Add stats report (README.md)

README.md

@@ -1,29 +1,141 @@
----
-dataset_info:
-  features:
-  - name: column
-    dtype: string
-  - name: feature_type
-    dtype: string
-  - name: kind
-    dtype: string
-  - name: n_total
-    dtype: int64
-  - name: n_null
-    dtype: int64
-  - name: null_pct
-    dtype: float64
-  - name: stats_json
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 1815
-    num_examples: 3
-  download_size: 7113
-  dataset_size: 1815
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
----
+# Stats for `stanfordnlp/sst2`
+
+Generated by [`dataset-stats.py`](https://github.com/tcondello/uv-scripts-colab/blob/main/recipes/dataset-stats.py) over the `train` split.
+
+- **Total rows in split:** 67,349
+- **Rows profiled:** 5,000
+- **Columns:** 3
+
+## Column overview
+
+| column | type | kind | null % | highlights |
+|---|---|---|---:|---|
+| `idx` | `Value(int32)` | numeric | 0.0% | min=0.00 · p50=2,499.50 · max=4,999.00 · distinct=5000 |
+| `sentence` | `Value(string)` | string | 0.0% | distinct=4,999 · len p50=39 · max=255 |
+| `label` | `ClassLabel` | class_label | 0.0% | `positive`=2758 · `negative`=2242 |
+
+## Per-column detail
+
+### `idx`
+
+- **type:** `Value(int32)`  •  **kind:** numeric  •  **null:** 0 / 5,000 (0.0%)
+
+```json
+{
+  "kind": "numeric",
+  "n": 5000,
+  "min": 0.0,
+  "p1": 49.99,
+  "p25": 1249.75,
+  "p50": 2499.5,
+  "p75": 3749.25,
+  "p99": 4949.01,
+  "max": 4999.0,
+  "mean": 2499.5,
+  "std": 1443.3756441065507,
+  "distinct": 5000
+}
+```
+
+### `sentence`
+
+- **type:** `Value(string)`  •  **kind:** string  •  **null:** 0 / 5,000 (0.0%)
+
+```json
+{
+  "kind": "string",
+  "n": 5000,
+  "distinct": 4999,
+  "len_min": 2,
+  "len_p50": 39,
+  "len_mean": 53.479,
+  "len_p99": 194,
+  "len_max": 255,
+  "top_values": [
+    {
+      "value": "unfunny ",
+      "count": 2
+    },
+    {
+      "value": "hide new secretions from the parental units ",
+      "count": 1
+    },
+    {
+      "value": "contains no wit , only labored gags ",
+      "count": 1
+    },
+    {
+      "value": "that loves its characters and communicates something rather beautiful about human nature ",
+      "count": 1
+    },
+    {
+      "value": "remains utterly satisfied to remain the same throughout ",
+      "count": 1
+    },
+    {
+      "value": "on the worst revenge-of-the-nerds clichés the filmmakers could dredge up ",
+      "count": 1
+    },
+    {
+      "value": "that 's far too tragic to merit such superficial treatment ",
+      "count": 1
+    },
+    {
+      "value": "demonstrates that the director of such hollywood blockbusters as patriot games can still turn out a ",
+      "count": 1
+    },
+    {
+      "value": "of saucy ",
+      "count": 1
+    },
+    {
+      "value": "a depressed fifteen-year-old 's suicidal poetry ",
+      "count": 1
+    }
+  ],
+  "samples": {
+    "shortest": "( ",
+    "median": "piece , a model of menacing atmosphere ",
+    "longest": "director george hickenlooper has had some success with documentaries , but here his sense of story and his juvenile camera movements smack of a film school undergrad , and his maudlin ending might not"
+  }
+}
+```
+
+### `label`
+
+- **type:** `ClassLabel`  •  **kind:** class_label  •  **null:** 0 / 5,000 (0.0%)
+
+```json
+{
+  "kind": "class_label",
+  "n": 5000,
+  "num_classes": 2,
+  "value_counts": [
+    {
+      "id": 1,
+      "name": "positive",
+      "count": 2758,
+      "pct": 55.16
+    },
+    {
+      "id": 0,
+      "name": "negative",
+      "count": 2242,
+      "pct": 44.84
+    }
+  ]
+}
+```
+
+## Reproduce
+
+```bash
+INPUT_DATASET=stanfordnlp/sst2 \
+OUTPUT_DATASET=Tim-Pinecone/sst2-stats \
+SPLIT=train LIMIT=5000 \
+  bin/colab-hf-run recipes/dataset-stats.py
+```
+
+## Machine-readable
+
+Per-column rows live in the dataset's `train` split: one row per column with `column`, `feature_type`, `kind`, `n_total`, `n_null`, `null_pct`, and `stats_json` (full per-kind detail).