train_df = pd.read_pickle(data_path + "seq-train.pkl")
la = ["age", "gender", "demand"]
train_df["age"] = train_df["age"] - 1
train_df["gender"] = train_df["gender"] - 1
for l in la:
print(l, train_df[l].unique())
val_df = train_df.sample(n=10000, random_state=999999)
train_df = train_df.drop(val_df.index)
train_df.reset_index(drop=True, inplace=True)
val_df.reset_index(drop=True, inplace=True)
train_ds = dataset.seq_dataset(train_df, SEQ_LEN, "gender")
val_ds = dataset.seq_dataset(val_df, SEQ_LEN, "gender")
model = model.baseline_model(VOCAB_SIZE, EMBED_DIM, NUN_CLASS,
MODE).to(device)
criterion = torch.nn.CrossEntropyLoss() #.to(device)
optimizer = torch.optim.SGD(model.parameters(), weight_decay=1e-5, lr=1.0)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer,
step_size=1,
gamma=0.75)
len_train = len(train_ds)
len_val = len(val_ds)
loss_plt = []
for epoch in range(N_EPOCHS):
train_dl = DataLoader(train_ds,
batch_size=BATCH_SIZE,
shuffle=True,
collate_fn=train_ds.generate_batch)
val_dl = DataLoader(val_ds,
batch_size=BATCH_SIZE,
collate_fn=val_ds.generate_batch)
train_loss, train_acc = train_epoc(train_dl, model, criterion,
#3412772
train_df=pd.read_pickle(data_path+"seq-train.pkl")
la=["age","gender","demand"]
train_df["age"]=train_df["age"]-1
train_df["gender"]=train_df["gender"]-1
for l in la:
print(l,train_df[l].unique())
val_df=train_df.sample(frac=0.2)#,random_state=999999
train_df=train_df.drop(val_df.index)
train_df.reset_index(drop=True, inplace=True)
val_df.reset_index(drop=True, inplace=True)
train_ds=dataset.seq_dataset(train_df,SEQ_LEN,"age")
val_ds=dataset.seq_dataset(val_df,SEQ_LEN,"age")
model = model.baseline_model(VOCAB_SIZE, EMBED_DIM, NUN_CLASS,MODE).to(device)
criterion = torch.nn.CrossEntropyLoss().to(device)
optimizer = torch.optim.Adam(model.parameters(),lr=0.01)#Adam 0.01
scheduler = torch.optim.lr_scheduler.StepLR(optimizer,step_size=1,gamma=0.7)
len_train=len(train_ds)
len_val=len(val_ds)
loss_plt=[]
for epoch in range(N_EPOCHS):
train_dl = DataLoader(train_ds, batch_size=BATCH_SIZE, shuffle=True,
collate_fn=train_ds.generate_batch)
val_dl = DataLoader(val_ds, batch_size=BATCH_SIZE,
collate_fn=val_ds.generate_batch)
train_loss, train_acc = train_epoc(train_dl,model,criterion,optimizer,scheduler)
valid_loss, valid_acc = val_epoc(val_dl,model,criterion)
print("epoch",epoch)
print(f'\tLoss: {train_loss/len_train:.8f}(train)\t|\tAcc: {train_acc/len_train * 100:.3f}%(train)')
print(f'\tLoss: {valid_loss/len_val:.8f}(valid)\t|\tAcc: {valid_acc/len_val * 100:.3f}%(valid)')
loss_plt.append(train_acc/len_train)
batch_size=args.batch_size,
shuffle=True,
**kwargs)
print("Done loading datasets!")
# LSTM Model:
model = model.ReinforcedLSTM(IMAGE_SIZE, HIDDEN_NODES, HIDDEN_LAYERS,
OUTPUT_CLASSES, args.batch_size, args.cuda)
if args.cuda:
print("\n---Activating GPU Training---\n")
model.cuda()
### PRINTING AMOUNT OF PARAMETERS ###
print('Number of model parameters: {}'.format(
sum([p.data.nelement() for p in model.parameters()])))
best_accuracy = 0.0
acc_dict = {1: [], 2: [], 5: [], 10: []}
test_acc_dict = {1: [], 2: [], 5: [], 10: []}
total_accuracy = []
total_loss = []
### LOADING PREVIOUS NETWORK ###
if args.load_checkpoint:
if os.path.isfile(args.load_checkpoint):
print("=> loading checkpoint '{}'".format(args.load_checkpoint))
checkpoint = torch.load(args.load_checkpoint)
args.start_epoch = checkpoint['epoch']
episode = checkpoint['episode']