In the function "train_model()" the for-loop "for asset_name in asset_name_all:" finishes before the model-optimization. So basically the optimized LSTM neural network is always based on the last asset. This can't be on purpose, can it?

best
eduard

def train_model(data):
"""
train the LSTM network
"""

asset_name_all = data.coords['asset'].values

features_all = get_features(data)
target_all = get_target_classes(data)

model = get_model()

for asset_name in asset_name_all:
    
    # drop missing values:
    target_cur = target_all.sel(asset=asset_name).dropna('time', 'any')
    features_cur = features_all.sel(asset=asset_name).dropna('time', 'any')

    # align features and targets:
    target_for_learn_df, feature_for_learn_df = xr.align(target_cur, features_cur, join='inner')
    
criterion = nn.MSELoss() # define loss function

optimiser = optim.LBFGS(model.parameters(), lr=0.08) # we use an LBFGS solver as optimiser
epochs = 1 #how many epochs 
for i in range(epochs):
        def closure(): # reevaluates the model and returns the loss (forward pass)
            optimiser.zero_grad()
            
            #input tensor
            in_ = torch.zeros(1,len(feature_for_learn_df.values))
            in_[0,:]=torch.tensor(np.array(feature_for_learn_df.values))
            
            #output
            out = model(in_)
            
            #target tensor
            target = torch.zeros(1,len(target_for_learn_df.values))
            target[0,:]=torch.tensor(np.array(target_for_learn_df.values))
            
            #evaluate loss
            loss = criterion(out, target)
            loss.backward()
            
            return loss
        optimiser.step(closure) #updates weights
        
return model

The loop "for asset_name in asset_name_all:" creates a model for each asset, but the individual models are never saved. At the end, the model for the last asset is returned and all the predictions are created based on this last model (asset 'XRP').

def train_model(data):
"""Create and train the models working on an asset-by-asset basis."""

asset_name_all = data.coords['asset'].values

data = data.sel(time=slice('2013-05-01',None)) # cut the noisy data head before 2013-05-01

features_all = get_features(data)
target_all = get_target_classes(data)

model = create_model()

for asset_name in asset_name_all:
    
    # drop missing values:
    target_cur = target_all.sel(asset=asset_name).dropna('time', 'any')
    features_cur = features_all.sel(asset=asset_name).dropna('time', 'any')

    # align features and targets:
    target_for_learn_df, feature_for_learn_df = xr.align(target_cur, features_cur, join='inner')

    if len(features_cur.time) < 10:
        # not enough points for training
        continue

    
    
    try:
        model.fit(feature_for_learn_df.values, target_for_learn_df)
    except KeyboardInterrupt as e:
        raise e
    except:
        logging.exception('model training failed')

return model

Thanks for your support, I really appreciate, together we are better!

This is how I solved the step four issue:

Kind Regards
Eddie

The last day shown ist currenty August 2th.
Best
Eddie

Hi guys, local development doesn't work.
I followed all the steps for installing local development.

I run "conda activate qntdev"
Then I create a Jupyter Notebook in this folder:
"C:\Users\baiti\anaconda3\envs\qntdev"

But when I run this Notebook I get always the error: "No module named 'qnt'"

Can you help me out?

Best
Eddie

Dear Quantiacs Team,
I have some questions on the Q19 Contest.

Are the rules and the stocks universe exactly the same as in the Q18 contest? If yes, why?

Can I use the codes from the Q18 contest for the search of investment strategies or are code adjustments necessary?

Best
Eddie

I have numerous submissions with "calculation time exceeded".
Just to be on the safe side, do I have to actively resubmit those strategies or is it not necessary?

@support Thanks a lot!

Dear Quantiacs Team, why is the "is_liquid" dataset flawed?
Especially in the first years, the sum of liquid assets just drops to zero (for one day!) several times. Some investment strategies, I backtested, are screwed up by this issue. These strategies have a certain time dependency and are just reset to zero from one day to another.

Specifically, I am interested in the exact time period of the first train period.

In this example, the train period equals 4x365 days and the backtest starts at 2006-01-01.
Does is it mean that the first train period spans from 2006-01-01 to 2009-12-31?
So the first four years of the backtest periods are actually in-sample predictions?

When the model is retrained every 5x365th day, is the train period always 4x365 days long (rolling window)? Is it possible to implement the expanding window approach for model retraining?

weights = qnbt.backtest_ml(
train=train_model,
predict=predict,
train_period=4x365, # the data length for training in calendar days
retrain_interval=5x365, # how often we have to retrain models (calendar days)
retrain_interval_after_submit=50, # how often retrain models after submission during evaluation (calendar days)
predict_each_day=True, # Is it necessary to call prediction for every day during backtesting?
competition_type='stocks_nasdaq100', # competition type
lookback_period=365, # how many calendar days are needed by the predict function to generate the output
start_date='2006-01-01', # backtest start date
build_plots=True # do you need the chart?
)

I have a different issue with testing and submitting.

I tested my strategy (Ridge Regression, random state = 18) with multi-pass backtesting using qnbt.backtest_ml. The Sharpe Ratio is 1.13, there is no forward looking, because the Sharpe Ratio is the same when predict_each_day is set to True.

Now comes the problem: I submitted this strategy (written in multi-pass backtesting format) and this strategy always gets rejected, because the Sharpe Ratio is drastically smaller 1 (exactly
0.49).

Why is there such a big difference in the Sharpe Ratio?

weights = qnbt.backtest_ml(
train=train_model,
predict=predict,
train_period=4x365, # the data length for training in calendar days
retrain_interval=5x365, # how often we have to retrain models (calendar days)
retrain_interval_after_submit=50, # how often retrain models after submission during evaluation (calendar days)
predict_each_day=True, # Is it necessary to call prediction for every day during backtesting?
# Set it to true if you suspect that get_features is looking forward.
competition_type='stocks_nasdaq100', # competition type
lookback_period=365, # how many calendar days are needed by the predict function to generate the output
start_date='2006-01-01', # backtest start date
build_plots=True # do you need the chart?
)

The last day shown ist currenty August 2th.
Best
Eddie

@vyacheslav_b said in Strategy Optimization in local development environment is not working:

result = qnop.optimize_strategy(
data,
single_pass_strategy,
qnop.full_range_args_generator(
wma_period=range(10, 20, 5), # min, max, step
roc_period=range(5, 15, 5) # min, max, step
),
workers=1 # you can set more workers when you run this code on your local PC to speed it up
)

I tried other ranges, it still doesn't work, any ideas?

Dear Support Team, I can not replicate the optimization example in the local development environment. My code just got stuck in the first iteration, I waited 8 hours and absolutely nothing happened. Do you have any idea?
Best
Eddie

Hello friends,

how can I update my local development to the Q18 contest? Loading nasdaq data doesn't work.

Best
Eddie