--- /dev/null
+#!/usr/bin/env python
+
+# @XREMOTE_HOST: elk.fleuret.org
+# @XREMOTE_EXEC: python
+# @XREMOTE_PRE: source ${HOME}/misc/venv/pytorch/bin/activate
+# @XREMOTE_PRE: killall -u ${USER} -q -9 python || true
+# @XREMOTE_PRE: ln -sf ${HOME}/data/pytorch ./data
+# @XREMOTE_SEND: *.py *.sh
+
+# Any copyright is dedicated to the Public Domain.
+# https://creativecommons.org/publicdomain/zero/1.0/
+
+# Written by Francois Fleuret <francois@fleuret.org>
+
+import math, sys
+
+import torch, torchvision
+
+from torch import nn
+from torch.nn import functional as F
+
+######################################################################
+
+nb_quantization_levels = 101
+
+
+def quantize(x, xmin, xmax):
+ return (
+ ((x - xmin) / (xmax - xmin) * nb_quantization_levels)
+ .long()
+ .clamp(min=0, max=nb_quantization_levels - 1)
+ )
+
+
+def dequantize(q, xmin, xmax):
+ return q / nb_quantization_levels * (xmax - xmin) + xmin
+
+
+######################################################################
+
+
+def create_model():
+ hidden_dim = 32
+
+ model = nn.Sequential(
+ nn.Linear(2, hidden_dim),
+ nn.ReLU(),
+ nn.Linear(hidden_dim, hidden_dim),
+ nn.ReLU(),
+ nn.Linear(hidden_dim, 2),
+ )
+
+ return model
+
+
+######################################################################
+
+
+def generate_sets_and_params(
+ nb_mlps,
+ nb_samples,
+ batch_size,
+ nb_epochs,
+ device=torch.device("cpu"),
+ print_log=False,
+):
+ data_input = torch.zeros(nb_mlps, 2 * nb_samples, 2, device=device)
+ data_targets = torch.zeros(
+ nb_mlps, 2 * nb_samples, dtype=torch.int64, device=device
+ )
+
+ while (data_targets.float().mean(-1) - 0.5).abs().max() > 0.1:
+ i = (data_targets.float().mean(-1) - 0.5).abs() > 0.1
+ nb = i.sum()
+ print(f"{nb=}")
+
+ nb_rec = 2
+ support = torch.rand(nb, nb_rec, 2, 3, device=device) * 2 - 1
+ support = support.sort(-1).values
+ support = support[:, :, :, torch.tensor([0, 2])].view(nb, nb_rec, 4)
+
+ x = torch.rand(nb, 2 * nb_samples, 2, device=device) * 2 - 1
+ y = (
+ (
+ (x[:, None, :, 0] >= support[:, :, None, 0]).long()
+ * (x[:, None, :, 0] <= support[:, :, None, 1]).long()
+ * (x[:, None, :, 1] >= support[:, :, None, 2]).long()
+ * (x[:, None, :, 1] <= support[:, :, None, 3]).long()
+ )
+ .max(dim=1)
+ .values
+ )
+
+ data_input[i], data_targets[i] = x, y
+
+ train_input, train_targets = (
+ data_input[:, :nb_samples],
+ data_targets[:, :nb_samples],
+ )
+ test_input, test_targets = data_input[:, nb_samples:], data_targets[:, nb_samples:]
+
+ q_train_input = quantize(train_input, -1, 1)
+ train_input = dequantize(q_train_input, -1, 1)
+ train_targets = train_targets
+
+ q_test_input = quantize(test_input, -1, 1)
+ test_input = dequantize(q_test_input, -1, 1)
+ test_targets = test_targets
+
+ hidden_dim = 32
+ w1 = torch.randn(nb_mlps, hidden_dim, 2, device=device) / math.sqrt(2)
+ b1 = torch.zeros(nb_mlps, hidden_dim, device=device)
+ w2 = torch.randn(nb_mlps, 2, hidden_dim, device=device) / math.sqrt(hidden_dim)
+ b2 = torch.zeros(nb_mlps, 2, device=device)
+
+ w1.requires_grad_()
+ b1.requires_grad_()
+ w2.requires_grad_()
+ b2.requires_grad_()
+ optimizer = torch.optim.Adam([w1, b1, w2, b2], lr=1e-2)
+
+ criterion = nn.CrossEntropyLoss()
+ criterion.to(device)
+
+ for k in range(nb_epochs):
+ acc_train_loss = 0.0
+ nb_train_errors = 0
+
+ for input, targets in zip(
+ train_input.split(batch_size, dim=1), train_targets.split(batch_size, dim=1)
+ ):
+ h = torch.einsum("mij,mnj->mni", w1, input) + b1[:, None, :]
+ h = F.relu(h)
+ output = torch.einsum("mij,mnj->mni", w2, h) + b2[:, None, :]
+ loss = F.cross_entropy(
+ output.reshape(-1, output.size(-1)), targets.reshape(-1)
+ )
+ acc_train_loss += loss.item() * input.size(0)
+
+ wta = output.argmax(-1)
+ nb_train_errors += (wta != targets).long().sum(-1)
+
+ optimizer.zero_grad()
+ loss.backward()
+ optimizer.step()
+
+ with torch.no_grad():
+ for p in [w1, b1, w2, b2]:
