def find_by(df: pd.DataFrame, uniques: List[Union[str, List[str]]]) -> Result:
"""Find equal items rows in `df` by `uniques`. I.e. if two items have the same
uniques's element value, they are considered duplicates.
Args:
uniques: list containing columns and list of columns to identify duplicates.
List of columns means that all list columns values should be equal.
Returns:
Any duplicates
"""
result = Result("Duplicates")
result.items_count = len(df)
df = df.dropna(subset=list(set(flatten(uniques))), how="all")
for columns in uniques:
mask = columns if isinstance(columns, list) else [columns]
duplicates = df[df.duplicated(columns, keep=False)][mask]
if duplicates.empty:
continue
errors = {}
grouped = duplicates.groupby(columns)
for _, d in grouped:
msgs = [f"'{d[c].iloc[0]}' `{c}`" for c in mask]
errors[f"same {', '.join(msgs)}"] = list(d.index)
result.add_error(
f"{', '.join(mask)} contains {len(grouped)} duplicated value(s)",
errors=errors,
)
return result
def compare_names_for_same_urls(
source_df: pd.DataFrame,
target_df: pd.DataFrame,
tagged_fields: Dict[str, List[str]],
):
"""For each pair of items that have the same `product_url_field` tagged field,
compare `name_field` field"""
result = Result("Compare Names Per Url")
url_field = tagged_fields.get("product_url_field")
if not url_field:
result.add_info("product_url_field tag is not set")
return result
url_field = url_field[0]
name_field = tagged_fields.get("name_field")
diff_names_count = 0
if not name_field:
result.add_info("name_field tag is not set")
return result
name_field = name_field[0]
if any([
name_field not in source_df.columns.values,
name_field not in target_df.columns.values,
]):
return
same_urls = source_df[(source_df[url_field].isin(
target_df[url_field].values))][url_field]
detailed_messages = []
for url in same_urls:
if url.strip() != "nan":
source_name = source_df[source_df[url_field] ==
url][name_field].iloc[0]
target_name = target_df[target_df[url_field] ==
url][name_field].iloc[0]
if (source_name != target_name and source_name.strip() != "nan"
and target_name.strip() != "nan"):
diff_names_count += 1
source_key = source_df[source_df[url_field] ==
url]["_key"].iloc[0]
target_key = target_df[target_df[url_field] ==
url]["_key"].iloc[0]
msg = (
f"different names for url: {url}\nsource name is {source_name} "
f"for {source_key}\ntarget name is {target_name} for {target_key}"
)
detailed_messages.append(msg)
res = f"{len(same_urls)} checked, {diff_names_count} errors"
if detailed_messages:
result.add_error(res, detailed="\n".join(detailed_messages))
else:
result.add_info(res)
return result
def check_outcome(job: Job) -> Result:
state = api.get_job_state(job)
reason = api.get_job_close_reason(job)
result = Result("Job Outcome")
if state != "finished" or reason != "finished":
result.add_error(f"Job has '{state}' state, '{reason}' close reason")
return result
def anomalies(target: str, sample: List[str]) -> Result:
"""Find fields with significant deviation. Significant means `dev > 2 * std()`
Args:
target: where to look for anomalies
sample: a list of jobs keys to infer metadata from
Returns:
A Result with a dataframe of significant deviations
"""
result = Result("Anomalies")
raw_stats = [job.items.stats() for job in api.get_jobs(sample + [target])]
counts = (pd.DataFrame(
rs.get("counts") for rs in raw_stats).fillna(0).drop(columns="_type"))
items_len = [rs["totals"]["input_values"] for rs in raw_stats]
stats = counts.apply(lambda x: x / items_len)
stats.index = sample + [target]
stats.rename(index={target: "target"}, inplace=True)
stats.loc["mean"] = stats.loc[sample].mean()
stats.loc["std"] = stats.loc[sample].std()
stats = stats.T
stats["target deviation"] = stats["target"] - stats["mean"]
devs = stats[(stats["target deviation"].abs() > 2 * stats["std"])]
devs.name = "Anomalies"
if not devs.empty:
result.add_error(
f"{len(devs.index)} field(s) with significant coverage deviation")
result.stats = [devs]
return result
def check_tags(source_columns, target_columns, tags):
result = Result("Tags")
found_tags = list(tags)
if found_tags:
