v3.3 EDA update (#3)

- Need to track pre-webdataset catalog with LFS. (8dca1b6045337337112546548a552e44a0369131)
- Update data files, including pre-webdataset catalog, this is v3.3. (91beb3aab41c0f7af2ddf8459508b06e9231a4ed)
- Updated EDA in notebooks for v3.3, also added analysis of predicted entries prior to webdataset creation, shows 27K more images expectd. (6b5820466dacab211eafad859fb1ba385cc996ae)
- Added explanation of new files. (67e2a4af8580f9d2ac3f1b7a9ea7a0c74cb0ad9b)
- Added count of unique 7-tuples and genus-species pairs. (e00f09d636815a1caaab72cf72cf23482abdd994)

.gitattributes

@@ -53,7 +53,9 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.jpg filter=lfs diff=lfs merge=lfs -text
 *.jpeg filter=lfs diff=lfs merge=lfs -text
 *.webp filter=lfs diff=lfs merge=lfs -text
+# Large notebooks and data
 data/v1-dev-names.csv filter=lfs diff=lfs merge=lfs -text
 notebooks/BioCLIP_taxa_viz_bySource.ipynb filter=lfs diff=lfs merge=lfs -text
 statistics.csv filter=lfs diff=lfs merge=lfs -text
 data/catalog*.csv filter=lfs diff=lfs merge=lfs -text
+data/predicted-catalog.csv filter=lfs diff=lfs merge=lfs -text

README.md

@@ -13,6 +13,7 @@ The `data` folder contains
  - `catalog.csv`: the metadata and identifiers for all images in the dataset. This includes original data source, unique identifier within TreeOfLife10M, and the associated taxa information for the image. This is the file that will be updated, while preserving the `catalog-v1-dev.csv` file for ease of comparison.
 - `stats_avg_std_byClass.csv`: average and standard distribution of images given by class in `catalog.csv`. This is for both all 
 images, and images that have labels.
+- `predicted-catalog.csv`: the metadata and identifiers for all entries expected in the dataset, prior to webdataset creation. This includes original data source, unique identifier within TreeOfLife10M, and the associated taxa information. In version 3.3 (first time it was generated), it showed 27K more entries than in the webdataset and they all had full taxonomic rank filled.
 
  - `catalog-v1-dev.csv`: the metadata and identifiers for all images in the dataset. This includes original data source, unique identifier within TreeOfLife10M, and the associated taxa information for the image. This file will be maintained for v1 reference, and as such will not be updated.
  - `taxa_counts.csv`: count of distinct lower taxa within each higher taxon from `kingdom` down to 
@@ -26,6 +27,8 @@ images, and images that have labels.
 The `notebooks` folder contains
  - `ToL_catalog_EDA.ipynb`: more full EDA of TreeOfLife10M dataset using `catalog.csv`. To be updated as `catalog.csv` is updated, i.e., as the dataset is updated.
  - `ToL_catalog_EDA.py`: py file paired to `ToL_catalog_EDA.ipynb` to facilitate diff checking in case of cell text changes in notebook.
+ - `ToL_predicted-catalog_EDA.ipynb`: more full EDA of TreeOfLife10M dataset using `predicted-catalog.csv`. To be updated as `predicted-catalog.csv` is updated, i.e., as the dataset is updated.
+ - `ToL_predicted-catalog_EDA.py`: py file paired to `ToL_predicted-catalog_EDA.ipynb` to facilitate diff checking in case of cell text changes in notebook.
 
  - `BioCLIP_data_viz.ipynb`: notebook with quick basic stats for `catalog-v1-dev.csv`, generates `taxa_counts.csv`.
  - `BioCLIP_taxa_viz_bySource.ipynb`: generates data visualizations, in particular, the generation of visualizations in `visuals` folder and some histograms. The treemaps produced in the notebook are interactive.
@@ -33,7 +36,7 @@ The `notebooks` folder contains
 direction of standardization efforts. Maintained for v1 reference, should not be updated.
  - `missing_taxa_testGen.ipynb`: generates `tol_hierarchy_test.csv` to test [`check_taxa` script](https://github.com/Imageomics/open_clip/blob/main/scripts/evobio10m/check_taxa.py). Also observes species labeled as `(unidentified)` in EOL data.
 
-**Note:** run `pip install -r requirements.txt` before starting the visualization notebooks.
+**Note:** run `pip install -r requirements.txt` before starting the visualization notebooks. EDA notebooks only require `pandas` and `seaborn`.
 
 ### Visuals
 

data/catalog.csv

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:315576dd390920972ad48accbef002b5f19ca6838a6035cd1f32f09e3ee6d411
-size 2018327986
+oid sha256:ef5edf6a6e6048c98e74787cc143981095efaa27ba0aa4cebbac8116b24c9735
+size 2024176607

data/missing_taxa_output.txt

@@ -70,3 +70,27 @@ Output of taxa check for catalog.csv v3.2 (taxa check now also removes duplicate
 [2023-10-26 13:44:14,425] [WARNING] [root] There are 149057 samples for which the species column may have genus and species.
 
 ------------------------------------------------------------------------------------------------------
+
+Output of taxa check for catalog.csv v3.3 (from pr3):
+
+[2023-10-30 17:31:38,357] [WARNING] [root] There are 12 kingdoms instead of 3.
+[2023-10-30 17:31:39,420] [WARNING] [root] 39574 entries are missing rank kingdom, but have genus label.
+[2023-10-30 17:31:39,681] [WARNING] [root] 38324 entries are missing rank phylum, but have genus label.
+[2023-10-30 17:31:39,942] [WARNING] [root] 44828 entries are missing rank class, but have genus label.
+[2023-10-30 17:31:40,199] [WARNING] [root] 42771 entries are missing rank order, but have genus label.
+[2023-10-30 17:31:40,456] [WARNING] [root] 40131 entries are missing rank family, but have genus label.
+[2023-10-30 17:31:40,733] [WARNING] [root] 232 entries are missing rank kingdom, but have family label.
+[2023-10-30 17:31:40,759] [WARNING] [root] 223 entries are missing rank phylum, but have family label.
+[2023-10-30 17:31:40,785] [WARNING] [root] 493 entries are missing rank class, but have family label.
+[2023-10-30 17:31:40,812] [WARNING] [root] 345 entries are missing rank order, but have family label.
+[2023-10-30 17:31:40,853] [WARNING] [root] 170 entries are missing rank kingdom, but have order label.
+[2023-10-30 17:31:40,854] [WARNING] [root] 105 entries are missing rank phylum, but have order label.
+[2023-10-30 17:31:40,855] [WARNING] [root] 1263 entries are missing rank class, but have order label.
+[2023-10-30 17:31:40,893] [WARNING] [root] 41 entries have kingdom and species labels but no genus.
+[2023-10-30 17:31:40,897] [WARNING] [root] 41 entries have phylum and species labels but no genus.
+[2023-10-30 17:31:40,901] [WARNING] [root] 41 entries have class and species labels but no genus.
+[2023-10-30 17:31:40,905] [WARNING] [root] 41 entries have order and species labels but no genus.
+[2023-10-30 17:31:40,908] [WARNING] [root] 41 entries have family and species labels but no genus.
+[2023-10-30 17:31:42,311] [WARNING] [root] There are 137661 samples with species label longer than one word.
+
+------------------------------------------------------------------------------------------------------

