Skip to content

Commit 4c76639

Browse files
tobias-liaudattobias-liaudat
committed
Add notebook descriptions
1 parent 238f473 commit 4c76639

7 files changed

+117
-175
lines changed

‎examples/RI_imaging_QuantifAI_LCIs.ipynb‎

Lines changed: 18 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,22 @@
11
{
22
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"metadata": {},
6+
"source": [
7+
"# Bayesian error bars (i.e. Length of the credible interval (LCI)) example with the `QuantifAI` model\n",
8+
"\n",
9+
"In this notebook we:\n",
10+
"- set hyperparameters,\n",
11+
"- prepare the synthetic observations,\n",
12+
"- define the model, likelihood and prior,\n",
13+
"- estimate the MAP reconstruction through a convex optimisation algorithm,\n",
14+
"- compute the high posterior density (HPD) region approximation,\n",
15+
"- compute the Bayesian error bars (LCIs),\n",
16+
"- plot the LCIs at different superpixel sizes.\n",
17+
"\n"
18+
]
19+
},
320
{
421
"cell_type": "code",
522
"execution_count": 1,
@@ -27,7 +44,7 @@
2744
"# Import torch and select GPU\n",
2845
"import torch\n",
2946
"\n",
30-
"os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"2\"\n",
47+
"os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"0\"\n",
3148
"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
3249
"if torch.cuda.is_available():\n",
3350
" print(torch.cuda.is_available())\n",

‎examples/RI_imaging_QuantifAI_MAP_estimation.ipynb‎

Lines changed: 15 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,20 @@
11
{
22
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"metadata": {},
6+
"source": [
7+
"# Compute the MAP example with the `QuantifAI` model\n",
8+
"\n",
9+
"In this notebook we:\n",
10+
"- set hyperparameters,\n",
11+
"- prepare the synthetic observations,\n",
12+
"- define the model, likelihood and prior,\n",
13+
"- estimate the MAP reconstruction through a convex optimisation algorithm,\n",
14+
"- plot the MAP estimation result and the error.\n",
15+
"\n"
16+
]
17+
},
318
{
419
"cell_type": "code",
520
"execution_count": 1,

‎examples/RI_imaging_QuantifAI_fast_pixel_UQ.ipynb‎

Lines changed: 17 additions & 171 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,21 @@
11
{
22
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"metadata": {},
6+
"source": [
7+
"# Fast pixel uncertainty quantification (UQ) example with the `QuantifAI` model\n",
8+
"\n",
9+
"In this notebook we:\n",
10+
"- set hyperparameters,\n",
11+
"- prepare the synthetic observations,\n",
12+
"- define the model, likelihood and prior,\n",
13+
"- estimate the MAP reconstruction through a convex optimisation algorithm,\n",
14+
"- compute the fast pixel UQ maps,\n",
15+
"- compare the predicted error maps and the true error maps.\n",
16+
"\n"
17+
]
18+
},
319
{
420
"cell_type": "code",
521
"execution_count": 5,
@@ -24,12 +40,10 @@
2440
"\n",
2541
"import time as time\n",
2642
"\n",
27-
"from functools import partial\n",
28-
"\n",
2943
"# Import torch and select GPU\n",
3044
"import torch\n",
3145
"\n",
32-
"os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"2\"\n",
46+
"os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"0\"\n",
3347
"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
3448
"if torch.cuda.is_available():\n",
3549
" print(torch.cuda.is_available())\n",
@@ -559,174 +573,6 @@
559573
"outputs": [],
560574
"source": []
561575
},
562-
{
563-
"cell_type": "code",
564-
"execution_count": 21,
565-
"metadata": {},
566-
"outputs": [
567-
{
568-
"data": {
569-
"text/plain": [
570-
"65536"
571-
]
572-
},
573-
"execution_count": 21,
574-
"metadata": {},
575-
"output_type": "execute_result"
576-
}
577-
],
578-
"source": [
579-
"to_numpy(x_map).size"
580-
]
581-
},
582-
{
583-
"cell_type": "code",
584-
"execution_count": 15,
585-
"metadata": {},
586-
"outputs": [],
587-
"source": [
588-
"# Function handle for the potential\n",
589-
"def _fun(_x, CRR_model, mu, lmbd):\n",
590-
" return (lmbd / mu) * CRR_model.cost(mu * _x) + likelihood.fun(_x)\n",
591-
"\n",
592-
"\n",
593-
"# Evaluation of the potential\n",
594-
"fun = partial(_fun, CRR_model=CRR_model, mu=mu, lmbd=lmbd)\n",
595-
"# Evaluation of the potential in numpy\n",
596-
"fun_np = lambda _x: fun(qai.utils.to_tensor(_x, dtype=myType)).item()\n",
597-
"\n",
598-
"# Compute HPD region bound\n",
599-
"N = np_x_map.size\n",
600-
"tau_alpha = np.sqrt(16 * np.log(3 / alpha_prob))\n",
601-
"gamma_alpha = fun(x_map).item() + tau_alpha * np.sqrt(N) + N\n",
602-
"\n",
603-
"# Define the wavelet dict\n",
604-
"# Define the l1 norm with dict psi\n",
605-
"Psi = qai.operators.DictionaryWv_torch(wavs_list, levels)\n",
606-
"oper2wavelet = qai.operators.Operation2WaveletCoeffs_torch(Psi=Psi)\n",
607-
"\n",
608-
"# Clone MAP estimation and cast type for wavelet operations\n",
609-
"torch_map = torch.clone(x_map).to(torch.float64)\n",
610-
"torch_x = to_tensor(x_gt).to(torch.float64)"
611-
]
612-
},
613-
{
614-
"cell_type": "code",
615-
"execution_count": 16,
616-
"metadata": {},
617-
"outputs": [
618-
{
619-
"name": "stderr",
620-
"output_type": "stream",
621-
"text": [
622-
"/disk/xray0/tl3/repos/QuantifAI/quantifai/utils.py:48: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
623-
" return torch.tensor(z, device=device, dtype=dtype, requires_grad=False).reshape(\n"
624-
]
625-
}
626-
],
627-
"source": [
628-
"def _potential_to_bisect(thresh, fun_np, oper2wavelet, torch_map):\n",
629-
" thresh_img = oper2wavelet.full_op_threshold_img(\n",
630-
" torch_map, thresh, thresh_type=\"hard\"\n",
631-
" )\n",
632-
"\n",
633-
" return gamma_alpha - fun_np(thresh_img)\n",
634-
"\n",
635-
"\n",
636-
"# Evaluation of the potential\n",
637-
"potential_to_bisect = partial(\n",
638-
" _potential_to_bisect, fun_np=fun_np, oper2wavelet=oper2wavelet, torch_map=torch_map\n",
639-
")\n",
640-
"\n",
641-
"\n",
642-
"selected_thresh, bisec_iters = qai.map_uncertainty.bisection_method(\n",
643-
" potential_to_bisect, start_interval, iters, tol, return_iters=True\n",
644-
")\n",
645-
"select_thresh_img = oper2wavelet.full_op_threshold_img(torch_map, selected_thresh)"
646-
]
647-
},
648-
{
649-
"cell_type": "code",
650-
"execution_count": 17,
651-
"metadata": {},
652-
"outputs": [
653-
{
654-
"name": "stdout",
655-
"output_type": "stream",
656-
"text": [
657-
"SNR (thresh vs MAP) at lvl 0: 20.090000\n",
658-
"SNR (MAP vs GT) at lvl 0: 43.690000\n",
659-
"SNR (thresh vs MAP) at lvl 1: 17.080000\n",
660-
"SNR (MAP vs GT) at lvl 1: 39.120000\n",
661-
"SNR (thresh vs MAP) at lvl 2: 16.550000\n",
662-
"SNR (MAP vs GT) at lvl 2: 32.000000\n",
663-
"SNR (thresh vs MAP) at lvl 3: 22.430000\n",
664-
"SNR (MAP vs GT) at lvl 3: 31.100000\n",
665-
"SNR (thresh vs MAP) at lvl 4: 41.220000\n",
666-
"SNR (MAP vs GT) at lvl 4: 36.690000\n"
667-
]
668-
}
669-
],
670-
"source": [
671-
"modif_img_list = []\n",
672-
"GT_modif_img_list = []\n",
673-
"SNR_at_lvl_list = []\n",
674-
"SNR_at_lvl_map_vs_GT_list = []\n",
675-
"\n",
676-
"for modif_level in range(levels + 1):\n",
677-
" op = lambda x1, x2: x2\n",
678-
"\n",
679-
" modif_img = oper2wavelet.full_op_two_img(\n",
680-
" torch.clone(torch_map), torch.clone(select_thresh_img), op, level=modif_level\n",
681-
" )\n",
682-
" GT_modif_img = oper2wavelet.full_op_two_img(\n",
683-
" torch.clone(torch_x), torch.clone(torch_map), op, level=modif_level\n",
684-
" )\n",
685-
" print(\n",
686-
" \"SNR (thresh vs MAP) at lvl {:d}: {:f}\".format(\n",
687-
" modif_level, qai.utils.eval_snr(to_numpy(torch_map), to_numpy(modif_img))\n",
688-
" )\n",
689-
" )\n",
690-
" print(\n",
691-
" \"SNR (MAP vs GT) at lvl {:d}: {:f}\".format(\n",
692-
" modif_level, qai.utils.eval_snr(to_numpy(torch_x), to_numpy(GT_modif_img))\n",
693-
" )\n",
694-
" )\n",
695-
" modif_img_list.append(to_numpy(modif_img))\n",
696-
" GT_modif_img_list.append(to_numpy(GT_modif_img))\n",
697-
" SNR_at_lvl_list.append(qai.utils.eval_snr(to_numpy(torch_map), to_numpy(modif_img)))\n",
698-
" SNR_at_lvl_map_vs_GT_list.append(\n",
699-
" qai.utils.eval_snr(to_numpy(torch_x), to_numpy(GT_modif_img))\n",
700-
" )"
701-
]
702-
},
703-
{
704-
"cell_type": "code",
705-
"execution_count": 4,
706-
"metadata": {},
707-
"outputs": [
708-
{
709-
"data": {
710-
"text/plain": [
711-
"torch.dtype"
712-
]
713-
},
714-
"execution_count": 4,
715-
"metadata": {},
716-
"output_type": "execute_result"
717-
}
718-
],
719-
"source": [
720-
"type(torch.float)"
721-
]
722-
},
723-
{
724-
"cell_type": "code",
725-
"execution_count": null,
726-
"metadata": {},
727-
"outputs": [],
728-
"source": []
729-
},
730576
{
731577
"cell_type": "code",
732578
"execution_count": null,

‎examples/RI_imaging_QuantifAI_hypothesis_test.ipynb‎

Lines changed: 18 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,22 @@
11
{
22
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"metadata": {},
6+
"source": [
7+
"# Carry out hypothesis test on an inpainted surrogate image with the `QuantifAI` model\n",
8+
"\n",
9+
"In this notebook we:\n",
10+
"- set hyperparameters,\n",
11+
"- prepare the synthetic observations,\n",
12+
"- define the model, likelihood and prior,\n",
13+
"- estimate the MAP reconstruction through a convex optimisation algorithm,\n",
14+
"- define the region of interest,\n",
15+
"- inpaint the surrogate's region of interest using the prior,\n",
16+
"- compute the high posterior density region approximation,\n",
17+
"- compute potentials and carry out the hypothesis test.\n"
18+
]
19+
},
320
{
421
"cell_type": "code",
522
"execution_count": 1,
@@ -32,7 +49,7 @@
3249
"if M1:\n",
3350
" device = torch.device(\"mps\" if torch.backends.mps.is_available() else \"cpu\")\n",
3451
"else:\n",
35-
" os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"2\"\n",
52+
" os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"0\"\n",
3653
" device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
3754
" if torch.cuda.is_available():\n",
3855
" print(torch.cuda.is_available())\n",

‎examples/RI_imaging_QuantifAI_sampling.ipynb‎

Lines changed: 18 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,22 @@
11
{
22
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"metadata": {},
6+
"source": [
7+
"# Sampling from the posterior example with the `QuantifAI` model\n",
8+
"\n",
9+
"In this notebook we:\n",
10+
"- set hyperparameters,\n",
11+
"- prepare the synthetic observations,\n",
12+
"- define the model, likelihood and prior,\n",
13+
"- estimate the MAP reconstruction through a convex optimisation algorithm,\n",
14+
"- define sampling hyperparameters,\n",
15+
"- run the MCMC sampling algorithm,\n",
16+
"- compute posterior sample-based Bayesian error bars (a.k.a. LCIs) and posterior standard deviations at different superpixel sizes\n",
17+
"- plot results comparing posterior sample-based Bayesian error bars, posterior means and posterior standard deviation at different superpixel sizes.\n"
18+
]
19+
},
320
{
421
"cell_type": "code",
522
"execution_count": 1,
@@ -30,7 +47,7 @@
3047
"# Import torch and select GPU\n",
3148
"import torch\n",
3249
"\n",
33-
"os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"2\"\n",
50+
"os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"0\"\n",
3451
"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
3552
"if torch.cuda.is_available():\n",
3653
" print(torch.cuda.is_available())\n",

‎examples/RI_imaging_wavelets_LCIs.ipynb‎

Lines changed: 17 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,21 @@
11
{
22
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"metadata": {},
6+
"source": [
7+
"# Bayesian error bars (i.e. Length of the credible interval (LCI)) example with the wavelet-based model\n",
8+
"\n",
9+
"In this notebook we:\n",
10+
"- set hyperparameters,\n",
11+
"- prepare the synthetic observations,\n",
12+
"- define the model, likelihood and prior,\n",
13+
"- estimate the MAP reconstruction through a convex optimisation algorithm,\n",
14+
"- compute the high posterior density (HPD) region approximation,\n",
15+
"- compute the Bayesian error bars (LCIs),\n",
16+
"- plot the LCIs at different superpixel sizes.\n"
17+
]
18+
},
319
{
420
"cell_type": "code",
521
"execution_count": 1,
@@ -28,7 +44,7 @@
2844
"# Import torch and select GPU\n",
2945
"import torch\n",
3046
"\n",
31-
"os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"2\"\n",
47+
"os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"0\"\n",
3248
"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
3349
"if torch.cuda.is_available():\n",
3450
" print(torch.cuda.is_available())\n",

‎examples/RI_imaging_wavelets_MAP_estimation.ipynb‎

Lines changed: 14 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,19 @@
11
{
22
"cells": [
3+
{
4+
"cell_type": "markdown",
5+
"metadata": {},
6+
"source": [
7+
"# Compute the MAP example with the wavelet-based model\n",
8+
"\n",
9+
"In this notebook we:\n",
10+
"- set hyperparameters,\n",
11+
"- prepare the synthetic observations,\n",
12+
"- define the model, likelihood and prior,\n",
13+
"- estimate the MAP reconstruction through a convex optimisation algorithm,\n",
14+
"- plot the MAP estimation result and the error.\n"
15+
]
16+
},
317
{
418
"cell_type": "code",
519
"execution_count": 1,

0 commit comments

Comments
 (0)