Merge iodic-diva-branch into main

dvc.lock

@@ -8,53 +8,53 @@ stages:
     deps:
     - path: http://www.manythings.org/anki/fra-eng.zip
       hash: md5
-      checksum: '29366693926503416364759562686349725439'
-      size: 7757635
+      checksum: '58835075968442113841504768009775441090'
+      size: 7833145
     outs:
     - path: fra.txt
       hash: md5
-      md5: f16099673fd64e9fda1e17927ad02248
-      size: 34292144
+      md5: efac451a5f83015366bfe6ac117b9ba4
+      size: 34591986
   train:
     cmd: python train.py
     deps:
     - path: fra.txt
       hash: md5
-      md5: f16099673fd64e9fda1e17927ad02248
-      size: 34292144
+      md5: efac451a5f83015366bfe6ac117b9ba4
+      size: 34591986
     - path: train.py
       hash: md5
-      md5: 9505d81175e09e8cc80a8905c95abed7
-      size: 7136
+      md5: 51d5640636dc54ef6171b81ec77b599f
+      size: 7224
     params:
       params.yaml:
         data_path: fra.txt
         model:
           batch_size: 512
           latent_dim: 8
           duration: 00:00:30:00
-          max_epochs: 2
+          max_epochs: 20
           optim:
-            lr: 0.01
+            lr: 0.004
         num_samples: 10000
         seed: 423
     outs:
     - path: model
       hash: md5
-      md5: b4e402de57bdfa7871c9ea74562482f5.dir
-      size: 18561
+      md5: deae7576b332077b74aaf9470428840d.dir
+      size: 18930
       nfiles: 1
     - path: results/artifacts
       hash: md5
-      md5: b4e402de57bdfa7871c9ea74562482f5.dir
-      size: 18561
+      md5: deae7576b332077b74aaf9470428840d.dir
+      size: 18930
       nfiles: 1
     - path: results/metrics.json
       hash: md5
-      md5: e525dbdc2466b77f199fe5fe24afc549
-      size: 256
+      md5: 1e13f4dfbfa45966ad62b7df95e79e47
+      size: 366
     - path: results/plots
       hash: md5
-      md5: 1dd4ea44497391604fc8c379789b52b4.dir
-      size: 240
-      nfiles: 5
+      md5: ac5e4f9a863532eccbdf9e2ab81cad46.dir
+      size: 1255
+      nfiles: 7

dvc.yaml

@@ -31,5 +31,5 @@ plots:
     x: step
 artifacts:
   best:
-    path: results/artifacts/epoch=0-step=2.ckpt
+    path: results/artifacts/epoch=0-step=16.ckpt
     type: model

params.yaml

@@ -2,9 +2,9 @@ model:
   batch_size: 512
   latent_dim: 8
   duration: 00:00:30:00
-  max_epochs: 2
+  max_epochs: 20
   optim:
-    lr: 0.01
+    lr: 0.004
 data_path: fra.txt
 num_samples: 10000
 seed: 423

results/metrics.json

@@ -1,14 +1,18 @@
 {
     "val": {
-        "loss": 1.7237510681152344,
-        "acc": 0.014084506779909134
+        "loss": 0.6839841604232788,
+        "acc": 0.01470289845019579
     },
-    "epoch": 1,
-    "step": 3,
+    "epoch": 9,
+    "step": 159,
     "train": {
         "epoch": {
-            "loss": 2.709216356277466,
-            "acc": 0.013341710902750492
+            "loss": 0.6881032586097717,
+            "acc": 0.01480439305305481
+        },
+        "step": {
+            "loss": 0.7074033617973328,
+            "acc": 0.013838487677276134
         }
     }
 }

results/params.yaml

@@ -1,3 +1,3 @@
 latent_dim: 8
 optim_params:
-  lr: 0.01
+  lr: 0.004

results/plots/metrics/epoch.tsv

@@ -1,5 +1,24 @@
 step	epoch
-1	0
-1	0
-3	1
-3	1
+15	0
+15	0
+31	1
+31	1
+47	2
+47	2
+49	3
+63	3
+63	3
+79	4
+79	4
+95	5
+95	5
+99	6
+111	6
+111	6
+127	7
+127	7
+143	8
+143	8
+149	9
+159	9
+159	9