result.add_info(", ".join(found_tags))
tagged_fields = []
for tag in tags:
tagged_fields.extend(tags[tag])
missing_in_source = sorted(set(tagged_fields) - set(source_columns))
if missing_in_source:
result.add_error(
f"{str(missing_in_source)[1:-1]} field(s) was not found in "
"source, but specified in schema")
if target_columns.size > 0:
missing_in_target = sorted(set(tagged_fields) - set(target_columns))
if missing_in_target:
result.add_error(
f"{str(missing_in_target)[1:-1]} field(s) was not found "
"in target, but specified in schema")
if result.errors:
result.add_error("Skipping tag rules")
return result
def check_response_ratio(job: Job) -> Result:
requests_number = api.get_requests_count(job)
items_count = api.get_items_count(job)
result = Result("Responses Per Item Ratio")
result.add_info(
f"Number of responses / Number of scraped items - "
f"{round(requests_number / items_count, 2)}"
)
return result
def compare_was_now(df: pd.DataFrame, tagged_fields: Dict[str, List[str]]):
"""Compare price_was and price_now tagged fields"""
price_was_fields = tagged_fields.get("product_price_was_field")
price_fields = tagged_fields.get("product_price_field")
items_number = len(df.index)
result = Result("Compare Price Was And Now")
if (price_was_fields and price_was_fields[0] in df.columns and price_fields
and price_fields[0] in df.columns):
price_field = price_fields[0]
price_was_field = price_was_fields[0]
prices = df.copy()
prices[price_was_field] = prices[price_was_field].astype(float)
prices[price_field] = prices[price_field].astype(float)
df_prices_less = pd.DataFrame(
prices[prices[price_was_field] < prices[price_field]],
columns=["_key", price_was_field, price_field],
)
price_less_percent = "{:.2%}".format(
len(df_prices_less) / items_number)
if not df_prices_less.empty:
error = f"Past price is less than current for {len(df_prices_less)} items"
result.add_error(
f"{price_less_percent} ({len(df_prices_less)}) of "
f"items with {price_was_field} < {price_field}",
detailed=f"{error}:\n{list(df_prices_less['_key'])}",
)
df_prices_equals = pd.DataFrame(
prices[prices[price_was_field] == prices[price_field]],
columns=["_key", price_was_field, price_field],
)
price_equal_percent = "{:.2%}".format(
len(df_prices_equals) / items_number)
if not df_prices_equals.empty:
result.add_warning(
(f"{price_equal_percent} ({len(df_prices_equals)}) "
f"of items with {price_was_field} = {price_field}"),
detailed=(f"Prices equal for {len(df_prices_equals)} items:\n"
f"{list(df_prices_equals['_key'])}"),
)
result.err_items_count = len(df_prices_equals) + len(df_prices_less)
result.items_count = len(df.index)
else:
result.add_info(
"product_price_field or product_price_was_field tags were not "
"found in schema")
return result
def compare_was_now(df: pd.DataFrame, tagged_fields: TaggedFields):
"""Compare price_was and price_now tagged fields"""
price_was_fields = tagged_fields.get("product_price_was_field")
price_fields = tagged_fields.get("product_price_field")
items_number = len(df.index)
result = Result("Compare Price Was And Now")
if not price_was_fields or not price_fields:
result.add_info(Outcome.SKIPPED)
return result
price_field = price_fields[0]
price_was_field = price_was_fields[0]
prices = df.copy()
prices[price_was_field] = prices[price_was_field].astype(float)
prices[price_field] = prices[price_field].astype(float)
df_prices_less = pd.DataFrame(
prices[prices[price_was_field] < prices[price_field]],
columns=[price_was_field, price_field],
)
price_less_percent = "{:.2%}".format(len(df_prices_less) / items_number)
if not df_prices_less.empty:
error = f"Past price is less than current for {len(df_prices_less)} items"
result.add_error(
f"{price_less_percent} ({len(df_prices_less)}) of "
f"items with {price_was_field} < {price_field}",
detailed=f"{error}:\n{list(df_prices_less.index)}",
)
df_prices_equals = pd.DataFrame(
prices[prices[price_was_field] == prices[price_field]],
columns=[price_was_field, price_field],
)
price_equal_percent = "{:.2%}".format(len(df_prices_equals) / items_number)
if not df_prices_equals.empty:
result.add_warning(
