jens/Rbm
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Wire up temperature-based sampling for poet text generation

Browse Source 
Greedy argmax decoding in RbmListener::forward always produces the
exact same character sequence and quickly falls into short repeating
loops once the 5-character lookback state revisits a prior cycle.

Add an optional temperature argument (poet f <seed> [temperature]) that
switches decoding to RnnStack::sample_one_hot, which now does proper
categorical sampling (temperature-scaled, renormalized draw) instead of
the old per-code Bernoulli approach that could leave the result as a
non-one-hot probability vector. Also seed Armadillo's RNG in main(),
since it otherwise defaults to a fixed seed and every run would sample
identically.

Add docs/RNN_ARCHITECTURE.md documenting how the RnnStack/Layer stack
implements the RNN (context chaining across layers, training/generation
data flow, and the decoding behavior above).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016K8Gu7Qejd11JbdiHZqYAs
This commit is contained in:
jens
2026-07-27 12:34:44 +02:00
co-authored by Claude Sonnet 5
parent 33a0647a51
commit 656a0252d3
4 changed files with 247 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/RNN_ARCHITECTURE.md
+188
View File
@@ -0,0 +1,188 @@
# The RNN used by `poet` (`RnnStack`)
This document explains how `RnnStack` — the model driving the `poet` character-generation
CLI — actually works. It is not an RNN in the LSTM/GRU sense: there is no shared weight matrix
applied repeatedly across time. Instead it's a **stack of independently-weighted RBMs, one per
"tap" in a fixed-size lookback window, chained together by feeding each layer's hidden
activations forward as extra "context" input to the next layer**. The recurrence is emulated by
re-running this same chain one step at a time during generation, sliding a window of the last N
characters through it.
Relevant sources: `source/RnnStack.{hpp,cpp}`, `source/Layer.{hpp,cpp}`, `source/AStack.cpp`,
`source/RnnTextHelper.{hpp,cpp}`, `source/matutils.hpp`, `source/poet.cpp`.
## The alphabet
`RnnTextHelper` encodes characters as one-hot vectors over `NUM_CODES = 37` symbols:
- index `0` = space (also the fallback for anything that isn't a letter or digit)
- indices `1..26` = `A`..`Z` (case is folded via `toupper` — **lowercase is not represented**)
- indices `27..36` = `0`..`9`
`ch2idx`/`idx2ch` do the conversion; `Matutils::char2vec`/`vec2char` wrap them as one-hot
`arma::mat` rows. Because case is discarded at encoding time, everything `poet` generates comes
out upper-case, regardless of the casing in the source text or seed string.
## Project geometry: `poet_2v_5s`
`poet.cpp` hardcodes the project name `poet_2v_5s`, loaded via `StackCreator::fromFile`. The name
encodes its shape: **2v** = 2 stacked visible character-frames per layer, **5s** = 5 layers deep
(the RBM "stack"). Concretely (`poet_2v_5s.prj`):
| layer id | visible (X×Y) | hidden | context |
|---|---|---|---|
| 0 | 37×2 = 74 | 128 | 0 |
| 1 | 37×2 = 74 | 128 | 128 |
| 2 | 37×2 = 74 | 128 | 128 |
| 3 | 37×2 = 74 | 128 | 128 |
| 4 | 37×2 = 74 | 128 | 128 |
Each `Layer` is a plain `Rbm` (`Layer.hpp:29` → `Rbm(numVisibleX*numVisibleY + numContext,
numHidden)`), so layers 1–4 are actually 202-visible-unit RBMs (74 character bits + 128 context
bits inherited from the previous layer); layer 0 has no context and is a plain 74-unit RBM.
`RnnStack::getSeqLen()` returns `numLayers()` (5) — the "depth" of the stack doubles as the length
of the character lookback window used during generation.
## Training data: character bigrams
`RnnTextHelper::createTraining(file, seq_len=2)` turns a text file into a training batch:
- keeps a rolling 2-row window (`pattern`) of one-hot vectors: `row0` = current character,
`row1` = previous character (each new character shifts the window and overwrites row 0).