data/predicted-catalog.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e17a9325927e2dff21bc30bb8b3151a300fd456e4f725a626692e76fca6a0f2b
+size 1847398611

data/stats_avg_std_byClass.csv

@@ -1,9 +1,9 @@
 class,average_all_imgs,standard_deviation,avg_labeled,std_dev_labeled
-kingdom,1490931.5714285714,1221.0370884737988,1383385.142857143,1176.1739424324717
-phylum,115961.34444444445,340.53097428052627,107705.3111111111,328.1848733733947
-class,36748.31338028169,191.69849603030715,34047.58450704225,184.5198756422794
-order,7841.1126972201355,88.55005757886403,7262.970698722765,85.22306435890911
-family,1339.3892453798767,36.59766721226746,1236.6025410677619,35.165359959308844
-genus,141.30713405635214,11.887267728807664,120.2536794075037,10.966023864988792
-species,63.3115005702361,7.9568524285823035,52.917740409114074,7.274458083535438
-common,23.458496762142808,4.8433972335688935,22.625316649171822,4.756607682915612
+kingdom,869710.0833333334,932.582480713279,817063.9166666666,903.9158791982065
+phylum,96634.45370370371,310.86082690442635,90799.6574074074,301.32981499912586
+class,30076.42939481268,173.4255730704462,28203.440922190202,167.93880112168898
+order,6980.9505016722405,83.55208256933061,6542.110367892977,80.88331328458904
+family,1309.8043423694778,36.19121913350637,1223.4554467871485,34.977927994481725
+genus,141.30713405635214,11.887267728807664,120.25013201186076,10.96586211895174
+species,63.770697250957795,7.9856557182837395,53.29990773385801,7.300678580368952
+common,23.511079121780227,4.848822446922575,22.675425435573036,4.761872051575203

notebooks/ToL_catalog_EDA.py

@@ -39,22 +39,29 @@ df = df.loc[df.split != 'train_small']
 df.info(show_counts = True)
 
 # %% [markdown]
-# Original version had 10,436,521 entries; we expected loss of about 84K from the genera with label "unknown", but there's still another ~300K missing for this 10,065,845. Likely due to dropping subspecies, which will now be integrated back in under their species for the next round.
+# Original version had 10,436,521 entries; we expected loss of about 84K from the genera with label "unknown", but there's still another ~300K missing for this 10,065,576. It seems subspecies were not integrated back in under their species for this version, as we now have 269 less images than last time.
 #
-# Labeling definitely has far better coverage from original, and even improves over the last version (of course `common` is completely filled, but we also gained almost 50K more taxa labels across the board).
+# Coverage for species also dropped by 269, genus is similar.
+#
+# Labeling definitely has far better coverage from original still.
 
 # %%
 df.nunique()
 
 # %% [markdown]
-# There are 503,595 unique EOL page IDs, suggesting 503,595 unique classes among the 6,250,689 images pulled from EOL (and maintained here). Presumably this would represent the number of species or other lowest rank taxa covered by EOL.
+# There are 503,589 unique EOL page IDs, suggesting 503,589 unique classes among the 6,250,420 images pulled from EOL (and maintained here). Presumably this would represent the number of species or other lowest rank taxa covered by EOL.
+#
+# This is only 6 less pages than v3.2, but 269 less images.
 
 # %% [markdown]
-# Notice that we have 7 unique kingdoms, which we're sticking with.
+# Notice that we now have 12 unique kingdoms, instead of the 7 we were sticking with.
 
 # %%
 df['kingdom'].value_counts()
 
+# %% [markdown]
+# `Viridiplantae`, `Protista`, `Orthornavirae`, `Bamfordvirae`, and `Shotokuvirae` are all new. 
+
 # %% [markdown]
 # `Metazoa` and `Archaeplastida` have been replaced by `Animalia` and `Plantae`, setting them as the dominant kingdoms represented (which we'd expect). 
 #
@@ -72,39 +79,61 @@ df_all_taxa = df.dropna(subset = taxa)
 df_all_taxa[taxa].info(show_counts = True)
 
 # %% [markdown]
-# We have 8,320,837 images with full taxonomic labels.
+# We have 8,455,243 images with full taxonomic labels.
 #
-# Notice that we still have gaps in the taxonomic hierarchy (both higher and lower values), as this is less than the number of species labels in the dataset and species had the least non-null values. 
+# Notice that we still have gaps in the taxonomic hierarchy (both higher and lower values), as this is less than the number of species labels in the dataset and species had the least non-null values, but the gap is closing! 
 #
-# We did hit 8M+ entries with full taxonomic labels, so that's really good, especially considering 1,043,863 of the BIOSCAN sourced images don't have species labels.
+# We're at 8.4M+ entries with full taxonomic labels, so that's really good, especially considering 1,043,863 of the BIOSCAN sourced images don't have species labels.
+
+# %% [markdown]
+# How about number of unique 7-tuples?
+
+# %%
+#number of unique 7-tuples in full dataset
+df['duplicate'] = df.duplicated(subset = taxa, keep = 'first')
+df_unique_taxa = df.loc[~df['duplicate']]
+
+# %%
+df_unique_taxa.info(show_counts = True)
+
+# %% [markdown]
+# And number of unique `genus-species` pairs?
+
+# %%
+#number of unique genus-species pairs in full dataset
+df['sciName-duplicate'] = df.duplicated(subset = ['genus', 'species'], keep = 'first')
+df_unique_sciName = df.loc[~df['sciName-duplicate']]
+
+# %%
+df_unique_sciName.info(show_counts = True)
 