(
f"{price_equal_percent} ({len(df_prices_equals)}) "
f"of items with {price_was_field} = {price_field}"
),
detailed=(
f"Prices equal for {len(df_prices_equals)} items:\n"
f"{list(df_prices_equals.index)}"
),
)
result.err_items_count = len(df_prices_equals) + len(df_prices_less)
result.items_count = len(df.index)
return result
def compare_spider_names(source_job: Job, target_job: Job) -> Result:
s_name = source_job.metadata.get("spider")
t_name = target_job.metadata.get("spider")
result = Result("Spider Names")
if s_name != t_name:
result.add_warning(
f"{source_job.key} spider is {s_name}, {target_job.key} spider is {t_name}"
)
return result
def compare_spider_names(source_job: Job, target_job: Job) -> Result:
name1 = source_job.metadata.get("spider")
name2 = target_job.metadata.get("spider")
result = Result("Spider Names")
if name1 != name2:
result.add_warning(
f"{source_job.key} spider is {name1}, {target_job.key} spider is {name2}"
)
return result
def get_coverage_per_category(df: pd.DataFrame,
tagged_fields: Dict[str, List[str]]):
result = Result("Coverage For Scraped Categories")
category_fields = tagged_fields.get("category", [])
for f in category_fields:
value_counts = df[f].value_counts()
result.add_info(f"{len(value_counts)} categories in '{f}'",
stats=value_counts)
return result
def check_errors(source_job: Job) -> Result:
source_errs = api.get_errors_count(source_job)
result = Result("Job Errors")
if not source_errs:
return result
errors_url = "{}/{}/log?filterType=error&filterAndHigher"
result.add_error(
f"{source_errs} error(s) - {errors_url.format(SH_URL, source_job.key)}"
)
return result
def compare_prices_for_same_urls(source_df: pd.DataFrame,
target_df: pd.DataFrame,
tagged_fields: TaggedFields) -> Result:
"""For each pair of items that have the same `product_url_field` tagged field,
compare `product_price_field` field
Returns:
A result containing pairs of items from `source_df` and `target_df`
which `product_price_field` differ.
"""
result = Result("Compare Prices For Same Urls")
url_field_list: Optional[List[str]] = tagged_fields.get(
"product_url_field")
if not url_field_list:
result.outcome = Outcome.SKIPPED
return result
url_field = url_field_list[0]
source_df = source_df.dropna(subset=[url_field])
target_df = target_df.dropna(subset=[url_field])
same_urls = source_df[(source_df[url_field].isin(
target_df[url_field].values))][url_field]
price_fields = tagged_fields.get("product_price_field")
if not price_fields:
result.add_info("product_price_field tag is not set")
else:
price_field = price_fields[0]
detailed_messages = []
for url in same_urls:
if url.strip() != "nan":
source_price = source_df[source_df[url_field] ==
url][price_field].iloc[0]
target_price = target_df[target_df[url_field] ==
url][price_field].iloc[0]
if (is_number(source_price) and is_number(target_price)
and ratio_diff(source_price, target_price) > 0.1):
source_key = source_df[source_df[url_field] ==
url].index[0]
target_key = target_df[target_df[url_field] ==
url].index[0]
msg = (
f"different prices for url: {url}\nsource price is {source_price} "
f"for {source_key}\ntarget price is {target_price} for {target_key}"
)
detailed_messages.append(msg)
res = f"{len(same_urls)} checked, {len(detailed_messages)} errors"
if detailed_messages:
result.add_error(res, detailed="\n".join(detailed_messages))
else:
result.add_info(res)
return result
def get_difference(
source_df: pd.DataFrame,
target_df: pd.DataFrame,
category_names: List[str],
err_thr: float = 0.2,
warn_thr: float = 0.1,
) -> Result:
"""Find and show differences between categories coverage, including `nan` values.
Coverage means value counts divided on total size.
Args:
source_df: a data you want to compare
target_df: a data you want to compare with
category_names: list of columns which values to compare
err_thr: sets error threshold
warn_thr: warning threshold
Returns:
A result instance with messages containing significant difference defined by
thresholds, a dataframe showing all normalized value counts in percents and
a series containing significant difference.
