🐛 don't train with too small batches

- rather "bigger" batches with more training steps - update Fig. 2 plots generation to 25MNAR
RasmussenLab · Nov 17, 2023 · 89046b4 · 89046b4
1 parent 0b0d747
commit 89046b4
Show file tree

Hide file tree

Showing 5 changed files with 36 additions and 29 deletions.
diff --git a/project/03_2_best_models_comparison_fig2.ipynb b/project/03_2_best_models_comparison_fig2.ipynb
@@ -34,11 +34,12 @@
    "outputs": [],
    "source": [
     "# parameters\n",
-    "FOLDER = Path('runs/dev_dataset_large/')\n",
+    "FOLDER = Path('runs/mnar_mcar/')\n",
+    "SIZE = 'l'\n",
     "files_in = {\n",
-    "    'protein groups': FOLDER / 'proteinGroups/figures/performance_test.csv',\n",
-    "    'peptides': FOLDER / 'peptides/figures/performance_test.csv',\n",
-    "    'precursors': FOLDER / 'evidence/figures/performance_test.csv'\n",
+    "    'protein groups': FOLDER / 'pg_l_25MNAR/figures/2_1_performance_test.csv',\n",
+    "    'peptides': FOLDER / 'pep_l_25MNAR/figures/2_1_performance_test.csv',\n",
+    "    'precursors': FOLDER / 'evi_l_25MNAR/figures/2_1_performance_test.csv'\n",
     "}"
    ]
   },
@@ -49,11 +50,12 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "FOLDER = Path('runs/dev_dataset_small/')\n",
+    "FOLDER = Path('runs/mnar_mcar/')\n",
+    "SIZE = 'm'\n",
     "files_in = {\n",
-    "    'protein groups': FOLDER / 'proteinGroups_N50/figures/performance_test.csv',\n",
-    "    'peptides': FOLDER / 'peptides_N50/figures/performance_test.csv',\n",
-    "    'precursors': FOLDER / 'evidence_N50/figures/performance_test.csv'\n",
+    "    'protein groups': FOLDER / 'pg_m_25MNAR/figures/2_1_performance_test.csv',\n",
+    "    'peptides': FOLDER / 'pep_m_25MNAR/figures/2_1_performance_test.csv',\n",
+    "    'precursors': FOLDER / 'evi_m_25MNAR/figures/2_1_performance_test.csv'\n",
     "}"
    ]
   },
@@ -134,13 +136,13 @@
    },
    "outputs": [],
    "source": [
-    "fname = FOLDER / 'best_models_1_test_mpl.pdf'\n",
+    "fname = FOLDER / f'best_models_{SIZE}_test_mpl.pdf'\n",
     "metrics = df['metric_value'].unstack('model')\n",
     "ORDER_MODELS = metrics.mean().sort_values().index.to_list()\n",
     "metrics = metrics.loc[ORDER_DATA, ORDER_MODELS]\n",
     "\n",
     "plt.rcParams['figure.figsize'] = [4.0, 2.0]\n",
-    "matplotlib.rcParams.update({'font.size': 5})\n",
+    "matplotlib.rcParams.update({'font.size': 6})\n",
     "\n",
     "ax = (metrics\n",
     "      .plot\n",
@@ -149,7 +151,7 @@
     "           ylabel=f\"{METRIC} (log2 intensities)\",\n",
     "           color=COLORS_TO_USE_MAPPTING,\n",
     "           width=.85,\n",
-    "           fontsize=8\n",
+    "           fontsize=7\n",
     "           ))\n",
     "\n",
     "ax = vaep.plotting.add_height_to_barplot(ax, size=5)\n",
@@ -230,7 +232,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "fname = FOLDER / 'performance_summary.xlsx'\n",
+    "fname = FOLDER / f'performance_summary_{SIZE}.xlsx'\n",
     "perf.to_excel(fname)\n",
     "fname.as_posix()"
    ]

diff --git a/project/03_2_best_models_comparison_fig2.py b/project/03_2_best_models_comparison_fig2.py
@@ -6,7 +6,7 @@
 #       extension: .py
 #       format_name: percent
 #       format_version: '1.3'
-#       jupytext_version: 1.15.0
+#       jupytext_version: 1.15.2
 #   kernelspec:
 #     display_name: Python 3
 #     language: python
@@ -33,19 +33,21 @@
 