- collapses runs of whitespace/unknown characters down to a single space (skips a char if both it
and the previous character map to index 0).
- for every accepted character, flattens `pattern` row-major (`as_row()`, confirmed block order:
`[current(37) | previous(37)]`) and appends it as one training row.
So the batch has one 74-wide row per character in the source text: `[char_t | char_t-1]`. This
matches every layer's 74-wide visible geometry — every layer is trained on the *same* stream of
character bigrams, just offset in time from each other (next section).
## Training: `RnnStack::train`
```cpp
padding = zeros(numLayers - 1, batch.n_cols) // 4 zero rows
batch_padded = [batch ; padding] // append padding at the end
c = zeros(N, layer[0].numContext()) // starts as an N×0 matrix (no columns)
for i in 0..numLayers-1:
layer = getLayer(i)
batch_shifted = batch_padded.rows(i, N+i-1) // batch, shifted forward by i rows
vc = join_rows(batch_shifted, c) // attach inherited context
layer.train(vc, listener) // contrastive divergence on this layer
layer.gibbs_vh(vc, c) // c ← toHiddenProbs(vc); vc ← reconstruction (vc discarded)
```
The subtlety is `c`: it starts as a zero-column matrix (matching layer 0, which has no context),
and is **mutated in place** by `gibbs_vh` on every iteration — `gibbs_vh(v_probs, h_probs)` sets
`h_probs = toHiddenProbs(v_probs)`, so after layer *i* trains, `c` becomes layer *i*'s hidden-unit
probabilities, which is exactly what layer *i+1* expects as its context input. This is the whole
recurrence mechanism: **each layer's hidden activations become the next layer's extra visible
input**, propagated forward once per training step.
`batch_shifted` is `batch_padded` windowed with an offset of `i` rows, i.e. layer `i` is trained on
training-row `j+i` while the loop index is conceptually `j`. So for a given alignment index `j`,
layer 0 sees the bigram at `j`, layer 1 sees the bigram at `j+1`, ..., layer 4 sees the bigram at
`j+4` (the most recent). In other words: **layer 0 is the "oldest" tap, layer 4 the "newest"**, and
by the time training reaches layer 4, its context input already encodes information recursively
folded in from characters `j..j+3`. This is a 5-character lookback window, unrolled across depth
(distinct weights per tap) rather than across time (shared weights) as a true RNN would.
## Generation: `RnnStack::step_forward`
`state` is a rolling `numLayers × NUM_CODES` matrix of single-character one-hot rows: row 0 is the
newest character, row `numLayers-1` the oldest of the last 5. Each call:
```cpp
state = shift(state, 1, 0); // age every row down by one
state.row(0) = v_curr; // newest known character goes in at row 0
for j in 0..numLayers-1:
k = numLayers - j - 1 // j=0 → k=4 (oldest); j=4 → k=0 (newest/current)
layer = getLayer(j)
v = [zeros(37) | state.row(k)] // "unknown target" bigram: only the lagged half is known
vc = join_rows(v, c) // attach context inherited from layer j-1
layer.gibbs_vh(vc, c) // c ← this layer's hidden probs (context for next layer)
r = vc_to_v(vc, layer.numContext())// strip context columns back off
return r // == the last layer's (j=4, k=0) reconstruction
```
Each layer is asked "given only the known previous character (`state.row(k)`) plus context
inherited from older taps, reconstruct the full bigram" — the zeroed first half is what the RBM's
one round of Gibbs sampling (`toHiddenProbs` then `toVisibleProbs`) has to fill in. Only the
**last** layer's reconstruction (built from the newest character, `k=0`, with all four older
layers' context folded in) is returned; `to_next()` (`RnnStack.cpp:121`) takes the first 37 columns
of that as the network's guess for the *next* character.