"""
source_key = "source"
target_key = "target"
result = Result("Category Coverage Difference")
for c in category_names:
cats = (
pd.DataFrame(
{
source_key: source_df[c].value_counts(dropna=False, normalize=True),
target_key: target_df[c].value_counts(dropna=False, normalize=True),
}
)
.fillna(0)
.sort_values(by=[source_key, target_key], kind="mergesort")
)
cats.name = f"Coverage for {c}"
result.stats.append(cats)
cat_difs = cats[source_key] - cats[target_key]
cat_difs = cat_difs[cat_difs.abs() > warn_thr]
cat_difs.name = f"Coverage difference more than {warn_thr:.0%} for {c}"
if not cat_difs.empty:
result.stats.append(cat_difs)
errs = cat_difs[cat_difs.abs() > err_thr]
if not errs.empty:
result.add_warning(
f"The difference is greater than {err_thr:.0%} for {len(errs)} value(s) of {c}"
)
if not category_names:
result.outcome = Outcome.SKIPPED
return result
def check_items(df: pd.DataFrame, tagged_fields: Dict[str,
List[str]]) -> Result:
"""Check for items with the same name and url"""
name_fields = tagged_fields.get("name_field")
url_fields = tagged_fields.get("product_url_field")
result = Result("Duplicated Items")
if not name_fields or not url_fields:
result.add_info(
"'name_field' and 'product_url_field' tags were not found in schema"
)
else:
result.items_count = len(df)
errors = {}
name_field = name_fields[0]
url_field = url_fields[0]
df = df[[name_field, url_field, "_key"]]
duplicates = df[df[[name_field, url_field]].duplicated(keep=False)]
if duplicates.empty:
return result
result.err_items_count = len(duplicates)
for _, d in duplicates.groupby([name_field, url_field]):
msg = (
f"same '{d[name_field].iloc[0]}' name and '{d[url_field].iloc[0]}' url"
)
errors[msg] = list(d["_key"])
result.add_error(
f"{len(duplicates)} duplicate(s) with same name and url",
errors=errors)
return result
def check_uniqueness(df: pd.DataFrame,
tagged_fields: Dict[str, List[str]]) -> Result:
"""Verify if each item field tagged with `unique` is unique.
Returns:
A result containing field names and keys for non unique items
"""
unique_fields = tagged_fields.get("unique", [])
result = Result("Uniqueness")
if not unique_fields:
result.add_info("'unique' tag was not found in schema")
return result
err_keys = set()
for field in unique_fields:
result.items_count = df[field].count()
duplicates = df[df[field].duplicated(keep=False)][[field, "_key"]]
errors = {}
for _, d in duplicates.groupby([field]):
keys = list(d["_key"])
msg = f"same '{d[field].iloc[0]}' {field}"
errors[msg] = keys
err_keys = err_keys.union(keys)
if not duplicates.empty:
result.add_error(
f"'{field}' contains {len(duplicates[field].unique())} duplicated value(s)",
errors=errors,
)
result.err_items_count = len(err_keys)
return result
def get_categories(df: pd.DataFrame, max_uniques: int = 10) -> Result:
"""Find category columns. A category column is the column which holds a limited number
of possible values, including `NAN`.
Args:
df: data
max_uniques: filter which determines which columns to use. Only columns with
the number of unique values less than or equal to `max_uniques` are category columns.
Returns:
A result with stats containing value counts of categorical columns.
"""
result = Result("Categories")
columns = find_likely_cats(df, max_uniques)
result.stats = [
value_counts
for value_counts in tqdm(
map(lambda c: df[c].value_counts(dropna=False), columns),
desc="Finding categories",
total=len(columns),
)
if len(value_counts) <= max_uniques
]
if not result.stats:
result.add_info("Categories were not found")
return result
result.add_info(f"{len(result.stats)} category field(s)")
result.outcome = Outcome.INFO
return result
def find_by_unique(df: pd.DataFrame, tagged_fields: TaggedFields) -> Result:
"""Verify if each item field tagged with `unique` is unique.
