 0, 0.9)}
P_test.set_keys(name='Test_CUDA', CUDA = True,)
evaluate(P_test)
hyperopt_GAN(P_test.copy(),param_space,eval_step=2,max_evals=5)
P_test.set_keys(name='Test_CPU', CUDA = False, D_fake_step=2)
evaluate(P_test)
hyperopt_GAN(P_test.copy(),param_space,eval_step=2,max_evals=5)
if args.RUN:
P_run = P.copy().set_keys(
runs = 5,
)
# P_run.set_keys(
# cross_val = 'user1',
# save_step = 1,
# epochs = 150,
# GD_ratio = 0,
# runs = 10,
# sample_no = None,
# undersampling = True,
# oversampling = False,
# )
for step_size in [5,None,10]:
P_run.set_keys(name='eval_user1_fpos_step_'+str(step_size), cross_val='user1', D_fake_step = step_size, sample_no = 11136, undersampling = False, oversampling = False, )
evaluate(P_run,P_run.copy().set_keys( sample_no = None, undersampling = False, oversampling = False, ),epoch_lst = [50,100,150,200,250])
# P_run.set_keys(
# name='eval_LR_C_'+str(step_size),
# D_fake_step = step_size,
# RB1 = 0.8661148142428583,
# RLR = 8.299645247840653e-05,
# R_ac_func = 'leaky20',
# R_hidden = 1790,
# R_hidden_no = 2,
# R_optim = 'AdamW',
# CB1 = 0.8661148142428583,
# CLR = 8.299645247840653e-05,
# C_ac_func = 'leaky20',
# C_hidden = 1790,
# C_hidden_no = 2,
# C_optim = 'AdamW',)
# evaluate(P_run,P_run.copy().set_keys( sample_no = None, undersampling = False, oversampling = False, ),epoch_lst=[50,100])
# P_run.set_keys(
# name='eval_LR_R_'+str(step_size),
# D_fake_step = step_size,
# RB1 = 0.02621587421913803,
# RLR = 0.03451171211996072,
# R_ac_func = 'leaky20',
# R_hidden = 1790,
# R_hidden_no = 2,
# R_optim = 'AdamW',
# CB1 = 0.02621587421913803,
# CLR = 0.03451171211996072,
# C_ac_func = 'leaky20',
# C_hidden = 1790,
# C_hidden_no = 2,
# C_optim = 'AdamW',)
# evaluate(P_run,P_run.copy().set_keys( sample_no = None, undersampling = False, oversampling = False, ))
# P_run.set_keys(
# name='eval_R',
# RB1 = 0.02621587421913803,
# RLR = 0.03451171211996072,
# R_ac_func = 'leaky20',
# R_hidden = 1294,
# R_hidden_no = 4,
# R_optim = 'SGD',
# CB1 = 0.02621587421913803,
# CLR = 0.03451171211996072,
# C_ac_func = 'leaky20',
# C_hidden = 1294,
# C_hidden_no = 4,
# C_optim = 'SGD',)
# evaluate(P_run,P_run.copy().set_keys( sample_no = None, undersampling = False, oversampling = False, ))
# P_run.set_keys(name='eval_0_GD', epochs=50, GD_ratio = 0,)
# evaluate(P_run,P_run.copy().set_keys( sample_no = None, undersampling = False, oversampling = False, ))
# P_run.set_keys(name='eval_50_GD', epochs=100, GD_ratio = 0.5,)
# evaluate(P_run,P_run.copy().set_keys( sample_no = None, undersampling = False, oversampling = False, ))
# P_run.set_keys(name='eval_100_GD', epochs=150, GD_ratio = 2/3,)
# evaluate(P_run,P_run.copy().set_keys( sample_no = None, undersampling = False, oversampling = False, ))
if args.EVAL:
V = ds.load_data(P)
for sample_no in [512,1024,4096,11136]:
P.set_keys(name='eval_combined_'+str(sample_no), cross_val='combined', sample_no = sample_no, undersampling = False, oversampling = False, )
evaluate(P,P.copy().set_keys( sample_no = None, undersampling = False, oversampling = False, ),epoch_lst = [50,100,150,200,250],V=V)
#sklearn_baseline(P,V)
for sample_no in [512,1024,4096,11136]:
P.set_keys(name='eval_user1_'+str(sample_no), cross_val='user1', sample_no = sample_no, undersampling = False, oversampling = False, )
evaluate(P,P.copy().set_keys( sample_no = None, undersampling = False, oversampling = False, ),epoch_lst = [50,100,150,200,250],V=V)
#sklearn_baseline(P,V)
# for cross_val in ['user','none']:
# for basic_train in [True,False]:
# P.set_keys(
# name = '_'.join(['eval','C',('Complete' if basic_train else 'GAN'),cross_val,'cross']),
# C_basic_train = basic_train,
# cross_val = cross_val,
# )
# evaluate(P,P.copy().set_keys( sample_no = None, undersampling = False, oversampling = False, ))
if args.BASE:
P_base1 = P.copy().set_keys( name='eval_c_normal', dataset='SHL', epochs=1500, sample_no=400, undersampling=False, oversampling=False, )
P_base2 = P.copy().set_keys( name='eval_c_extended', dataset='SHL_ext', epochs=200, sample_no=None, undersampling=True, oversampling=False, )
pytorch_baseline(P_base1,P_base1.copy().set_keys( sample_no = None, undersampling = False, oversampling = False, ))
pytorch_baseline(P_base2,P_base2.copy().set_keys( sample_no = None, undersampling = False, oversampling = False, ))
indeces = None
if args.MRMR:
selected, indeces = mrmr(dataset='SHL_ext')
if args.SKLEARN:
P_sklearn = P.copy().set_keys(
name = 'eval_sklearn',
runs = 10,
)
sklearn_baseline(P_sklearn)
if args.FX_NUM is not None:
P_fx_num = P.copy().set_keys( name='fx_num', dataset='SHL_ext', runs=8, sample_no=512, undersampling=False, oversampling=False, )
plt_FX_num(P_fx_num,max_n=args.FX_NUM,P_val=P_fx_num.copy().set_keys(sample_no=None, undersampling=False, oversampling=False,),indeces=indeces)
if args.SEARCH:
hyper_GAN_3_5(P_args)
if args.SEARCH_C:
hyper_R_1_3(P_args)
if args.SEARCH_GD:
hyper_GD_1_3(P_args)