+ m = (
+ torch.rand(p.size(), device=p.device) <= k / (nb_epochs - 1)
+ ).long()
+ pq = quantize(p, -2, 2)
+ p[...] = (1 - m) * p + m * dequantize(pq, -2, 2)
+
+ train_error = nb_train_errors / train_input.size(1)
+ acc_train_loss = acc_train_loss / train_input.size(1)
+
+ # print(f"{k=} {acc_train_loss=} {train_error=}")
+
+ q_params = torch.cat(
+ [quantize(p.view(nb_mlps, -1), -2, 2) for p in [w1, b1, w2, b2]], dim=1
+ )
+ q_train_set = torch.cat([q_train_input, train_targets[:, :, None]], -1).reshape(
+ nb_mlps, -1
+ )
+ q_test_set = torch.cat([q_test_input, test_targets[:, :, None]], -1).reshape(
+ nb_mlps, -1
+ )
+
+ return q_train_set, q_test_set, q_params
+
+
+######################################################################
+
+
+def evaluate_q_params(q_params, q_set, batch_size=25, device=torch.device("cpu")):
+ nb_mlps = q_params.size(0)
+ hidden_dim = 32
+ w1 = torch.empty(nb_mlps, hidden_dim, 2, device=device)
+ b1 = torch.empty(nb_mlps, hidden_dim, device=device)
+ w2 = torch.empty(nb_mlps, 2, hidden_dim, device=device)
+ b2 = torch.empty(nb_mlps, 2, device=device)
+
+ with torch.no_grad():
+ k = 0
+ for p in [w1, b1, w2, b2]:
+ print(f"{p.size()=}")
+ x = dequantize(q_params[:, k : k + p.numel() // nb_mlps], -2, 2).view(
+ p.size()
+ )
+ p.copy_(x)
+ k += p.numel() // nb_mlps
+
+ q_set = q_set.view(nb_mlps, -1, 3)
+ data_input = dequantize(q_set[:, :, :2], -1, 1).to(device)
+ data_targets = q_set[:, :, 2].to(device)
+
+ print(f"{data_input.size()=} {data_targets.size()=}")
+
+ criterion = nn.CrossEntropyLoss()
+ criterion.to(device)
+
+ acc_loss = 0.0
+ nb_errors = 0
+
+ for input, targets in zip(
+ data_input.split(batch_size, dim=1), data_targets.split(batch_size, dim=1)
+ ):
+ h = torch.einsum("mij,mnj->mni", w1, input) + b1[:, None, :]
+ h = F.relu(h)
+ output = torch.einsum("mij,mnj->mni", w2, h) + b2[:, None, :]
+ loss = F.cross_entropy(output.reshape(-1, output.size(-1)), targets.reshape(-1))
+ acc_loss += loss.item() * input.size(0)
+ wta = output.argmax(-1)
+ nb_errors += (wta != targets).long().sum(-1)
+
+ error = nb_errors / data_input.size(1)
+ acc_loss = acc_loss / data_input.size(1)
+
+ return error
+
+
+######################################################################
+
+
+def generate_sequence_and_test_set(
+ nb_mlps,
+ nb_samples,
+ batch_size,
+ nb_epochs,
+ device,
+):
+ q_train_set, q_test_set, q_params = generate_sets_and_params(
+ nb_mlps,
+ nb_samples,
+ batch_size,
+ nb_epochs,
+ device=device,
+ )
+
+ input = torch.cat(
+ [
+ q_train_set,
+ q_train_set.new_full(
+ (
+ q_train_set.size(0),
+ 1,
+ ),
+ nb_quantization_levels,
+ ),
+ q_params,
+ ],
+ dim=-1,
+ )
+
+ print(f"SANITY #1 {q_train_set.size()=} {q_params.size()=} {input.size()=}")
+
+ ar_mask = (
+ (torch.arange(input.size(0), device=input.device) > q_train_set.size(0) + 1)
+ .long()
+ .view(1, -1)
+ .reshape(nb_mlps, -1)
+ )
+
+ return input, ar_mask, q_test_set
+
+
+######################################################################
+
+if __name__ == "__main__":
+ import time
+
+ nb_mlps, nb_samples = 128, 200
+
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+ start_time = time.perf_counter()
+
+ data = []
+
+ for n in range(2):
+ data.append(
+ generate_sequence_and_test_set(
+ nb_mlps=nb_mlps,
+ nb_samples=nb_samples,
+ device=device,
+ batch_size=25,
+ nb_epochs=250,
+ )
+ )
+
+ end_time = time.perf_counter()
+ nb = sum([i.size(0) for i, _, _ in data])
+ print(f"{nb / (end_time - start_time):.02f} samples per second")
+
+ for input, ar_mask, q_test_set in data:
+ q_train_set = input[:, : nb_samples * 3]
+ q_params = input[:, nb_samples * 3 + 1 :]
+ print(f"SANITY #2 {q_train_set.size()=} {q_params.size()=} {input.size()=}")
+ error_train = evaluate_q_params(q_params, q_train_set)
+ print(f"train {error_train*100}%")
+ error_test = evaluate_q_params(q_params, q_test_set)
+ print(f"test {error_test*100}%")