 # %% [markdown]
 # More detail on these missing values from [`check_taxa` script](https://github.com/Imageomics/open_clip/blob/main/scripts/evobio10m/check_taxa.py) are displayed below. Note: full record of taxa check output is in `data/missing_taxa_output.txt`.
 #
 # ```
-# [2023-10-26 13:44:09,780] [WARNING] [root] There are 7 kingdoms instead of 3.
-# [2023-10-26 13:44:11,325] [WARNING] [root] 160824 entries are missing rank kingdom, but have genus label.
-# [2023-10-26 13:44:11,613] [WARNING] [root] 151405 entries are missing rank phylum, but have genus label.
-# [2023-10-26 13:44:11,910] [WARNING] [root] 161871 entries are missing rank class, but have genus label.
-# [2023-10-26 13:44:12,199] [WARNING] [root] 156405 entries are missing rank order, but have genus label.
-# [2023-10-26 13:44:12,483] [WARNING] [root] 153279 entries are missing rank family, but have genus label.
-# [2023-10-26 13:44:12,782] [WARNING] [root] 315 entries are missing rank kingdom, but have family label.
-# [2023-10-26 13:44:12,809] [WARNING] [root] 289 entries are missing rank phylum, but have family label.
-# [2023-10-26 13:44:12,837] [WARNING] [root] 792 entries are missing rank class, but have family label.
-# [2023-10-26 13:44:12,865] [WARNING] [root] 414 entries are missing rank order, but have family label.
-# [2023-10-26 13:44:12,905] [WARNING] [root] 170 entries are missing rank kingdom, but have order label.
-# [2023-10-26 13:44:12,906] [WARNING] [root] 105 entries are missing rank phylum, but have order label.
-# [2023-10-26 13:44:12,908] [WARNING] [root] 1263 entries are missing rank class, but have order label.
-# [2023-10-26 13:44:12,949] [WARNING] [root] 41 entries have kingdom and species labels but no genus.
-# [2023-10-26 13:44:12,953] [WARNING] [root] 41 entries have phylum and species labels but no genus.
-# [2023-10-26 13:44:12,957] [WARNING] [root] 41 entries have class and species labels but no genus.
-# [2023-10-26 13:44:12,961] [WARNING] [root] 41 entries have order and species labels but no genus.
-# [2023-10-26 13:44:12,965] [WARNING] [root] 41 entries have family and species labels but no genus.
-# [2023-10-26 13:44:14,425] [WARNING] [root] There are 149057 samples for which the species column may have genus and species.
+# [2023-10-30 17:31:38,357] [WARNING] [root] There are 12 kingdoms instead of 3.
+# [2023-10-30 17:31:39,420] [WARNING] [root] 39574 entries are missing rank kingdom, but have genus label.
+# [2023-10-30 17:31:39,681] [WARNING] [root] 38324 entries are missing rank phylum, but have genus label.
+# [2023-10-30 17:31:39,942] [WARNING] [root] 44828 entries are missing rank class, but have genus label.
+# [2023-10-30 17:31:40,199] [WARNING] [root] 42771 entries are missing rank order, but have genus label.
+# [2023-10-30 17:31:40,456] [WARNING] [root] 40131 entries are missing rank family, but have genus label.
+# [2023-10-30 17:31:40,733] [WARNING] [root] 232 entries are missing rank kingdom, but have family label.
+# [2023-10-30 17:31:40,759] [WARNING] [root] 223 entries are missing rank phylum, but have family label.
+# [2023-10-30 17:31:40,785] [WARNING] [root] 493 entries are missing rank class, but have family label.
+# [2023-10-30 17:31:40,812] [WARNING] [root] 345 entries are missing rank order, but have family label.
+# [2023-10-30 17:31:40,853] [WARNING] [root] 170 entries are missing rank kingdom, but have order label.
+# [2023-10-30 17:31:40,854] [WARNING] [root] 105 entries are missing rank phylum, but have order label.
+# [2023-10-30 17:31:40,855] [WARNING] [root] 1263 entries are missing rank class, but have order label.
+# [2023-10-30 17:31:40,893] [WARNING] [root] 41 entries have kingdom and species labels but no genus.
+# [2023-10-30 17:31:40,897] [WARNING] [root] 41 entries have phylum and species labels but no genus.
+# [2023-10-30 17:31:40,901] [WARNING] [root] 41 entries have class and species labels but no genus.
+# [2023-10-30 17:31:40,905] [WARNING] [root] 41 entries have order and species labels but no genus.
+# [2023-10-30 17:31:40,908] [WARNING] [root] 41 entries have family and species labels but no genus.
+# [2023-10-30 17:31:42,311] [WARNING] [root] There are 137661 samples with species label longer than one word.
 # ```
 
 # %% [markdown]
-# Can we get some more information on those 41 entries that are still missing genus?
+# Can we get some more information on those 41 entries that are still missing genus? (should be same as before)
 
 # %%
 missing_genus = df.dropna(subset = ['kingdom', 'phylum', 'class', 'order', 'family', 'species'])
@@ -120,7 +149,7 @@ for taxon in taxa:
     print(taxon, ": ", missing_genus[taxon].unique())
 
 # %% [markdown]
-# "celebensis" is new, but next iteration should resolve these.
+# All the same as last time, will stay.
 
 # %% [markdown]
 # Let's add a column indicating the original data source so we can also get some stats by datasource.
@@ -138,7 +167,7 @@ missing_genus = missing_genus.loc[missing_genus.genus.isna()]
 missing_genus.data_source.unique()
 
 # %% [markdown]
-# Missing genus with all other taxa occurs only in EOL now, so should get fixed.
+# Missing genus with all other taxa occurs only in EOL still. It's what it is for this version.
 
 # %% [markdown]
 # First, check their unique class values (`common`).
@@ -155,9 +184,9 @@ df.loc[df['data_source'] == 'BIOSCAN', 'common'].nunique()
 # %% [markdown]
 # iNat's number of unique values in `common` is still up by 6 from the original --- probably pulled from ITIS instead and that has more listed?
 #
-# EOL's went back up a bit, though still about 70-80K less than the original version.
+# EOL's went back down, it's about 80K less than the original version.
 #
-# BIOSCAN's counts went down even more...
+# BIOSCAN's counts stabilized.
 
 # %% [markdown]
 # Make `df_taxa` with just taxa columns (+ `common` & `data_source`) so it's smaller to process faster.
@@ -228,6 +257,8 @@ bioscan_df.nunique()
 
 # %% [markdown]
 # We have half as many unique species and 3,000 less unique common names as in the original. Can this be accounted for by the adjustment in genus?
+#
+# We now seem to have gained an order over the last iteration--that would be the name resolver I guess.
 