## Decoding: `poet.cpp` / `RbmListener::forward`
`forward()` takes an optional `temperature` (default `0.0`):
```cpp
auto pickNext = [temperature](arma::mat &next) {
if (temperature > 0.0) stack->sample_one_hot(next, temperature);
else stack->clamp_one_hot(next);
};
// priming: feed the seed string character-by-character to populate `state`
for each char in seed:
r = step_forward(state, char2vec(char))
next = to_next(r); pickNext(next) // result unused except to keep looping
// generation: feed each predicted character back in as the next input
for i in seqLen..len:
curr = next
r = step_forward(state, curr)
next = to_next(r); pickNext(next)
```
`temperature <= 0` keeps the original behavior: `clamp_one_hot` is a hard argmax (`index_max()` →
one-hot), fully deterministic given a trained model and a seed. `temperature > 0` instead calls
`sample_one_hot` (`RnnStack.cpp`), which now does proper categorical sampling: the 37 independent
sigmoid activations in `next` are raised to `1/temperature` and renormalized into a distribution
(`temperature < 1` sharpens it towards argmax, `temperature > 1` flattens it towards uniform), then
one code is drawn from that distribution via a cumulative-sum draw. This replaced an older
implementation that sampled each code as an independent Bernoulli and only patched the result back
into a proper one-hot vector when *more than one* bit happened to fire — leaving `srcDst` as a
non-one-hot probability vector whenever exactly zero or one bits fired.
`main()` exposes this as a third CLI argument: `poet f <seed> [temperature]`, e.g. `poet f "the "
0.7`. Sampling only produces different output across process runs because `main()` now seeds
Armadillo's RNG (`arma::arma_rng::set_seed_random()`) — Armadillo otherwise defaults to a fixed
seed, which would have made every run's "random" draws identical.
`RnnStack::to_curr()` is also computed once per generation step but its result is immediately
discarded (overwritten by `curr = next` at the top of the next iteration) — dead code kept from
development/debugging.
Two consequences worth knowing when reading `poet` output:
- **Greedy argmax decoding (`temperature <= 0`, the default) is why generated text degenerates into
short repeating loops** (e.g. `...MURPURSE AN MURPURSE AN...`) once the rolling 5-character state
re-enters a cycle it has visited before — there's no randomness to break out. Passing a
`temperature > 0` avoids this by sampling instead.
- If the seed string is shorter than `numLayers` (5) characters, the oldest rows of `state` are
still zero (not a space one-hot) when generation begins — the model briefly sees "blank" rather
than real history for the first few generated characters.
## Summary picture
```
Text file --collapse whitespace, bigram window--> training batch [char_t | char_t-1] (74-wide rows)
│
┌─────────────────────────────────────┴─────────────────────────────────────┐
│ Layer 0 (oldest tap, no context) │
│ CD train on batch │
│ context₀ = hidden_probs(batch) ──► fed into Layer 1 │
│ Layer 1 (context=128) │
│ CD train on [batch shifted +1 | context₀] │
│ context₁ = hidden_probs(...) ──► fed into Layer 2 │
│ ... │
│ Layer 4 (newest tap) │
│ CD train on [batch shifted +4 | context₃] │
└───────────────────────────────────────────────────────────────────────────────┘
Generation: same 5-layer chain, run once per output character, consuming a sliding 5-character
window of state (oldest → layer 0 ... newest → layer 4), context threaded the same way; only
layer 4's reconstruction is decoded (argmax) into the next character.
```
source/RnnStack.cpp
+26 -13
View File
@@ -92,25 +92,38 @@ void RnnStack::clamp_one_hot(arma::mat& srcDst)
srcDst[index] = 1;
}
void RnnStack::sample_one_hot(arma::mat& srcDst)
void RnnStack::sample_one_hot(arma::mat& srcDst, double temperature)
{
double k = arma::accu(srcDst);
if (k != 0)
// srcDst holds independent per-code sigmoid activations, not a normalized
// distribution. Raising to 1/temperature before renormalizing sharpens
// (temperature < 1) or flattens (temperature > 1) the resulting
// categorical distribution, then we draw one index from it directly.
arma::mat p = arma::pow(arma::clamp(srcDst, 1e-9, 1.0), 1.0 / temperature);
double sum = arma::accu(p);
if (sum > 0)
{
srcDst = srcDst / k;
p /= sum;
}
else
{
p = arma::ones(arma::size(srcDst)) / srcDst.n_elem;
}
arma::mat ps = Matutils::sample(srcDst);
if (arma::accu(ps) > 1)
double r = arma::randu(1)[0];
double cumulative = 0.0;
size_t chosen = p.n_elem - 1;
for (size_t i = 0; i < p.n_elem; i++)
{
arma::uvec q1 = find(ps > 0);
int winner = (q1.n_elem - 1) * arma::randu(1)[0];
int wix = q1(winner);
srcDst = zeros(arma::size(srcDst));
srcDst[wix] = 1;
cumulative += p[i];
if (r <= cumulative)
{
chosen = i;
break;
}
}
srcDst = arma::zeros(arma::size(srcDst));
srcDst[chosen] = 1;
}
arma::mat RnnStack::to_curr(const arma::mat& v)
source/RnnStack.hpp
+1 -1
View File
@@ -34,7 +34,7 @@ public:
arma::mat vc_to_v(const arma::mat &vc, size_t numContext) const;
arma::mat step_forward(arma::mat &state, const arma::mat &v);
void sample_one_hot(arma::mat &srcDst);
void sample_one_hot(arma::mat &srcDst, double temperature=1.0);
void clamp_one_hot(arma::mat &srcDst);
arma::mat to_next(const arma::mat &v);
arma::mat to_curr(const arma::mat &v);
source/poet.cpp
+32 -6
View File
@@ -2,6 +2,7 @@
#include <fstream>
#include <streambuf>
#include <cstdio>
#include <cstdlib>
#include <cmath>
#include <string>
#include <cassert>
@@ -47,21 +48,37 @@ class RbmListener : public Rbm::IListener
return true;
}
void forward(std::string const &start, size_t len)
// temperature <= 0 means greedy argmax decoding (the original behavior);
// temperature > 0 samples the next character from the (temperature-scaled)
// output distribution instead, which avoids the short repeating loops that
// greedy decoding tends to fall into.
void forward(std::string const &start, size_t len, double temperature=0.0)
{
RnnStack *stack = reinterpret_cast<RnnStack*>(&m_stack);
arma::mat state;
arma::mat curr;
arma::mat next;
std::string curr_str;
auto pickNext = [stack, temperature](arma::mat &next)
{
if (temperature > 0.0)
{
stack->sample_one_hot(next, temperature);
}
else
{
stack->clamp_one_hot(next);
}
};
for (int i=0; i < start.size(); i++)
{
curr = Matutils::char2vec(start.at(i), RnnTextHelper::NUM_CODES);
curr_str.append(1, Matutils::vec2char(curr));
arma::mat r = stack->step_forward(state, curr);
next = stack->to_next(r);
stack->clamp_one_hot(next);
pickNext(next);
}
cout << "Start: " << curr_str << std::endl;
@@ -71,7 +88,7 @@ class RbmListener : public Rbm::IListener
curr_str.append(1, Matutils::vec2char(curr));
arma::mat r = stack->step_forward(state, curr);
next = stack->to_next(r);
stack->clamp_one_hot(next);
pickNext(next);
curr = stack->to_curr(r);
}
cout << "Curr: " << curr_str << std::endl;
@@ -83,6 +100,10 @@ class RbmListener : public Rbm::IListener
int main(int argc, char *argv[])
{
// Armadillo's RNG otherwise defaults to a fixed seed, which would make
// sample_one_hot() produce the exact same "random" text on every run.
arma::arma_rng::set_seed_random();
enum Command {Nop, Create, Reset, Train, Forward};
Command command = Nop;
@@ -145,12 +166,17 @@ int main(int argc, char *argv[])
if (command == Command::Forward)
{
std::string start(DEFAULT_START_WORD);
double temperature = 0.0;
if (argv[2] != nullptr)
{
start = std::string(argv[2]);
}
listener.forward(start, 100);
if (argv[3] != nullptr)
{
temperature = std::atof(argv[3]);
}
listener.forward(start, 100, temperature);
}
printf("\nEnd of program\n");
return 0;