def trainprocess():
sym = txt.get()
if sym != "":
TR.process(sym, 4)
tm.showinfo("Input", "Training Successfully Finished")
else:
tm.showinfo("Input error", "Select Dataset")
def get_sql():
df = pd.read_sql('SELECT * FROM stock_table', con=db_connection)
df.drop(columns=['index'], inplace=True)
print(df.columns)
l = process(df, db_connection)
# print("ppppppppppppppppppp")
return l
nparams["training"]["loss_func"] = 'cosine'
nparams["training"]["optimizer"] = 'adam'
nparams["training"]["normalize_y"] = True
nparams["cnn"]["architecture"] = '33'
nparams["cnn"]["n_dense"] = 0
nparams["cnn"]["dropout_factor"] = 0.7
nparams["cnn"]["final_activation"] = 'linear'
nparams["dataset"]["nsamples"] = 'all'
nparams["dataset"]["dataset"] = 'MuMu-albums'
nparams["dataset"]["meta-suffix"] = meta_suffix
nparams["dataset"]["meta-suffix2"] = meta_suffix2
nparams["dataset"]["meta-suffix3"] = meta_suffix3
add_extra_params(nparams, extra_params)
params['cosine_multilabel_tri'] = copy.deepcopy(nparams)
return params[suffix]
if __name__ == '__main__':
parser = argparse.ArgumentParser(
description='Run experiment',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('suffix', default="class_bow", help='Suffix of experiment params')
parser.add_argument('meta_suffix', nargs='?', default="", help='Suffix of input matrix for experiment')
parser.add_argument('meta_suffix2', nargs='?', default="", help='Suffix of input matrix for experiment')
parser.add_argument('meta_suffix3', nargs='?', default="", help='Suffix of input matrix for experiment')
parser.add_argument('extra_params', nargs='?', default="", help='Specific extra parameters')
args = parser.parse_args()
print args.extra_params
params = get_configuration(args.suffix,args.meta_suffix,args.meta_suffix2,args.meta_suffix3,args.extra_params)
process(params)
model = models.model_from_json(f.read())
model.load_weights(wfile)
return model
if __name__ == '__main__':
model = load_model()
words, embeddings = load_data()
word_len = model.input_shape[1]
word = raw_input('Enter a word [enter nothing to exit]: ')
while word:
try:
vector = model.predict(
process(word, word_len).reshape(1, word_len))[0]
nn_words, nn_dists = rank_nearest_neighbors(
vector, words, embeddings)
print('Closest words to "%s":' % word)
s = len(words) - 4
for i, (word, dist) in enumerate(zip(nn_words[:5], nn_dists[:5]),
1):
print(' %d. "%s" [%.3f]' % (i, word, dist))
print(' ...')
for i, (word, dist) in enumerate(zip(nn_words[-5:], nn_dists[-5:]),
s):
print(' %d. "%s" [%.3f]' % (i, word, dist))
except ValueError, e:
print(e)
word = raw_input('Enter another word [enter nothing to exit]: ')
# optimizer and criterion
optimizer = optimizer_dict[optimizer_name](model.parameters(), **optimizer_options)
criterion = nn.CrossEntropyLoss()
criterion = criterion.to(device)
# initialize data
ds_trn, ds_val = imagenet_1k()
dl_trn = DataLoader(ds_trn, batch_size, shuffle=True, num_workers=4, pin_memory=True)
dl_val = DataLoader(ds_val, batch_size, num_workers=4, pin_memory=True)
# release unused objects
del ds_trn, ds_val
del parser, args, optimizer_name, optimizer_options, optimizer_dict
# obtain best accuracy from checkpoint
_, best_acc1, _ = process(dl_val, model, criterion, None, mode='eval', device=device, progress=True)
print(f" *** Starting Acc@1 {best_acc1:.4f}")
set_all_rng_seed(2019)
for epoch in range(epochs):
l, t1, t5 = process(dl_trn, model, criterion, optimizer, mode='train')
print(f" * Train Loss {l:.4f} Acc@1 {t1:.4f} Acc@5 {t5:.4f}")
l, t1, t5 = process(dl_trn, model, criterion, None, mode='eval')
print(f" ** Test Loss {l:.4f} Acc@1 {t1:.4f} Acc@5 {t5:.4f}")
is_best = t1 > best_acc1
best_acc1 = max(t1, best_acc1)
checkpoint_filename = f'checkpoint-epoch-{epoch}.pt'
torch.save(model.state_dict(), checkpoint_filename)
if is_best:
shutil.copyfile(checkpoint_filename, 'model_best.pt')