 # %%
 bioscan_df['kingdom'].value_counts()
@@ -320,7 +351,7 @@ bioscan_long_species = bioscan_species_len_df.loc[bioscan_species_len_df["len_sp
 bioscan_long_species.info(show_counts = True)
 
 # %% [markdown]
-# Not all species indicated have length greater than 1 now, though this number has actually gone up by 140 since the last iteration.
+# Not all species indicated have length greater than 1 now, consistent with the last iteration.
 
 # %%
 bioscan_species_len_df.len_species.value_counts()
@@ -352,13 +383,16 @@ eol_df.info(show_counts = True)
 # %%
 eol_df.nunique()
 
+# %% [markdown]
+# `kingdom` through `family` gained new unique labels and greater coverage, while `genus` lost images, but did not lose representation (still the same number of unique genera). Both the number of `species` and `common` went down along with the number of unique such labels.
+
 # %%
 eol_df.loc[eol_df.species.isna()].nunique()
 
 # %% [markdown]
-# There are 503,595 unique page IDs from EOL in this dataset, which clearly represent varying levels of taxa. 
+# There are 503,589 unique page IDs from EOL in this dataset, which clearly represent varying levels of taxa. 
 #
-# Unique species + unique common where species is null (39,233) does not reach this number. But we have added SO many more phyla through family!
+# Unique species + unique common where species is null (39,232) does not reach this number. But we still added more phyla through family!
 
 # %%
 eol_df['kingdom'].value_counts()
@@ -366,7 +400,7 @@ eol_df['kingdom'].value_counts()
 # %% [markdown]
 # We have much greater kingdom variety here.
 #
-# We have already observed that not all ranks are filled in at the higher levels, sometimes having just one gap. This has been greatly improved from the first iteration and we'll get one more pass.
+# We have already observed that not all ranks are filled in at the higher levels, sometimes having just one gap. This has been greatly improved from the first iteration and again with this last one.
 
 # %%
 #number of unique 7-tuples in EOL
@@ -387,7 +421,7 @@ eol_df.loc[eol_df.species.notna()].info(show_counts = True)
 # %% [markdown]
 # Not all labeled species have higher order taxa, but they do all have `common` (likely because if there is no common name, it pulls scientific name (genus-species)).
 #
-# This will hopefully be filled with the next iteration.
+# Still a 200-300K difference.
 
 # %% [markdown]
 # Let's get a quick sample of the `common` column for images both with and without `species` labels. 
@@ -445,6 +479,18 @@ eol_long_species.loc[eol_long_species["len_species"] < 7].sample(7)
 #
 # Ex from earlier: "Adelpha godmani" is a butterfly with species name given by Fruhstorfer in 1913 ([source](https://butterfliesofamerica.com/L/t/Adelpha_godmani_a.htm)), it was listed as "adelpha godmani fruhstorfer".
 
+# %% [markdown]
+# #### Check for webdataset loss
+#
+# Are we potentially losing entries that have a longer number of words in species column? Let's see if any of these have full taxa.
+
+# %%
+eol_long_all_taxa = eol_long_species.dropna(subset = taxa)
+eol_long_all_taxa[taxa].info(show_counts = True)
+
+# %% [markdown]
+# That's a good number of images, so unlikely to be the cause.
+
 # %% [markdown]
 # ### Label Overlap Check
 
@@ -471,6 +517,8 @@ print(f"There are {len(list(set(gen_overlap) & set(bioscan_genera)))} genera sha
 