Returns:
A result containing field names and keys for non unique items
"""
unique_fields = tagged_fields.get("unique", [])
result = Result("Duplicates By **unique** Tag")
if not unique_fields:
result.add_info(Outcome.SKIPPED)
return result
err_keys = set()
for field in unique_fields:
result.items_count = df[field].count()
duplicates = df[df.duplicated(field, keep=False)][[field]]
errors = {}
for _, d in duplicates.groupby([field]):
keys = list(d.index)
msg = f"same '{d[field].iloc[0]}' `{field}`"
errors[msg] = keys
err_keys = err_keys.union(keys)
if not duplicates.empty:
result.add_error(
f"{field} contains {len(duplicates[field].unique())} duplicated value(s)",
errors=errors,
)
result.err_items_count = len(err_keys)
return result
def compare_boolean_fields(source_df: pd.DataFrame,
target_df: pd.DataFrame) -> Result:
"""Compare booleans distribution between two dataframes
Returns:
A result containing dataframe with distributions and messages if differences
are in thresholds
"""
warn_thr = 0.05
err_thr = 0.10
source_bool = source_df.select_dtypes(include="bool")
target_bool = target_df.select_dtypes(include="bool")
result = Result("Boolean Fields")
if not fields_to_compare(source_bool, target_bool):
result.add_info("SKIPPED")
return result
dummy = pd.DataFrame(columns=[True, False])
source_counts = pd.concat(
[dummy, source_bool.apply(pd.value_counts, normalize=True).T],
sort=False).fillna(0.0)
target_counts = pd.concat(
[dummy, target_bool.apply(pd.value_counts, normalize=True).T],
sort=False).fillna(0.0)
difs = (source_counts - target_counts).abs()[True]
bool_covs = pd.concat([
source_counts.rename("{}_source".format),
target_counts.rename("{}_target".format),
]).sort_index()
bool_covs.name = "Coverage for boolean fields"
result.stats.append(bool_covs)
err_diffs = difs[difs > err_thr]
if not err_diffs.empty:
result.add_error(
f"{', '.join(err_diffs.index)} relative frequencies differ "
f"by more than {err_thr:.0%}")
warn_diffs = difs[(difs > warn_thr) & (difs <= err_thr)]
if not warn_diffs.empty:
result.add_warning(
f"{', '.join(warn_diffs.index)} relative frequencies differ by "
f"{warn_thr:.0%}-{err_thr:.0%}")
if err_diffs.empty and warn_diffs.empty:
result.add_info("PASSED")
return result
def compare_errors(source_job: Job, target_job: Job) -> Result:
errors_count1 = api.get_errors_count(source_job)
errors_count2 = api.get_errors_count(target_job)
result = Result("Compare Job Errors")
if errors_count1:
errors_url = "{}/{}/log?filterType=error&filterAndHigher"
detailed_msg = (
f"{errors_count1} error(s) for {source_job.key} - "
f"{errors_url.format(SH_URL, source_job.key)}\n"
f"{errors_count2} error(s) for {target_job.key} - "
f"{errors_url.format(SH_URL, target_job.key)}"
)
result.add_error(f"{errors_count1} and {errors_count2} errors", detailed_msg)
return result
def find_by_name_url(df: pd.DataFrame, tagged_fields: TaggedFields) -> Result:
"""Check for items with the same name and url"""
name_fields = tagged_fields.get("name_field")
url_fields = tagged_fields.get("product_url_field")
name = "Duplicates By **name_field, product_url_field** Tags"
result = Result(name)
if not name_fields or not url_fields:
result.add_info(Outcome.SKIPPED)
return result
name_field = name_fields[0]
url_field = url_fields[0]
result = find_by(df, [name_field, url_field])
result.name = name
return result
def find_by_tags(df: pd.DataFrame, tagged_fields: TaggedFields) -> Result:
"""Check for duplicates based on schema tags. In particular, look for items with
the same `name_field` and `product_url_field`, and for uniqueness among `unique` field"""
name_fields = tagged_fields.get("name_field")
url_fields = tagged_fields.get("product_url_field")
columns_to_check: List = tagged_fields.get("unique", [])
if (not name_fields or not url_fields) and not columns_to_check:
result = Result("Duplicates")
result.add_info(Outcome.SKIPPED)
return result
if name_fields and url_fields:
columns_to_check.extend([[name_fields[0], url_fields[0]]])
return find_by(df, columns_to_check)
def get_coverage_per_category(df: pd.DataFrame,
category_names: List) -> Result:
"""Get value counts per column, excluding nan.