results/plots/metrics/train/epoch/acc.tsv

@@ -1,3 +1,11 @@
 step	acc
-1	0.0067295110784471035
-3	0.013341710902750492
+15	0.013551988638937473
+31	0.014475204981863499
+47	0.014578690752387047
+63	0.012611094862222672
+79	0.012640479020774364
+95	0.0126886535435915
+111	0.012571862898766994
+127	0.012932837940752506
+143	0.013741872273385525
+159	0.01480439305305481

results/plots/metrics/train/epoch/loss.tsv

@@ -1,3 +1,11 @@
 step	loss
-1	4.339257717132568
-3	2.709216356277466
+15	2.651099920272827
+31	1.1034443378448486
+47	0.9179204106330872
+63	0.8584996461868286
+79	0.7974901795387268
+95	0.7613820433616638
+111	0.7376429438591003
+127	0.7168747782707214
+143	0.7019720673561096
+159	0.6881032586097717

results/plots/metrics/train/step/acc.tsv

@@ -0,0 +1,4 @@
+step	acc
+49	0.012820512987673283
+99	0.01176421158015728
+149	0.013838487677276134

results/plots/metrics/train/step/loss.tsv

@@ -0,0 +1,4 @@
+step	loss
+49	0.8957815766334534
+99	0.7495381236076355
+149	0.7074033617973328

results/plots/metrics/val/acc.tsv

@@ -1,3 +1,11 @@
 step	acc
-1	0.014448276720941067
-3	0.014084506779909134
+15	0.01582844741642475
+31	0.012163701467216015
+47	0.012383817695081234
+63	0.012163614854216576
+79	0.013182752765715122
+95	0.01217393297702074
+111	0.012239444069564342
+127	0.012904092669487
+143	0.014017394743859768
+159	0.01470289845019579

results/plots/metrics/val/loss.tsv

@@ -1,3 +1,11 @@
 step	loss
-1	3.0123467445373535
-3	1.7237510681152344
+15	1.277770757675171
+31	0.9838346242904663
+47	0.8614449501037598
+63	0.8240187168121338
+79	0.7715871334075928
+95	0.7506844997406006
+111	0.7279419302940369
+127	0.7120298147201538
+143	0.696197509765625
+159	0.6839841604232788

train.py

@@ -1,10 +1,12 @@
 import os
 import numpy as np
 import torch
-import pytorch_lightning as pl
+import lightning
+from lightning.pytorch import callbacks
 import torchmetrics
 from ruamel.yaml import YAML
 from dvclive import Live
+# from lightning.pytorch.loggers import DVCLiveLogger
 from dvclive.lightning import DVCLiveLogger
 
 yaml = YAML(typ="safe")
@@ -76,7 +78,7 @@
 
 
 # Define the model 
-class LSTMSeqToSeq(pl.LightningModule):
+class LSTMSeqToSeq(lightning.LightningModule):
     def __init__(self, latent_dim, optim_params):
         super().__init__()
         # Log parameters (saves them to self.hparams)
@@ -163,14 +165,14 @@ def __getitem__(self, idx):
 
 exp = Live("results", save_dvc_exp=True)
 live = DVCLiveLogger(report=None, experiment=exp, log_model=True)
-checkpoint = pl.callbacks.ModelCheckpoint(
+checkpoint = callbacks.ModelCheckpoint(
         dirpath="model",
         monitor="val_acc",
         mode="max",
         save_weights_only=True, every_n_epochs=1)
-timer = pl.callbacks.Timer(duration=params["model"]["duration"])
+timer = callbacks.Timer(duration=params["model"]["duration"])
 
-trainer = pl.Trainer(max_epochs=params["model"]["max_epochs"], logger=[live],
+trainer = lightning.Trainer(max_epochs=params["model"]["max_epochs"], logger=[live],
                      callbacks=[timer, checkpoint])
 trainer.fit(model=arch, train_dataloaders=train_loader,
         val_dataloaders=val_loader)