 # %% [markdown]
 # BIOSCAN and iNat21's overlap of genera is completely contained in EOL.
+#
+# No changes here.
 
 # %% [markdown]
 # ## Overall Stats

notebooks/ToL_predicted-catalog_EDA.ipynb

@@ -0,0 +1,929 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import pandas as pd\n",
+    "import seaborn as sns\n",
+    "\n",
+    "sns.set_style(\"whitegrid\")\n",
+    "sns.set(rc = {'figure.figsize': (10,10)})"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/var/folders/nv/f0fq1p1n1_3b11x579py_0q80000gq/T/ipykernel_21854/4112647395.py:1: DtypeWarning: Columns (5,6,7,14,15) have mixed types. Specify dtype option on import or set low_memory=False.\n",
+      "  df = pd.read_csv(\"../data/predicted-catalog.csv\")\n"
+     ]
+    }
+   ],
+   "source": [
+    "df = pd.read_csv(\"../data/predicted-catalog.csv\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/html": [
+       "<div>\n",
+       "<style scoped>\n",
+       "    .dataframe tbody tr th:only-of-type {\n",
+       "        vertical-align: middle;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe tbody tr th {\n",
+       "        vertical-align: top;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe thead th {\n",
+       "        text-align: right;\n",
+       "    }\n",
+       "</style>\n",
+       "<table border=\"1\" class=\"dataframe\">\n",
+       "  <thead>\n",
+       "    <tr style=\"text-align: right;\">\n",
+       "      <th></th>\n",
+       "      <th>split</th>\n",
+       "      <th>treeoflife_id</th>\n",
+       "      <th>eol_content_id</th>\n",
+       "      <th>eol_page_id</th>\n",
+       "      <th>bioscan_part</th>\n",
+       "      <th>bioscan_filename</th>\n",
+       "      <th>inat21_filename</th>\n",
+       "      <th>inat21_cls_name</th>\n",
+       "      <th>inat21_cls_num</th>\n",
+       "      <th>kingdom</th>\n",
+       "      <th>phylum</th>\n",
+       "      <th>class</th>\n",
+       "      <th>order</th>\n",
+       "      <th>family</th>\n",
+       "      <th>genus</th>\n",
+       "      <th>species</th>\n",
+       "      <th>common</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>train</td>\n",
+       "      <td>f2f0aa29-e87b-469c-bf5b-51a3611ab001</td>\n",
+       "      <td>22131926.0</td>\n",
+       "      <td>269504.0</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>Animalia</td>\n",
+       "      <td>Arthropoda</td>\n",
+       "      <td>Insecta</td>\n",
+       "      <td>Lepidoptera</td>\n",
+       "      <td>Lycaenidae</td>\n",
+       "      <td>Orthomiella</td>\n",
+       "      <td>rantaizana</td>\n",
+       "      <td>Chinese Straight-wing Blue</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>1</th>\n",
+       "      <td>train</td>\n",
+       "      <td>5faa4f55-32e9-4872-953d-567e5d232e52</td>\n",
+       "      <td>22291283.0</td>\n",
+       "      <td>6101931.0</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>Plantae</td>\n",
+       "      <td>Tracheophyta</td>\n",
+       "      <td>Polypodiopsida</td>\n",
+       "      <td>Polypodiales</td>\n",
+       "      <td>Woodsiaceae</td>\n",
+       "      <td>Woodsia</td>\n",
+       "      <td>subcordata</td>\n",
+       "      <td>Woodsia subcordata</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>2</th>\n",
+       "      <td>train</td>\n",
+       "      <td>2282f2bf-2b52-4522-b588-dd6f356d5fd6</td>\n",
+       "      <td>21802775.0</td>\n",
+       "      <td>45513632.0</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>Animalia</td>\n",
+       "      <td>Chordata</td>\n",
+       "      <td>Aves</td>\n",
+       "      <td>Passeriformes</td>\n",
+       "      <td>Laniidae</td>\n",
+       "      <td>Lanius</td>\n",
+       "      <td>minor</td>\n",
+       "      <td>Lesser Grey Shrike</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>3</th>\n",
+       "      <td>train</td>\n",
+       "      <td>76b57c36-2181-4e6d-a5c4-b40e22a09449</td>\n",
+       "      <td>12784812.0</td>\n",
+       "      <td>51655800.0</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>tenuis</td>\n",
+       "      <td>Tenuis</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>4</th>\n",
+       "      <td>train</td>\n",
+       "      <td>f57d3ab6-2cf5-484b-a590-e2a3d49a3ca2</td>\n",
+       "      <td>29713643.0</td>\n",
+       "      <td>45515896.0</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>Animalia</td>\n",
+       "      <td>Chordata</td>\n",
+       "      <td>Aves</td>\n",
+       "      <td>Casuariiformes</td>\n",
+       "      <td>Casuariidae</td>\n",
+       "      <td>Casuarius</td>\n",
+       "      <td>casuarius</td>\n",
+       "      <td>Southern Cassowary</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ],
+      "text/plain": [
+       "   split                         treeoflife_id  eol_content_id  eol_page_id   \n",
+       "0  train  f2f0aa29-e87b-469c-bf5b-51a3611ab001      22131926.0     269504.0  \\\n",
+       "1  train  5faa4f55-32e9-4872-953d-567e5d232e52      22291283.0    6101931.0   \n",
+       "2  train  2282f2bf-2b52-4522-b588-dd6f356d5fd6      21802775.0   45513632.0   \n",
+       "3  train  76b57c36-2181-4e6d-a5c4-b40e22a09449      12784812.0   51655800.0   \n",
+       "4  train  f57d3ab6-2cf5-484b-a590-e2a3d49a3ca2      29713643.0   45515896.0   \n",
+       "\n",
+       "   bioscan_part bioscan_filename inat21_filename inat21_cls_name   \n",
+       "0           NaN              NaN             NaN             NaN  \\\n",
+       "1           NaN              NaN             NaN             NaN   \n",
+       "2           NaN              NaN             NaN             NaN   \n",
+       "3           NaN              NaN             NaN             NaN   \n",
+       "4           NaN              NaN             NaN             NaN   \n",
+       "\n",
+       "   inat21_cls_num   kingdom        phylum           class           order   \n",
+       "0             NaN  Animalia    Arthropoda         Insecta     Lepidoptera  \\\n",
+       "1             NaN   Plantae  Tracheophyta  Polypodiopsida    Polypodiales   \n",
+       "2             NaN  Animalia      Chordata            Aves   Passeriformes   \n",
+       "3             NaN       NaN           NaN             NaN             NaN   \n",
+       "4             NaN  Animalia      Chordata            Aves  Casuariiformes   \n",
+       "\n",
+       "        family        genus     species                      common  \n",
+       "0   Lycaenidae  Orthomiella  rantaizana  Chinese Straight-wing Blue  \n",
+       "1  Woodsiaceae      Woodsia  subcordata          Woodsia subcordata  \n",
+       "2     Laniidae       Lanius       minor          Lesser Grey Shrike  \n",
+       "3          NaN          NaN      tenuis                      Tenuis  \n",
+       "4  Casuariidae    Casuarius   casuarius          Southern Cassowary  "
+      ]
+     },
+     "execution_count": 3,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "df.head()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "<class 'pandas.core.frame.DataFrame'>\n",
+      "RangeIndex: 10092530 entries, 0 to 10092529\n",
+      "Data columns (total 17 columns):\n",
+      " #   Column            Non-Null Count     Dtype  \n",
+      "---  ------            --------------     -----  \n",
+      " 0   split             10092530 non-null  object \n",
+      " 1   treeoflife_id     10092530 non-null  object \n",
+      " 2   eol_content_id    6277374 non-null   float64\n",
+      " 3   eol_page_id       6277374 non-null   float64\n",
+      " 4   bioscan_part      1128313 non-null   float64\n",
+      " 5   bioscan_filename  1128313 non-null   object \n",
+      " 6   inat21_filename   2686843 non-null   object \n",
+      " 7   inat21_cls_name   2686843 non-null   object \n",
+      " 8   inat21_cls_num    2686843 non-null   float64\n",
+      " 9   kingdom           9831721 non-null   object \n",
+      " 10  phylum            9833317 non-null   object \n",
+      " 11  class             9813548 non-null   object \n",
+      " 12  order             9807409 non-null   object \n",
+      " 13  family            9775447 non-null   object \n",
+      " 14  genus             8908268 non-null   object \n",
+      " 15  species           8749857 non-null   object \n",
+      " 16  common            10092530 non-null  object \n",
+      "dtypes: float64(4), object(13)\n",
+      "memory usage: 1.3+ GB\n"
+     ]
+    }
+   ],
