Project Meeting (24/11/2017)

Topics Discussed

Data augmentation

Data augmentation could be useful if the amount of data is not enough. Different techniques of the data augmentation discussed were:

• Interpolation
• Using equations to recursively predict one variable while performing one step ahead predictions for more.
• Adding random noise to existing data

At this stage it is too early to determine weather or not data augmentation is required and could be useful. It remains a possibility for future experiments. Also transfer learning could be performed, training the network first on the augmented data and then retraining on the observed data.

Transfer Learning

Transfer learning could be applied using augmented/simulated data, data from other countries and data from EA countries for pretraining. The most promising option of the above would be using EA individual country data to pretrain and then train on the EA data. It is still early to decide weather or not transfer learning would be beneficial. However it does not seem like the best idea at the moment.

Multi task learning

Multitask learning is the concept of training the deep model to predict multiple macroeconomic variables at the same time. This ideas seems very promising weather it will be applied to the variables for the EA/US, the pairs of variables between the two economies or a mixture. Theoretically multi task learning should only improve performance.

Writing report

Down-top approach of writing is actually the way articles in some more general journals are done and it is useful for writing up experiments.

Design and Implementation vs Problem Analysis

Design and implementation should cover the specific choices of how an experiment was performed whereas Problem Analysis should discuss possibilities of how they could be performed. Content from one could accidentally end up in the other. Distinction not that important at this early stage - main focus is on writing something and tidy up later.

Tasks Completed

• Written Design and Implementation section about ARIMA models in report.
• Added evaluation metrics to ARIMA experiment results - ME, MAE, RMSE, ACF1.
• Added ARIMA experiment results to report appendix.
• Extra - took Structuring Machiene Learning Projects course and wrote a blog post about it.

Tasks for Next Week

• Write results from ARIMA forecasts in Results and Evaluation part of report.
• Write Problem Analysis part for ARIMA .
• Write Literature Review for ARIMA.
• Write up literature review outline in Latex and further break it down into smaller topics.
• Figure out correct formatting for project report (fonts, margins etc.) and modify document accordingly.
• Experiment with first difference univariate models .
• Experiment with decomposition univariate models.
• Create plots from hyper-parameter grid search
• Read deep learning book on RNNs.
• Read more papers.