accu | Mohanson 's Blog

null === undefined is false in Javascript.

I'm not quite sure what's going on in the code, but the following would work fine:

You are incrementing N in place via +=.

Your function fails to ensure it is initialized at zero. Rcpp tends to do that by default (as I think it is prudent) -- but this can be suppressed for speed if you know you are doing.

A minimally repaired version of your code (with the correct header, and a call to .fill(0)) follows.

The following implementation is based on a min-heap (std::priority_queue in C++), that memorizes the "best" future candidates.

One important point is to treat the basic solutions k *(k+1) differently. As it is likely that these numbers are in the majority, this allows to greatly reduce the size of the heap.

At each given time, we either decide to use a k(k+1)number, or to use the current top value of the min-heap. Each used value led to insertion of a new candidate in the min-heap.

Another aspect is to only insert in the heap the values less then the estimated max value, n(n+1).

The complexity is estimated to be O(n log M), where M is the mean size of the heap.

For n = 10^6, the programme measures that the maximum size of the heap is equal to 9998, much less than n.

On my PC, I get the result for n = 10^6 in 11 ms. Result: 977410038240

Here is the C++ code.

This code memorizes all the sequence, mainly for debugging. In practice, if we only need the nth value, such memorization can be avoided. The measurement of the maximum heap (useful for debugging) size can be removed too, if efficiency is still a concern.

1. In the root directory, these should be a file called mylib.opam with content similar to this one. Then, the (public_name mylib) s-expression should be added to library in mylib/src/dune. Finally, making sure to be in the root of the project, the library shall be built with dune build and installed with opam install ..
   For actually using the library in another project, even in a different dune workspace, all it's needed is to add (libraries mylib) to the dune file of project that will use the library.
2. The (private_modules priv) s-expression should be added to library in mylib/src/dune. Then, mylib.ml needs to be created with the following content: module Mymodule = Mymodule: this will make sure that only Mymodule will be exposed under the package Mylib.

As @kopecs pointed out, this is caused by case xs of when you want case xs_left of.

Here is a cleaned up (whitespace, naming) version of your function:

1. No - because valuation loss not increasing
2. Your plots look fine. It is expected that the training process goes slower
3. Yes, but it doesn't make sense. If you train any model for infinity - its performance will permanently improved - e.g. you can get 89.5% accuracy (which is better than 89.48%) if you train it for year.
4. Try decaying learning rate with different schedules

The LSTM requires two hidden states, not one. So instead of

Force the values of T to be string | number | symbol: