Fix usflows

.github/workflows/testing.yml

@@ -53,16 +53,9 @@ jobs:
           python-version: ${{ matrix.python-version }}
       - name: install setuptools
         run: pip${{ matrix.python-version }} install setuptools
-      - name: Build requirments.txt
-        uses: divideprojects/poetry-export-requirements-action@v1
-        with:
-          without-hashes: true
-          outfile-name: requirements.txt
       #----------------------------------------------
       # install dependencies if cache does not exist
       #----------------------------------------------
-      - name: Install dependencies
-        run: pip${{ matrix.python-version }} install -r requirements.txt
       - name: Install pytest
         run: pip${{ matrix.python-version }} install pytest
       #----------------------------------------------

experiments/cifar/cifar.yaml

@@ -0,0 +1,113 @@
+---
+__object__: src.explib.base.ExperimentCollection
+name: fashion_ablation
+experiments:
+  - &exp_rad_logN
+    __object__: src.explib.hyperopt.HyperoptExperiment
+    name: cfair_full_radial_logN
+    skip: true
+    device: cpu
+    skip: true
+    scheduler: &scheduler 
+      __object__: ray.tune.schedulers.ASHAScheduler
+      max_t: 1000000
+      grace_period: 1000000
+      reduction_factor: 2
+    num_hyperopt_samples: &num_hyperopt_samples 1
+    gpus_per_trial: &gpus_per_trial 0
+    cpus_per_trial: &cpus_per_trial 1
+    tuner_params: &tuner_params
+      metric: val_loss
+      mode: min
+    trial_config:
+      logging:
+        images: false
+        "image_shape": [28, 28]
+      dataset: &dataset
+        class:
+          __class__: src.explib.datasets.Cifar10Split
+        params:
+          space_to_depth_factor: 4
+          dataloc: /home/faried/Projects/USFlows/data/cifar10
+
+      epochs: &epochs 200000
+      patience: &patience 1
+      batch_size: &batch_size 
+        __eval__: tune.choice([32])
+      optim_cfg: &optim 
+        optimizer:
+          __class__: src.usflows.sophia.SophiaG
+        params:
+          lr: 
+            __eval__: 1e-3
+          weight_decay: 0.0
+
+      model_cfg:
+        type:
+          __class__: src.usflows.flows.USFlow
+        params:
+          soft_training: 
+            __eval__: tune.choice([False])
+          training_noise_prior:
+            __object__: pyro.distributions.Uniform
+            low:
+              __eval__: 1e-20
+            high: 0.01
+          prior_scale: 1.0
+          coupling_blocks: 
+            __eval__: tune.choice([10])
+          lu_transform: 1
+          householder: 0
+          conditioner_cls:
+            __class__: src.usflows.networks.ConvNet2D
+          conditioner_args:
+            c_in: 48
+            c_hidden: 
+              __eval__: tune.choice([32])
+            num_layers: 
+              __eval__: tune.choice([3])
+            padding: same
+            kernel_size: 3
+            rescale_hidden: 1
+            normalize_layers: 
+              __eval__: tune.choice([True])
+            gating: 
+              __eval__: tune.choice([True])
+          in_dims: [48, 8, 8]
+          affine_conjugation: true
+          nonlinearity:
+            __eval__: tune.choice([torch.nn.ReLU()])
+          base_distribution:
+            __object__: src.usflows.distributions.RadialDistribution
+            device: cpu
+            p: 
+              __eval__: float("1")
+            loc:
+              __eval__: torch.zeros([48, 8, 8]).to("cpu")
+            norm_distribution:
+              __object__: src.usflows.distributions.LogNormal
+              loc:
+                __eval__: torch.ones([1]).to("cpu") * 6
+              scale:
+                __eval__: torch.ones([1]).to("cpu") * .35
+              device: cpu
+  - &exp_normal
+    __overwrites__: *exp_rad_logN
+    name: fashion_full_laplace
+    skip: false
+    trial_config:
+      optim_cfg: 
+        params:
+          lr: 
+            __eval__: 1e-5
+      model_cfg: 
+          params:
+            lu_transform: 0
+            affine_conjugation: false
+            base_distribution:
+              __exact__: 
+                __object__: pyro.distributions.Normal
+                loc: 
+                  __eval__: torch.zeros([48, 8, 8]).to("cpu")
+                scale: 
+                  __eval__: torch.ones([48, 8, 8]).to("cpu")

experiments/fashion/fashion.yaml

@@ -4,92 +4,110 @@ name: fashion_ablation
 experiments:
   - &exp_rad_logN
     __object__: src.explib.hyperopt.HyperoptExperiment
-    name: fashion_full_radial_logN
-    device: cuda
+    name: cfair_full_radial_logN
+    skip: true
+    device: cpu
     skip: true
     scheduler: &scheduler 
       __object__: ray.tune.schedulers.ASHAScheduler
       max_t: 1000000
       grace_period: 1000000
       reduction_factor: 2
     num_hyperopt_samples: &num_hyperopt_samples 1
-    gpus_per_trial: &gpus_per_trial 1
+    gpus_per_trial: &gpus_per_trial 0
     cpus_per_trial: &cpus_per_trial 1
     tuner_params: &tuner_params
       metric: val_loss
       mode: min
     trial_config:
       logging:
-        images: true
+        images: false
         "image_shape": [28, 28]
       dataset: &dataset
-        __object__: src.explib.datasets.FashionMnistSplit
+        class:
+          __class__: src.explib.datasets.FashionMnistSplit
+        params:
+          space_to_depth_factor: 4
+          dataloc: /home/faried/Projects/USFlows/data/fashion
+
       epochs: &epochs 200000
-      patience: &patience 2
+      patience: &patience 1
       batch_size: &batch_size 
         __eval__: tune.choice([32])
       optim_cfg: &optim 
         optimizer:
-          __class__: torch.optim.Adam 
+          __class__: src.usflows.sophia.SophiaG
         params:
           lr: 
-            __eval__: 1e-4
+            __eval__: 1e-3
           weight_decay: 0.0
 
-      model_cfg: 
+      model_cfg:
         type:
-          __class__: &model src.veriflow.flows.NiceFlow
+          __class__: src.usflows.flows.USFlow
         params:
-          soft_training: true
+          soft_training: 
+            __eval__: tune.choice([False])
           training_noise_prior:
             __object__: pyro.distributions.Uniform
-            low: 
+            low:
               __eval__: 1e-20
             high: 0.01
           prior_scale: 1.0
-          coupling_layers: 10
-          coupling_nn_layers: [300, 300, 300] 
-          nonlinearity: &nonlinearity 
+          coupling_blocks: 
+            __eval__: tune.choice([10])
+          lu_transform: 1
+          householder: 0
+          conditioner_cls:
+            __class__: src.usflows.networks.ConvNet2D
+          conditioner_args:
+            c_in: 48
+            c_hidden: 
+              __eval__: tune.choice([32])
+            num_layers: 
+              __eval__: tune.choice([3])
+            padding: same
+            kernel_size: 3
+            rescale_hidden: 1
+            normalize_layers: 
+              __eval__: tune.choice([True])
+            gating: 
+              __eval__: tune.choice([True])
+          in_dims: [48, 8, 8]
+          affine_conjugation: true
+          nonlinearity:
             __eval__: tune.choice([torch.nn.ReLU()])
-          split_dim: 392
-          base_distribution: 
-            __object__: src.veriflow.distributions.RadialDistribution       
-            device: cuda
-            p: 1.0
-            loc: 
-              __eval__: torch.zeros(784).to("cuda")
+          base_distribution:
+            __object__: src.usflows.distributions.RadialDistribution
+            device: cpu
+            p: 
+              __eval__: float("1")
+            loc:
+              __eval__: torch.zeros([48, 8, 8]).to("cpu")
             norm_distribution:
-              __object__: pyro.distributions.LogNormal
-              loc: 
-                __eval__: torch.zeros(1).to("cuda")
-              scale: 
-                __eval__: (.5 * torch.ones(1)).to("cuda")
-          use_lu: true
-  - &exp_laplace
-    __overwrites__: *exp_rad_logN
-    name: fashion_full_laplace
-    skip: false
-    trial_config:
-      model_cfg: 
-          params:
-            base_distribution:
-              __exact__: 
-                __object__: pyro.distributions.Laplace
-                loc: 
-                  __eval__: torch.zeros(784).to("cuda")
-                scale: 
-                  __eval__: torch.ones(784).to("cuda")
+              __object__: src.usflows.distributions.LogNormal
+              loc:
+                __eval__: torch.ones([1]).to("cpu") * 6
+              scale:
+                __eval__: torch.ones([1]).to("cpu") * .35
+              device: cpu
   - &exp_normal
     __overwrites__: *exp_rad_logN
     name: fashion_full_laplace
     skip: false
     trial_config:
+      optim_cfg: 
+        params:
+          lr: 
+            __eval__: 1e-5
       model_cfg: 
           params:
+            lu_transform: 0
+            affine_conjugation: false
             base_distribution:
               __exact__: 
                 __object__: pyro.distributions.Normal
                 loc: 
-                  __eval__: torch.zeros(784).to("cuda")
+                  __eval__: torch.zeros([48, 8, 8]).to("cpu")
                 scale: 
-                  __eval__: torch.ones(784).to("cuda")
+                  __eval__: torch.ones([48, 8, 8]).to("cpu")