+   "source": [
+    "df.info(show_counts = True)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "The `train_small` is duplicates of `train`, so we will drop those to analyze the full training set plus val."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "df = df.loc[df.split != 'train_small']"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "<class 'pandas.core.frame.DataFrame'>\n",
+      "Index: 10092530 entries, 0 to 10092529\n",
+      "Data columns (total 17 columns):\n",
+      " #   Column            Non-Null Count     Dtype  \n",
+      "---  ------            --------------     -----  \n",
+      " 0   split             10092530 non-null  object \n",
+      " 1   treeoflife_id     10092530 non-null  object \n",
+      " 2   eol_content_id    6277374 non-null   float64\n",
+      " 3   eol_page_id       6277374 non-null   float64\n",
+      " 4   bioscan_part      1128313 non-null   float64\n",
+      " 5   bioscan_filename  1128313 non-null   object \n",
+      " 6   inat21_filename   2686843 non-null   object \n",
+      " 7   inat21_cls_name   2686843 non-null   object \n",
+      " 8   inat21_cls_num    2686843 non-null   float64\n",
+      " 9   kingdom           9831721 non-null   object \n",
+      " 10  phylum            9833317 non-null   object \n",
+      " 11  class             9813548 non-null   object \n",
+      " 12  order             9807409 non-null   object \n",
+      " 13  family            9775447 non-null   object \n",
+      " 14  genus             8908268 non-null   object \n",
+      " 15  species           8749857 non-null   object \n",
+      " 16  common            10092530 non-null  object \n",
+      "dtypes: float64(4), object(13)\n",
+      "memory usage: 1.4+ GB\n"
+     ]
+    }
+   ],
+   "source": [
+    "df.info(show_counts = True)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "`predicted-catalog` doesn't have `train_small`, hence, it's a smaller file."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Original version had 10,436,521 entries; we expected loss of about 84K from the genera with label \"unknown\".\n",
+    "We have still lost about 300K images from the original 10,092,530, but the pre-webdataset generation gained about 30K back from v3.2 (adding subspecies back in?).\n",
+    "\n",
+    ".......\n",
+    "\n",
+    "after webdataset generation:\n",
+    "\n",
+    ", but there's still another ~300K missing for this 10,065,576. It seems subspecies were not integrated back in under their species for this version, as we now have 269 less images than last time.\n",
+    "\n",
+    "Coverage for species and genus has also dropped by 269 and similar, resp."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Focus here on the difference between `predicted-catalog` and `catalog`, which is only in EOL data."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "split                      2\n",
+       "treeoflife_id       10092530\n",
+       "eol_content_id       6277374\n",
+       "eol_page_id           504018\n",
+       "bioscan_part             113\n",
+       "bioscan_filename     1128313\n",
+       "inat21_filename      2686843\n",
+       "inat21_cls_name        10000\n",
+       "inat21_cls_num         10000\n",
+       "kingdom                   12\n",
+       "phylum                   108\n",
+       "class                    347\n",
+       "order                   1495\n",
+       "family                  7973\n",
+       "genus                  73896\n",
+       "species               163741\n",
+       "common                444284\n",
+       "dtype: int64"
+      ]
+     },
+     "execution_count": 8,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "df.nunique()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "There are 504,018 unique EOL page IDs (total 6,277,374 entries), compared to the 503,589 in the webdataset (total 6,250,420 images)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 12,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "There are 26954 less entries in the webdataset.\n"
+     ]
+    }
+   ],
+   "source": [
+    "# Number that get dropped in webdataset\n",
+    "print(f\"There are {6277374 - 6250420} less entries in the webdataset.\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Notice that we have 12 unique kingdoms, which we're sticking with."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "kingdom\n",
+       "Animalia         6097759\n",
+       "Plantae          3339372\n",
+       "Fungi             359553\n",
+       "Chromista          15227\n",
+       "Protozoa           13755\n",
+       "Bacteria            3309\n",
+       "Viridiplantae       1960\n",
+       "Protista             422\n",
+       "Orthornavirae        289\n",
+       "Bamfordvirae          41\n",
+       "Archaea               25\n",
+       "Shotokuvirae           9\n",
+       "Name: count, dtype: int64"
+      ]
+     },
+     "execution_count": 9,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "df['kingdom'].value_counts()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "There is 1 more member of `Plantae` predicted, but 27K more `Animalia`."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 10,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "['kingdom', 'phylum', 'class', 'order', 'family', 'genus', 'species']"
+      ]
+     },
+     "execution_count": 10,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "taxa = list(df.columns[9:16])\n",
+    "taxa"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Check the number of images with all 7 taxonomic labels."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 11,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "<class 'pandas.core.frame.DataFrame'>\n",
+      "Index: 8482197 entries, 0 to 10092529\n",
+      "Data columns (total 7 columns):\n",
+      " #   Column   Non-Null Count    Dtype \n",
+      "---  ------   --------------    ----- \n",
+      " 0   kingdom  8482197 non-null  object\n",
+      " 1   phylum   8482197 non-null  object\n",
+      " 2   class    8482197 non-null  object\n",
+      " 3   order    8482197 non-null  object\n",
+      " 4   family   8482197 non-null  object\n",
+      " 5   genus    8482197 non-null  object\n",
+      " 6   species  8482197 non-null  object\n",
+      "dtypes: object(7)\n",
+      "memory usage: 517.7+ MB\n"
+     ]
+    }
+   ],
+   "source": [
+    "df_all_taxa = df.dropna(subset = taxa)\n",
+    "df_all_taxa[taxa].info(show_counts = True)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "We have 8,482,197 entries with full taxonomic labels, compared to  8,455,243 in the webdataset, so \n",
+    "### _**all**_ of our lost entries have all taxonomic ranks filled."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Let's add a column indicating the original data source so we can also get some stats by datasource, specifically focusing on EOL now."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 13,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Add data_source column for easier slicing\n",
+    "df.loc[df['inat21_filename'].notna(), 'data_source'] = 'iNat21'\n",
+    "df.loc[df['bioscan_filename'].notna(), 'data_source'] = 'BIOSCAN'\n",
+    "df.loc[df['eol_content_id'].notna(), 'data_source'] = 'EOL'"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "First, check their unique class values (`common`)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 14,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "440296"
+      ]
+     },
+     "execution_count": 14,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "df.loc[df['data_source'] == 'EOL', 'common'].nunique()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "EOL number of unique classes is 439,910 in the webdataset, so we do lose 386..."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Make `df_taxa` with just taxa columns (+ `common` & `data_source`) so it's smaller to process faster."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 15,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "taxa_com = list(df.columns[9:17]) # taxa + common\n",
+    "taxa_com.insert(0, 'data_source')\n",
+    "df_taxa = df[taxa_com]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 16,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/html": [
+       "<div>\n",
+       "<style scoped>\n",
+       "    .dataframe tbody tr th:only-of-type {\n",
+       "        vertical-align: middle;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe tbody tr th {\n",
