I would like to perform cross validation on minutely data.
I tried using the TimeSeriesSplit method in Scikit learn. But it creates folds based on index.
Hence what happens is the Train fold might contain data from
'2017-01-01 00:00:00' to '2019-01-01 10:00:00', and the
Test fold will have data from '2019-01-01 10:01:00' to '2019-01-02 23:59:00'.
As you can see this will not be an accurate way to perform cross validation because the hour 10 is included in train and test folds, hence it will lead to misleading results.
What i would like to achieve is perform TimeSeriesSplit with grouping
For example -
train_fold_1 - '2017-01-01 00:00:00' to '2019-01-01 23:59:00'
test_fold_1 - '2019-01-02 00:00:00' to '2019-01-02 23:59:00'
train_fold_2 - '2017-01-01 00:00:00' to '2019-01-02 23:59:00'
test_fold_2 - '2019-01-03 00:00:00' to '2019-01-03 23:59:00'
I have minutely data and currently using XGBoost for predicting every minute of tomorrow.
PFA - a screenshot of pandas Dataframe
In both the below cases past 400 days were used to perform cross validation
For example - If we want to predict from 2019-01-01 00:00:00 to 2019-01-01 23:59:00, we will use past 400 days till 2018-12-31 23:59:00 for performing cross validation.
The first value in below picture is MSE and the second one is standard deviation, calculated using TimeSeriesSplit from Scikit learn.
cross validation score 1-January-2019 using TimeSeriesSplit
cross validation score 1-february-2019 using TimeSeriesSplit
As you can notice even when the same amount of data has been used , we still observe a huge change in the MSE and standard deviation when the date range has been changed .
Related
I am trying to create a model that predicts if it will rain in the next 5 days (multi-step) or not, so I dont need the precipitation value, just a "yes" or "no". I've been testing with some different tools/algorithms and I guess the big challenge here is dealing with the zero skewed data.
The dataset consists of hourly data that has columns such as precipitation, temperature, pressure, wind speed, humidity. It has around 1 milion rows. There is no requisite to use a multivariate approach.
Rain occurs mostly on months 1,2,3,11 and 12.
So I tried using a univariate LSTM on the data, and with hourly sample I had the best results. I used the following architecture:
model=Sequential()
model.add(LSTM(150,return_sequences=True,input_shape=(1,look_back)))
model.add(LSTM(50,return_sequences=True))
model.add(LSTM(50))
model.add(Dense(1))
model.compile(loss='mean_squared_error', optimizer='adam')
history = model.fit(trainX, trainY, epochs=15, batch_size=4096, validation_data=(testX, testY), shuffle=False)
I'm using a lookback value of 24*60, which should mean 2 months.
Train/Validation Loss:
https://i.stack.imgur.com/CjDbR.png
Final result:
https://i.stack.imgur.com/p6SnD.png
So I read that this train/validation loss means the model is underfitting, is it? What could I do to prevent this?
Before using LSTM I tried using Prophet, which rendered really bad results and tried used autoarima, but it couldn't handle a yearly seasonality (365 days).
In case of underfitting what you can do is icreasing the learning rate, increasing training duration and number of training data.
It is also worth having some external metric such as the F1 score because loss isn't a good metrics for human evaluation.
Just looking at your example I would start with experimenting a bit with the loss function, it seems like your data is binary so it would be wiser to use a binary loss instead of a regression loss
I have a dataset (timeseries from 2010 to 2019 rainfall data from various districts near vellore). When I ran the ADF(Augmented Dickey-Fuller Test) i got my dataset to be Stationary! meaning no seasonality!
My question is that am I doing something wrong? because normally rainfall occurs more in particular months(rainy season ofc) So shouldn't there be seasonality in my dataset?
ADF Result
Results of Dickey-Fuller Test:
Test Statistic -1.770941e+01
p-value 3.507811e-30
#Lags Used 7.000000e+00
Number of Observations Used 3.644000e+03
Critical Value (1%) -3.432146e+00
Critical Value (5%) -2.862333e+00
Critical Value (10%) -2.567192e+00
According to this result my test statistic of -17.7 is very small compared to critical values -2.56(10%) Hence this means my data is already stationary!.
Dataset contains daily data so there are a lot of 0's too, does this affect the seasonality?
Thank you!
Check the same with KPPS test with checking the seasonal Trend
kpps(df,regression='ct')
The parameter regression = 'ct' will check over the seasonal trend
In a database there are time-series data with records:
device - timestamp - temperature - min limit - max limit
device - timestamp - temperature - min limit - max limit
device - timestamp - temperature - min limit - max limit
...