 # %%
 # parameters
-FOLDER = Path('runs/dev_dataset_large/')
+FOLDER = Path('runs/mnar_mcar/')
+SIZE = 'l'
 files_in = {
-    'protein groups': FOLDER / 'proteinGroups/figures/performance_test.csv',
-    'peptides': FOLDER / 'peptides/figures/performance_test.csv',
-    'precursors': FOLDER / 'evidence/figures/performance_test.csv'
+    'protein groups': FOLDER / 'pg_l_25MNAR/figures/2_1_performance_test.csv',
+    'peptides': FOLDER / 'pep_l_25MNAR/figures/2_1_performance_test.csv',
+    'precursors': FOLDER / 'evi_l_25MNAR/figures/2_1_performance_test.csv'
 }
 
 # %%
-FOLDER = Path('runs/dev_dataset_small/')
+FOLDER = Path('runs/mnar_mcar/')
+SIZE = 'm'
 files_in = {
-    'protein groups': FOLDER / 'proteinGroups_N50/figures/performance_test.csv',
-    'peptides': FOLDER / 'peptides_N50/figures/performance_test.csv',
-    'precursors': FOLDER / 'evidence_N50/figures/performance_test.csv'
+    'protein groups': FOLDER / 'pg_m_25MNAR/figures/2_1_performance_test.csv',
+    'peptides': FOLDER / 'pep_m_25MNAR/figures/2_1_performance_test.csv',
+    'precursors': FOLDER / 'evi_m_25MNAR/figures/2_1_performance_test.csv'
 }
 
 # %%
@@ -84,7 +86,7 @@
 df
 
 # %%
-fname = FOLDER / 'best_models_1_test_mpl.pdf'
+fname = FOLDER / f'best_models_{SIZE}_test_mpl.pdf'
 metrics = df['metric_value'].unstack('model')
 ORDER_MODELS = metrics.mean().sort_values().index.to_list()
 metrics = metrics.loc[ORDER_DATA, ORDER_MODELS]
@@ -148,6 +150,6 @@
 perf
 
 # %%
-fname = FOLDER / 'performance_summary.xlsx'
+fname = FOLDER / f'performance_summary_{SIZE}.xlsx'
 perf.to_excel(fname)
 fname.as_posix()
diff --git a/project/config/single_dev_dataset/mnar_mcar/evi_m.yaml b/project/config/single_dev_dataset/mnar_mcar/evi_m.yaml
@@ -30,7 +30,8 @@ models:
                   model: DAE
                   file_format: csv
                   latent_dim: 25
-                  batch_size: 10
+                  batch_size: 25
+                  patience: 50
                   epochs_max: 200
                   hidden_layers: "256"
                   sample_idx_position: 0
@@ -40,7 +41,7 @@ models:
                   model: VAE
                   file_format: csv
                   latent_dim: 10
-                  batch_size: 10
+                  batch_size: 25
                   epochs_max: 200
                   hidden_layers: "256"
                   sample_idx_position: 0

diff --git a/project/config/single_dev_dataset/mnar_mcar/pep_m.yaml b/project/config/single_dev_dataset/mnar_mcar/pep_m.yaml
@@ -30,7 +30,8 @@ models:
               model: DAE
               file_format: csv
               latent_dim: 75
-              batch_size: 10
+              batch_size: 25
+              patience: 50    
               epochs_max: 200
               hidden_layers: "256_128"
               sample_idx_position: 0
@@ -40,7 +41,7 @@ models:
               model: VAE
               file_format: csv
               latent_dim: 50
-              batch_size: 10
+              batch_size: 25
               epochs_max: 200
               hidden_layers: "256"
               sample_idx_position: 0

diff --git a/project/config/single_dev_dataset/mnar_mcar/pg_m.yaml b/project/config/single_dev_dataset/mnar_mcar/pg_m.yaml
@@ -30,7 +30,8 @@ models:
       model: DAE
       file_format: csv
       latent_dim: 10
-      batch_size: 10
+      batch_size: 25
+      patience: 50
       epochs_max: 200
       hidden_layers: "512"
       sample_idx_position: 0
@@ -40,7 +41,7 @@ models:
       model: VAE
       file_format: csv
       latent_dim: 25
-      batch_size: 10
+      batch_size: 25
       epochs_max: 200
       hidden_layers: "512_256"
       sample_idx_position: 0