+       "        vertical-align: top;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe thead th {\n",
+       "        text-align: right;\n",
+       "    }\n",
+       "</style>\n",
+       "<table border=\"1\" class=\"dataframe\">\n",
+       "  <thead>\n",
+       "    <tr style=\"text-align: right;\">\n",
+       "      <th></th>\n",
+       "      <th>data_source</th>\n",
+       "      <th>kingdom</th>\n",
+       "      <th>phylum</th>\n",
+       "      <th>class</th>\n",
+       "      <th>order</th>\n",
+       "      <th>family</th>\n",
+       "      <th>genus</th>\n",
+       "      <th>species</th>\n",
+       "      <th>common</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>EOL</td>\n",
+       "      <td>Animalia</td>\n",
+       "      <td>Arthropoda</td>\n",
+       "      <td>Insecta</td>\n",
+       "      <td>Lepidoptera</td>\n",
+       "      <td>Lycaenidae</td>\n",
+       "      <td>Orthomiella</td>\n",
+       "      <td>rantaizana</td>\n",
+       "      <td>Chinese Straight-wing Blue</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>1</th>\n",
+       "      <td>EOL</td>\n",
+       "      <td>Plantae</td>\n",
+       "      <td>Tracheophyta</td>\n",
+       "      <td>Polypodiopsida</td>\n",
+       "      <td>Polypodiales</td>\n",
+       "      <td>Woodsiaceae</td>\n",
+       "      <td>Woodsia</td>\n",
+       "      <td>subcordata</td>\n",
+       "      <td>Woodsia subcordata</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>2</th>\n",
+       "      <td>EOL</td>\n",
+       "      <td>Animalia</td>\n",
+       "      <td>Chordata</td>\n",
+       "      <td>Aves</td>\n",
+       "      <td>Passeriformes</td>\n",
+       "      <td>Laniidae</td>\n",
+       "      <td>Lanius</td>\n",
+       "      <td>minor</td>\n",
+       "      <td>Lesser Grey Shrike</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>3</th>\n",
+       "      <td>EOL</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>NaN</td>\n",
+       "      <td>tenuis</td>\n",
+       "      <td>Tenuis</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>4</th>\n",
+       "      <td>EOL</td>\n",
+       "      <td>Animalia</td>\n",
+       "      <td>Chordata</td>\n",
+       "      <td>Aves</td>\n",
+       "      <td>Casuariiformes</td>\n",
+       "      <td>Casuariidae</td>\n",
+       "      <td>Casuarius</td>\n",
+       "      <td>casuarius</td>\n",
+       "      <td>Southern Cassowary</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ],
+      "text/plain": [
+       "  data_source   kingdom        phylum           class           order   \n",
+       "0         EOL  Animalia    Arthropoda         Insecta     Lepidoptera  \\\n",
+       "1         EOL   Plantae  Tracheophyta  Polypodiopsida    Polypodiales   \n",
+       "2         EOL  Animalia      Chordata            Aves   Passeriformes   \n",
+       "3         EOL       NaN           NaN             NaN             NaN   \n",
+       "4         EOL  Animalia      Chordata            Aves  Casuariiformes   \n",
+       "\n",
+       "        family        genus     species                      common  \n",
+       "0   Lycaenidae  Orthomiella  rantaizana  Chinese Straight-wing Blue  \n",
+       "1  Woodsiaceae      Woodsia  subcordata          Woodsia subcordata  \n",
+       "2     Laniidae       Lanius       minor          Lesser Grey Shrike  \n",
+       "3          NaN          NaN      tenuis                      Tenuis  \n",
+       "4  Casuariidae    Casuarius   casuarius          Southern Cassowary  "
+      ]
+     },
+     "execution_count": 16,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "df_taxa.head()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Let's look a little closer at EOL."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 17,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "inat21_df = df_taxa.loc[df_taxa.data_source == 'iNat21']\n",
+    "bioscan_df = df_taxa.loc[df_taxa.data_source == 'BIOSCAN']\n",
+    "eol_df = df_taxa.loc[df_taxa.data_source == 'EOL']"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### EOL"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 18,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "<class 'pandas.core.frame.DataFrame'>\n",
+      "Index: 6277374 entries, 0 to 6277373\n",
+      "Data columns (total 9 columns):\n",
+      " #   Column       Non-Null Count    Dtype \n",
+      "---  ------       --------------    ----- \n",
+      " 0   data_source  6277374 non-null  object\n",
+      " 1   kingdom      6016565 non-null  object\n",
+      " 2   phylum       6018161 non-null  object\n",
+      " 3   class        5998392 non-null  object\n",
+      " 4   order        5992253 non-null  object\n",
+      " 5   family       5975682 non-null  object\n",
+      " 6   genus        5967267 non-null  object\n",
+      " 7   species      5978567 non-null  object\n",
+      " 8   common       6277374 non-null  object\n",
+      "dtypes: object(9)\n",
+      "memory usage: 478.9+ MB\n"
+     ]
+    }
+   ],
+   "source": [
+    "eol_df.info(show_counts = True)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 19,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "data_source         1\n",
+       "kingdom            12\n",
+       "phylum            108\n",
+       "class             344\n",
+       "order            1489\n",
+       "family           7929\n",
+       "genus           73104\n",
+       "species        162275\n",
+       "common         440296\n",
+       "dtype: int64"
+      ]
+     },
+     "execution_count": 19,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "eol_df.nunique()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "It seems the webdataset loses 5 families, 39 genera, and 84 species. As noted above, 386 common labels are lost."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "There are no missing species that get lost, as observed above, all lost entries have all taxonomic labels."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 22,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/var/folders/nv/f0fq1p1n1_3b11x579py_0q80000gq/T/ipykernel_21854/52183210.py:2: SettingWithCopyWarning: \n",
+      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
+      "Try using .loc[row_indexer,col_indexer] = value instead\n",
+      "\n",
+      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
+      "  eol_df['duplicate'] = eol_df.duplicated(subset = taxa, keep = 'first')\n"
+     ]
+    }
+   ],
+   "source": [
+    "#number of unique 7-tuples in EOL\n",
+    "eol_df['duplicate'] = eol_df.duplicated(subset = taxa, keep = 'first')\n",
+    "eol_df_unique_taxa = eol_df.loc[~eol_df['duplicate']]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 23,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "<class 'pandas.core.frame.DataFrame'>\n",
+      "Index: 449310 entries, 0 to 6277357\n",
+      "Data columns (total 10 columns):\n",
+      " #   Column       Non-Null Count   Dtype \n",
+      "---  ------       --------------   ----- \n",
+      " 0   data_source  449310 non-null  object\n",
+      " 1   kingdom      420156 non-null  object\n",
+      " 2   phylum       420619 non-null  object\n",
+      " 3   class        419503 non-null  object\n",
+      " 4   order        419849 non-null  object\n",
+      " 5   family       419720 non-null  object\n",
+      " 6   genus        422707 non-null  object\n",
+      " 7   species      409745 non-null  object\n",
+      " 8   common       449310 non-null  object\n",
+      " 9   duplicate    449310 non-null  bool  \n",
+      "dtypes: bool(1), object(9)\n",
+      "memory usage: 34.7+ MB\n"
+     ]
+    }
+   ],
+   "source": [
+    "eol_df_unique_taxa.info(show_counts = True)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "400 unique taxa are lost."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Let's check the `species` length in EOL as well, we know there are some that have genus-species. And others with hybrids that get VERY long.\n",
+    "\n",
+    "Wonder if this is where we lose some?\n",
+    "Quick check in the catalog notebook shows there are still plenty with full taxa there."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "jupytext": {
+   "formats": "ipynb,py:percent"
+  },
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.11.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