Args:
df: a source data to assess
category_names: list of columns which values counts to see
Returns:
Number of categories per field, value counts series for each field.
"""
result = Result("Coverage For Scraped Categories")
for c in category_names:
value_counts = df[c].value_counts(ascending=True)
result.add_info(f"{len(value_counts)} categories in '{c}'")
result.stats.append(value_counts)
return result
def compare_prices_for_same_names(source_df: pd.DataFrame,
target_df: pd.DataFrame,
tagged_fields: TaggedFields):
result = Result("Compare Prices For Same Names")
name_field_tag = tagged_fields.get("name_field")
if not name_field_tag:
result.outcome = Outcome.SKIPPED
return result
name_field = name_field_tag[0]
source_df = source_df[source_df[name_field].notnull()]
target_df = target_df[target_df[name_field].notnull()]
same_names = source_df[(source_df[name_field].isin(
target_df[name_field].values))][name_field]
price_fields = tagged_fields.get("product_price_field")
if not price_fields:
result.add_info("product_price_field tag is not set")
return result
price_field = price_fields[0]
detailed_messages = []
for name in same_names:
if name.strip() != "nan":
source_price = source_df[source_df[name_field] ==
name][price_field].iloc[0]
target_price = target_df[target_df[name_field] ==
name][price_field].iloc[0]
if is_number(source_price) and is_number(target_price):
if ratio_diff(source_price, target_price) > 0.1:
source_key = source_df[source_df[name_field] ==
name].index[0]
target_key = target_df[target_df[name_field] ==
name].index[0]
msg = (
f"different price for {name}\nsource price is {source_price} "
f"for {source_key}\ntarget price is {target_price} for {target_key}"
)
detailed_messages.append(msg)
result_msg = f"{len(same_names)} checked, {len(detailed_messages)} errors"
if detailed_messages:
result.add_error(result_msg, detailed="\n".join(detailed_messages))
else:
result.add_info(result_msg)
return result
def garbage_symbols(df: pd.DataFrame) -> Result:
"""Find unwanted symbols in `np.object` columns.
Returns:
A result containing item keys per field which contained any trash symbol
"""
garbage = (
r"(?P<spaces>^\s|\s$)"
r"|(?P<html_entities>&[a-zA-Z]{2,}?;|&#\d*?;)"
r"|(?P<css>[.#@][^\d{}#.\s][^{}#.]+?{(?:[^:;{}]+?:[^:;{}]+?;)+?\s*?})"
r"|(?P<html_tags></??(?:h\d|b|u|i|div|ul|ol|li|table|tbody|th|tr|td|p|a|br|img|sup|SUP|"
r"blockquote)\s*?/??>|<!--|-->)")
errors = {}
row_keys = set()
rule_result = Result("Garbage Symbols", items_count=len(df))
for column in tqdm_notebook(df.select_dtypes([np.object]).columns,
desc="Garbage Symbols"):
matches = df[column].apply(str).str.extractall(garbage,
flags=re.IGNORECASE)
if not matches.empty:
error_keys = df.loc[matches.unstack().index.values].index
bad_texts = matches.stack().value_counts().index.sort_values(
).tolist()
# escape backslashes for markdown repr, `\n > \\n`
bad_texts = [
f"'{codecs.encode(bx, 'unicode_escape').decode()[:20]}'"
for bx in bad_texts
]
error = (f"{len(error_keys)/len(df)*100:.1f}% of '{column}' "
f"values contain `{', '.join(bad_texts)}`")
errors[error] = list(error_keys)
row_keys = row_keys.union(error_keys)
if errors:
rule_result.add_error(
f"{len(row_keys)/len(df) * 100:.1f}% ({len(row_keys)}) items affected",
errors=errors,
)
rule_result.err_items_count = len(row_keys)
return rule_result
def compare_finish_time(source_job: Job, target_job: Job) -> Result:
diff_in_days = api.get_finish_time_difference_in_days(source_job, target_job)
result = Result("Finish Time")
if diff_in_days == 0:
result.add_info("Less than 1 day difference")
else:
if diff_in_days is None:
result.add_warning("Jobs are not finished")
else:
result.add_warning(f"{diff_in_days} day(s) difference between 2 jobs")
return result
def garbage_symbols(items: Items) -> Result:
"""Find unwanted symbols in `np.object` columns.