For every device there are 4 hours of time series data (with an interval of 5 minutes) before an alarm was raised and 4 hours of time series data (again with an interval of 5 minutes) that didn't raise any alarm. This graph describes better the representation of the data, for every device:
I need to use RNN class in python for alarm prediction. We define alarm when the temperature goes below the min limit or above the max limit.
After reading the official documentation from tensorflow here, i'm having troubles understanding how to set the input to the model. Should i normalise the data beforehand or something and if yes how?
Also reading the answers here didn't help me as well to have a clear view on how to transform my data into an acceptable format for the RNN model.
Any help on how the X and Y in model.fit should look like for my case?
If you see any other issue regarding this problem feel free to comment it.
PS. I have already setup python in docker with tensorflow, keras etc. in case this information helps.
You can begin with a snippet that you mention in the question.
Any help on how the X and Y in model.fit should look like for my case?
X should be a numpy matrix of shape [num samples, sequence length, D], where D is a number of values per timestamp. I suppose D=1 in your case, because you only pass temperature value.
y should be a vector of target values (as in the snippet). Either binary (alarm/not_alarm), or continuous (e.g. max temperature deviation). In the latter case you'd need to change sigmoid activation for something else.
Should i normalise the data beforehand
Yes, it's essential to preprocess your raw data. I see 2 crucial things to do here:
Normalise temperature values with min-max or standardization (wiki, sklearn preprocessing). Plus, I'd add a bit of smoothing.
Drop some fraction of last timestamps from all of the time-series to avoid information leak.
Finally, I'd say that this task is more complex than it seems to be. You might want to either find a good starter tutorial on time-series classification, or a course on machine learning in general. I believe you can find a better method than RNN.
Yes you should normalize your data. I would look at differencing by every day. Aka difference interval is 24hours / 5 minutes. You can also try and yearly difference but that depends on your choice in window size(remember RNNs dont do well with large windows). You may possibly want to use a log-transformation like the above user said but also this seems to be somewhat stationary so I could also see that not being needed.
For your model.fit, you are technically training the equivelant of a language model, where you predict the next output. SO your inputs will be the preciding x values and preceding normalized y values of whatever window size you choose, and your target value will be the normalized output at a given time step t. Just so you know a 1-D Conv Net is good for classification but good call on the RNN because of the temporal aspect of temperature spikes.
Once you have trained a model on the x values and normalized y values and can tell that it is actually learning (converging) then you can actually use the model.predict with the preciding x values and preciding normalized y values. Take the output and un-normalize it to get an actual temperature value or just keep the normalized value and feed it back into the model to get the time+2 prediction
I'm very new to deep learning models, and trying to train a multiple time series model using LSTM with Keras Sequential. There are 25 observations per year for 50 years = 1250 samples, so not sure if this is even possible to use LSTM for such small data. However, I have thousands of feature variables, not including time lags. I'm trying to predict a sequence of the next 25 time steps of data. The data is normalized between 0 and 1. My problem is that, despite trying many obvious adjustments, I cannot get the LSTM validation loss anywhere close to the training loss (overfitting dramatically, I think).
I have tried adjusting number of nodes per hidden layer (25-375), number of hidden layers (1-3), dropout (0.2-0.8), batch_size (25-375), and train/ test split (90%:10% - 50%-50%). Nothing really makes much of a difference on the validation loss/ training loss disparity.
# SPLIT INTO TRAIN AND TEST SETS
# 25 observations per year; Allocate 5 years (2014-2018) for Testing
n_test = 5 * 25
test = values[:n_test, :]
train = values[n_test:, :]
# split into input and outputs
train_X, train_y = train[:, :-25], train[:, -25:]
test_X, test_y = test[:, :-25], test[:, -25:]
# reshape input to be 3D [samples, timesteps, features]
train_X = train_X.reshape((train_X.shape[0], 5, newdf.shape[1]))
test_X = test_X.reshape((test_X.shape[0], 5, newdf.shape[1]))
print(train_X.shape, train_y.shape, test_X.shape, test_y.shape)
# design network
model = Sequential()
model.add(Masking(mask_value=-99, input_shape=(train_X.shape[1], train_X.shape[2])))
model.add(LSTM(375, return_sequences=True))
model.add(Dropout(0.8))
model.add(LSTM(125, return_sequences=True))
model.add(Dropout(0.8))
model.add(LSTM(25))
model.add(Dense(25))
model.compile(loss='mse', optimizer='adam')
# fit network
history = model.fit(train_X, train_y, epochs=20, batch_size=25, validation_data=(test_X, test_y), verbose=2, shuffle=False)
Epoch 19/20
14s - loss: 0.0512 - val_loss: 188.9568
Epoch 20/20
14s - loss: 0.0510 - val_loss: 188.9537
I assume I must be doing something obvious wrong, but can't realize it since I'm a newbie. I am hoping to either get some useful validation loss achieved (compared to training), or know that my data observations are simply not large enough for useful LSTM modeling. Any help or suggestions is much appreciated, thanks!
