File size: 6,545 Bytes
93aebb3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
50186b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
---
dataset_info:
  features:
  - name: instruction
    dtype: string
  - name: input
    dtype: float64
  - name: output
    dtype: string
  splits:
  - name: train
    num_bytes: 743535
    num_examples: 800
  - name: validation
    num_bytes: 87504
    num_examples: 100
  - name: test
    num_bytes: 92879
    num_examples: 100
  download_size: 360936
  dataset_size: 923918
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
---
# Dataset Analysis Report

## Dataset Format

The dataset is provided in **Parquet** format with the following structure:

| Field         | Description                                     |
| ------------- | ----------------------------------------------- |
| `instruction` | Instruction or question provided to the model   |
| `input`       | Additional input or contextual information      |
| `output`      | Expected response generated for the instruction |

## Dataset Statistics

| Metric                      | Result |
| --------------------------- | -----: |
| Total rows                  |  1,000 |
| Valid rows                  |  1,000 |
| Invalid rows                |      0 |
| Columns                     |      3 |
| Expected fields per row     |      3 |
| Rows with missing fields    |      0 |
| Rows with unexpected fields |      0 |

### Fields

```text
['input', 'instruction', 'output']
```

## Duplicate Analysis

| Check                            | Result |
| -------------------------------- | -----: |
| Exact duplicate rows             |      0 |
| Duplicate groups                 |      0 |
| Duplicate instructions           |    215 |
| Duplicate responses              |      2 |
| Near-duplicate instruction pairs |     69 |

No exact duplicate rows were detected. However, **215 duplicate instructions**, **2 duplicate responses**, and **69 near-duplicate instruction pairs** were identified.

## Quality Analysis

The automated analysis identified the following quality flags:

| Quality Issue                    |  Rows | Percentage |
| -------------------------------- | ----: | ---------: |
| English words detected           | 1,000 |    100.00% |
| Missing values                   | 1,000 |    100.00% |
| Low Devanagari ratio             |    93 |      9.30% |
| HTML noise                       |    21 |      2.10% |
| URLs detected                    |    16 |      1.60% |
| Potentially incomplete responses |    10 |      1.00% |

> **Important:** The `missing_values` check flagged all 1,000 rows. This means the automated analysis identified at least one field as missing according to its configured missing-value rules. The result should be manually reviewed, particularly for the `input` field, because an intentionally empty `input` may be a valid characteristic of an instruction/input/output dataset rather than an actual data-quality error.

## Analysis Summary

```text
Total rows          : 1,000
Flagged rows        : 1,000
Flagged percentage  : 100.00%

Data status         : ANALYSIS ONLY
Cleaning performed  : NO
```

All 1,000 rows were flagged by at least one automated quality check. Being flagged does **not necessarily mean that every row is unusable**; it means that every row requires review according to the configured analysis rules.

## Overall Dataset Quality

| Metric                | Result         |
| --------------------- | -------------- |
| Overall Quality Score | 66.40 / 100    |
| Quality Grade         | D              |
| Dataset Status        | NEEDS CLEANING |

### Strengths

* No exact duplicate rows detected.
* No possible PII detected.
* All rows follow the expected `instruction` / `input` / `output` schema.
* All 1,000 rows are valid records according to the analysis.

### Identified Issues

* 215 duplicate instructions detected.
* 2 duplicate responses detected.
* English words were detected in all 1,000 rows.
* 21 rows contain HTML noise.
* 16 rows contain URLs.
* 10 responses were potentially incomplete.
* 93 rows have a low Devanagari ratio.
* The automated analysis flagged missing values in all 1,000 rows.
* 69 near-duplicate instruction pairs were identified.

## Recommendation

Significant cleaning and manual review are recommended before using this dataset for model training.

Priority should be given to reviewing:

1. Duplicate and near-duplicate instructions.
2. The 100% missing-value flag, especially whether empty `input` fields are intentional.
3. English-word detection across the dataset.
4. Low-Devanagari-ratio records.
5. HTML noise and URLs.
6. Potentially incomplete responses.

## Analysis Method

The dataset was analyzed without modifying the original data.

The analysis included:

* Dataset structure and schema validation
* Missing-value detection
* Exact duplicate detection
* Duplicate instruction detection
* Duplicate response detection
* Near-duplicate instruction detection
* English-word detection
* Devanagari-ratio analysis
* HTML-noise detection
* URL detection
* Potential incomplete-response detection
* Possible PII detection
* Overall dataset quality scoring and grading

### Cleaning Status

**No cleaning, deletion, modification, or automatic correction was performed during this analysis.**

The analysis script only identified and flagged potential quality issues for subsequent manual review and cleaning.

## Analysis Outputs

The analysis generated the following files:

```text
analysis_report(train).json
flagged_rows(train).jsonl
```

`analysis_report(train).json` contains the complete dataset-level analysis results, including dataset structure, schema consistency, duplicate statistics, language analysis, quality flags, incomplete-response analysis, and the overall quality score.

`flagged_rows(train).jsonl` contains the individual records flagged by the automated quality checks for manual review.

## Final Assessment

**Overall Quality Score:** 66.40 / 100
**Grade:** D
**Status:** NEEDS CLEANING

The dataset has a valid and consistent three-column `instruction` / `input` / `output` structure with 1,000 valid rows and no exact duplicate rows. However, the automated analysis identified duplicate instructions, near-duplicate instructions, duplicate responses, English-word occurrences, missing-value flags, low Devanagari ratios, HTML noise, URLs, and potentially incomplete responses.

The dataset should therefore undergo **manual review and targeted cleaning** before being considered ready for model training.