Returns:
A result containing item keys per field which contained any trash symbol
"""
garbage = (
r"(?P<spaces>^\s|\s$)"
r"|(?P<html_entities>&|®)"
r"|(?P<css>(?:(?:\.|#)[^#. ]+\s*){.+})"
r"|(?P<html_tags></?(h\d|b|u|i|div|ul|ol|li|table|tbody|th|tr|td|p|a|br|img|sup|SUP|"
r"blockquote)\s*/?>|<!--|-->)")
errors = {}
row_keys = set()
rule_result = Result("Garbage Symbols", items_count=items.size)
for column in items.flat_df.select_dtypes([np.object]):
matches = items.flat_df[column].str.extractall(garbage,
flags=re.IGNORECASE)
matches = matches[["spaces", "html_entities", "css", "html_tags"]]
if not matches.empty:
error_keys = items.flat_df.iloc[
matches.unstack().index.values]["_key"]
original_column = items.get_origin_column_name(column)
bad_texts = matches.stack().value_counts().index.sort_values(
).tolist()
error = (
f"{len(error_keys)/items.size*100:.1f}% of '{original_column}' "
f"values contain {[t[:20] for t in bad_texts]}")
errors[error] = list(error_keys)
row_keys = row_keys.union(error_keys)
if errors:
rule_result.add_error(
f"{len(row_keys)/items.size * 100:.1f}% ({len(row_keys)}) items affected",
errors=errors,
)
rule_result.err_items_count = len(row_keys)
return rule_result
def check_fields_coverage(df: pd.DataFrame) -> Result:
"""Get fields coverage from df. Coverage reflects the percentage of real values
(exluding `nan`) per column.
Args:
df: a data to count the coverage
Returns:
A result with coverage for all columns in provided df. If column contains only `nan`,
treat it as an error.
"""
fields_coverage = df.count().sort_values(ascending=False)
fields_coverage.name = f"Fields coverage for {len(df):_} items"
empty_fields = fields_coverage[fields_coverage == 0]
result = Result("Fields Coverage")
result.stats = [fields_coverage]
if not empty_fields.empty:
result.add_error(f"{len(empty_fields)} empty field(s)")
return result
def validate(
schema: RawSchema, raw_items: RawItems, keys: pd.Index, fast: bool = False
) -> Result:
"""Run JSON schema validation against data.
Args:
fast: defines if we use fastjsonschema or jsonschema validation
Returns:
Schema errors if any
"""
validate_func = fast_validate if fast else full_validate
errors = validate_func(schema, raw_items, keys)
result = Result("JSON Schema Validation")
err_items = len(set(itertools.chain.from_iterable(errors.values())))
if errors:
result.add_error(
f"{err_items} ({err_items/len(raw_items):.0%}) items have {len(errors)} errors",
errors=errors,
)
return result
def compare_scraped_fields(source_df, target_df):
source_field_coverage = dict(
source_df.count().sort_values(ascending=False))
target_field_coverage = dict(
target_df.count().sort_values(ascending=False))
result = Result("Scraped Fields")
missing_fields = set(target_df.columns.values) - set(
source_df.columns.values)
if missing_fields:
detailed_messages = ["Missing Fields"]
for field in missing_fields:
target_coverage = target_field_coverage[field] / len(
target_df) * 100
detailed_messages.append(
f"{field} - coverage - {int(target_coverage)}% - "
f"{target_field_coverage[field]} items")
result.add_error(f"{len(missing_fields)} field(s) are missing",
"\n".join(detailed_messages))
new_fields = set(source_df.columns.values) - set(target_df.columns.values)
if new_fields:
detailed_messages = ["New Fields"]
for field in new_fields:
source_coverage = source_field_coverage[field] / len(
source_df) * 100
detailed_messages.append(
f"{field} - coverage - {int(source_coverage)}% - "
f"{source_field_coverage[field]} items")
result.add_info(f"{len(new_fields)} field(s) are new",
"\n".join(detailed_messages))
return result
