Context In machine learning there is a long path from understanding to intuition. I have created many data files of traditional electronics test pattern to see the response of different activation, loss, optimizers, and metrics in Keras. These files should give some ability to test drive your chosen type of machine learning with a very deliberate data set. I wanted something that was infinitely predicable to see how all the different settings effected the algorithms to set a base line for me to gain intuition as to how they should behave once I make more complex models. Content These files most contain single line 10,000 example patterns in sine, cosine, triangle, and others. Frequency and amplitude in some change through out the set. One has 2,500 example with 4 features of a sine wave 90 degrees out of phase from each other. The values are all between zero and one so no scaling should be necessary. CosineDecAmpFreqInc, CosineDecreasingAmp, CosineIncAmpFreqInc, CosineIncAmpFreqSlowing, ExponentialDecayTenWaves, ExponentialRiseTenWaves, FourSineWaves, LinearFall, LinearRise, Lorentz, Multitone, Pulse10Waves, Pulse10WavesInverted, RandomSamples, SinFiveWaves, SinFourtyWaves, SinTenWaves, SinTwentyWaves, 30,000 SquareFiveWaves, SquareTenWaves, SweepOneToFive, SweepOneToTwo, SweepOneToTwoPointFive, SyncPattern, TriangleFiveWaves, TriangleTenWaves.csv Acknowledgements Some were generated using Tektronics ArbExpress and modified in Excel for scale. Some I generated in c#. Inspiration How about a good Toy example of a LSTM in Keras with multivariate data and a single prediction of one of the columns. I did the 4 sine wave .csv to try this. So far the examples I have found just average all of them.