Overfitting
In general, if you're seeing much higher validation loss than training loss, then it's a sign that your model is overfitting - it learns "superstitions" i.e. patterns that accidentally happened to be true in your training data but don't have a basis in reality, and thus aren't true in your validation data.
It's generally a sign that you have a "too powerful" model, too many parameters that are capable of memorizing the limited amount of training data. In your particular model you're trying to learn almost a million parameters (try printing model.summary()) from a thousand datapoints - that's not reasonable, learning can extract/compress information from data, not create it out of thin air.
What's the expected result?
The first question you should ask (and answer!) before building a model is about the expected accuracy. You should have a reasonable lower bound (what's a trivial baseline? For time series prediction, e.g. linear regression might be one) and an upper bound (what could an expert human predict given the same input data and nothing else?).
Much depends on the nature of the problem. You really have to ask, is this information sufficient to get a good answer? For many real life time problems with time series prediction, the answer is no - the future state of such a system depends on many variables that can't be determined by simply looking at historical measurements - to reasonably predict the next value, you need to bring in lots of external data other than the historical prices. There's a classic quote by Tukey: "The combination of some data and an aching desire for an answer does not ensure that a reasonable answer can be extracted from a given body of data."
I have been trying to get into more details of resampling methods and implemented them on a small data set of 1000 rows. The data was split into 800 training set and 200 validation set. I used K-fold cross validation and repeated K-fold cross validation to train the KNN using the training set. Based on my understanding I have done some interpretations of the results - however, I have certain doubts about them (see questions below):
Results :
10 Fold Cv
Resampling: Cross-Validated (10 fold)
Summary of sample sizes: 720, 720, 720, 720, 720, 720, ...
Resampling results across tuning parameters:
k Accuracy Kappa
5 0.6600 0.07010791
7 0.6775 0.09432414
9 0.6800 0.07054371
Accuracy was used to select the optimal model using the largest value.
The final value used for the model was k = 9.
Repeated 10 fold with 10 repeats
Resampling results across tuning parameters:
k Accuracy Kappa
5 0.670250 0.10436607
7 0.676875 0.09288219
9 0.683125 0.08062622
Accuracy was used to select the optimal model using the largest value.
The final value used for the model was k = 9.
10 fold, 1000 repeats
k Accuracy Kappa
5 0.6680438 0.09473128
7 0.6753375 0.08810406
9 0.6831800 0.07907891
Accuracy was used to select the optimal model using the largest value.
The final value used for the model was k = 9.
10 fold with 2000 repeats
k Accuracy Kappa
5 0.6677981 0.09467347
7 0.6750369 0.08713170
9 0.6826894 0.07772184
Doubts:
While selecting the parameter, K=9 is the optimal value for highest accuracy. However, I don't understand how to take Kappa into consideration while finally choosing parameter value?
Repeat number has to be increased until we get stabilised result, the accuracy changes when the repeats are increased from 10 to 1000. However,the results are similar for 1000 repeats and 2000 repeats. Will it be right to consider the results of 1000/2000 repeats to be stabilised performance estimate?
Any thumb rule for the repeat number?
Finally,should I train the model on my complete training data (800 rows) now test the accuracy on the validation set ?
Accuracy and Kappa are just different classification performance metrics. In a nutshell, their difference is that Accuracy does not take possible class imbalance into account when calculating the metrics, while Kappa does. Therefore, with imbalanced classes, you might be better off using Kappa. With R caret you can do so via the train::metric parameter.
You could see a similar effect of slightly different performance results when running e.g. the 10CV with 10 repeats multiple times - you will just get slightly different results for those as well. Something you should look out for is the variance of classification performance over your partitions and repeats. In case you obtain a small variance you can derive that you by training on all your data, you likely obtain a model that will give you similar (hence stable) results on new data. But, in case you obtain a huge variance, you can derive that just by chance (being lucky or unlucky) you might instead obtain a model that either gives you rather good or rather bad performance on new data. BTW: the prediction performance variance is something e.g. R caret::train will give you automatically, hence I'd advice on using it.
See above: look at the variance and increase the repeats until you can e.g. repeat the whole process and obtain a similar average performance and variance of performance.
Yes, CV and resampling methods exist to give you information about how well your model will perform on new data. So, after performing CV and resampling and obtaining this information, you will usually use all your data to train a final model that you use in your e.g. application scenario (this includes both train and test partition!).