多次安装 Keras-LSTM model

Question

Problem description:

Given a set X of 100 time series train/fit a Keras-LSTM model with 70/100 time series,generate 30 predictions and compare them to the 30 remaining time series of X.

Approach in pseudocode:

    lookBack = 3
    for i in range(0, 70):
        xData, YData = sample(timeSeries[i], lookBack)
        model.fit(xData, yData)
        
    for i in range(70, len(timeSeries)):
        xData, _ = sample(timeSeries[i], lookBack)
        model.predict(xData)

I was wondering if my approach would work as I don't know if consecutive calls to.fit() work. Does the model reset its state in each iteration?

It seems that this was already answered here but I don't know if this is still relevant as it was way back in 2016.

Furthermore is there an alternative way to do this?

My assignment focuses strongly on the tuning of the hyperparameters (total-layers,type-of-layers,neurons/layer, epoch and batch-size ) and not so much at the implementation itself. I was thinking of using fit with generators as it is indicated in the documentation but I have no clue on how to design one and even if I did I don't know if this allows the tuning of the hyperparameters

Answer 1

Yes you can call Fit method multiple times.

Yes, successive calls to fit will incrementally train the model.

Commented by @fchollet. https://github.com/keras-team/keras/issues/4446#issuecomment-261804574