notebooks/ToL_predicted-catalog_EDA.py

@@ -0,0 +1,162 @@
+# ---
+# jupyter:
+#   jupytext:
+#     formats: ipynb,py:percent
+#     text_representation:
+#       extension: .py
+#       format_name: percent
+#       format_version: '1.3'
+#       jupytext_version: 1.15.2
+#   kernelspec:
+#     display_name: Python 3 (ipykernel)
+#     language: python
+#     name: python3
+# ---
+
+# %%
+import pandas as pd
+import seaborn as sns
+
+sns.set_style("whitegrid")
+sns.set(rc = {'figure.figsize': (10,10)})
+
+# %%
+df = pd.read_csv("../data/predicted-catalog.csv")
+
+# %%
+df.head()
+
+# %%
+df.info(show_counts = True)
+
+# %% [markdown]
+# The `train_small` is duplicates of `train`, so we will drop those to analyze the full training set plus val.
+
+# %%
+df = df.loc[df.split != 'train_small']
+
+# %%
+df.info(show_counts = True)
+
+# %% [markdown]
+# `predicted-catalog` doesn't have `train_small`, hence, it's a smaller file.
+
+# %% [markdown]
+# Original version had 10,436,521 entries; we expected loss of about 84K from the genera with label "unknown".
+# We have still lost about 300K images from the original 10,092,530, but the pre-webdataset generation gained about 30K back from v3.2 (adding subspecies back in?).
+#
+# .......
+#
+# after webdataset generation:
+#
+# , but there's still another ~300K missing for this 10,065,576. It seems subspecies were not integrated back in under their species for this version, as we now have 269 less images than last time.
+#
+# Coverage for species and genus has also dropped by 269 and similar, resp.
+
+# %% [markdown]
+# ### Focus here on the difference between `predicted-catalog` and `catalog`, which is only in EOL data.
+
+# %%
+df.nunique()
+
+# %% [markdown]
+# There are 504,018 unique EOL page IDs (total 6,277,374 entries), compared to the 503,589 in the webdataset (total 6,250,420 images).
+
+# %%
+# Number that get dropped in webdataset
+print(f"There are {6277374 - 6250420} less entries in the webdataset.")
+
+# %% [markdown]
+# Notice that we have 12 unique kingdoms, which we're sticking with.
+
+# %%
+df['kingdom'].value_counts()
+
+# %% [markdown]
+# There is 1 more member of `Plantae` predicted, but 27K more `Animalia`.
+
+# %%
+taxa = list(df.columns[9:16])
+taxa
+
+# %% [markdown]
+# Check the number of images with all 7 taxonomic labels.
+
+# %%
+df_all_taxa = df.dropna(subset = taxa)
+df_all_taxa[taxa].info(show_counts = True)
+
+# %% [markdown]
+# We have 8,482,197 entries with full taxonomic labels, compared to  8,455,243 in the webdataset, so 
+# ### _**all**_ of our lost entries have all taxonomic ranks filled.
+
+# %% [markdown]
+# Let's add a column indicating the original data source so we can also get some stats by datasource, specifically focusing on EOL now.
+
+# %%
+# Add data_source column for easier slicing
+df.loc[df['inat21_filename'].notna(), 'data_source'] = 'iNat21'
+df.loc[df['bioscan_filename'].notna(), 'data_source'] = 'BIOSCAN'
+df.loc[df['eol_content_id'].notna(), 'data_source'] = 'EOL'
+
+# %% [markdown]
+# First, check their unique class values (`common`).
+
+# %%
+df.loc[df['data_source'] == 'EOL', 'common'].nunique()
+
+# %% [markdown]
+# EOL number of unique classes is 439,910 in the webdataset, so we do lose 386...
+
+# %% [markdown]
+# Make `df_taxa` with just taxa columns (+ `common` & `data_source`) so it's smaller to process faster.
+
+# %%
+taxa_com = list(df.columns[9:17]) # taxa + common
+taxa_com.insert(0, 'data_source')
+df_taxa = df[taxa_com]
+
+# %%
+df_taxa.head()
+
+# %% [markdown]
+# Let's look a little closer at EOL.
+
+# %%
+inat21_df = df_taxa.loc[df_taxa.data_source == 'iNat21']
+bioscan_df = df_taxa.loc[df_taxa.data_source == 'BIOSCAN']
+eol_df = df_taxa.loc[df_taxa.data_source == 'EOL']
+
+# %% [markdown]
+# ### EOL
+
+# %%
+eol_df.info(show_counts = True)
+
+# %%
+eol_df.nunique()
+
+# %% [markdown]
+# It seems the webdataset loses 5 families, 39 genera, and 84 species. As noted above, 386 common labels are lost.
+
+# %% [markdown]
+# There are no missing species that get lost, as observed above, all lost entries have all taxonomic labels.
+
+# %%
+#number of unique 7-tuples in EOL
+eol_df['duplicate'] = eol_df.duplicated(subset = taxa, keep = 'first')
+eol_df_unique_taxa = eol_df.loc[~eol_df['duplicate']]
+
+# %%
+eol_df_unique_taxa.info(show_counts = True)
+
+# %% [markdown]
+# 400 unique taxa are lost.
+
+# %% [markdown]
+# Let's check the `species` length in EOL as well, we know there are some that have genus-species. And others with hybrids that get VERY long.
+#
+# Wonder if this is where we lose some?
+# Quick check in the catalog notebook shows there are still plenty with full taxa there.
+
+# %%