def describe_float_1d(df, column, current_result, nrows):
stats_df = df.select(column).na.drop().agg(
mean(col(column)).alias("mean"),
df_min(col(column)).alias("min"),
df_max(col(column)).alias("max"),
variance(col(column)).alias("variance"),
kurtosis(col(column)).alias("kurtosis"),
stddev(col(column)).alias("std"),
skewness(col(column)).alias("skewness"),
df_sum(col(column)).alias("sum")).toPandas()
for x in np.array([0.05, 0.25, 0.5, 0.75, 0.95]):
stats_df[pretty_name(x)] = (df.select(column).na.drop().selectExpr(
"percentile_approx(`{col}`,CAST({n} AS DOUBLE))".format(
col=column, n=x)).toPandas().iloc[:, 0])
stats = stats_df.iloc[0].copy()
stats.name = column
stats["range"] = stats["max"] - stats["min"]
stats["iqr"] = stats[pretty_name(0.75)] - stats[pretty_name(0.25)]
stats["cv"] = stats["std"] / float(stats["mean"])
stats["mad"] = (df.select(column).na.drop().select(
df_abs(col(column) - stats["mean"]).alias("delta")).agg(
df_sum(col("delta"))).toPandas().iloc[0, 0] /
float(current_result["count"]))
stats["type"] = "NUM"
stats['n_zeros'] = df.select(column).where(col(column) == 0.0).count()
stats['p_zeros'] = stats['n_zeros'] / float(nrows)
hist_data = create_hist_data(df, column, stats["min"], stats["max"],
bins)
return stats
def describe_date_1d(df, column):
stats_df = df.select(column).na.drop().agg(
df_min(col(column)).alias("min"),
df_max(col(column)).alias("max")).toPandas()
stats = stats_df.iloc[0].copy()
stats.name = column
if isinstance(stats["max"], pd.Timestamp):
stats = stats.astype(object)
stats["max"] = str(stats["max"].to_pydatetime())
stats["min"] = str(stats["min"].to_pydatetime())
else:
stats["range"] = stats["max"] - stats["min"]
stats["type"] = "DATE"
return stats
def describe_date_1d(df, column):
stats_df = df.select(column).na.drop().agg(df_min(col(column)).alias("min"),
df_max(col(column)).alias("max")
).toPandas()
stats = stats_df.ix[0].copy()
stats.name = column
# Convert Pandas timestamp object to regular datetime:
if isinstance(stats["max"], pd.tslib.Timestamp):
stats = stats.astype(object)
stats["max"] = str(stats["max"].to_pydatetime())
stats["min"] = str(stats["min"].to_pydatetime())
# Range only got when type is date
else:
stats["range"] = stats["max"] - stats["min"]
stats["type"] = "DATE"
return stats
def describe_date_1d(df, column):
stats_df = df.select(column).na.drop().agg(
df_min(col(column)).alias("min"),
df_max(col(column)).alias("max")).toPandas()
stats = stats_df.ix[0].copy()
stats.name = column
# Convert Pandas timestamp object to regular datetime:
if isinstance(stats["max"], pd.tslib.Timestamp):
stats = stats.astype(object)
stats["max"] = str(stats["max"].to_pydatetime())
stats["min"] = str(stats["min"].to_pydatetime())
# Range only got when type is date
else:
stats["range"] = stats["max"] - stats["min"]
stats["type"] = "DATE"
return stats
def describe_float_1d(df, column, current_result, nrows):
if spark_version == "1.6+":
stats_df = df.select(column).na.drop().agg(mean(col(column)).alias("mean"),
df_min(col(column)).alias("min"),
df_max(col(column)).alias("max"),
variance(col(column)).alias("variance"),
kurtosis(col(column)).alias("kurtosis"),
stddev(col(column)).alias("std"),
skewness(col(column)).alias("skewness"),
df_sum(col(column)).alias("sum"),
count(col(column) == 0.0).alias('n_zeros')
).toPandas()
else:
stats_df = df.select(column).na.drop().agg(mean(col(column)).alias("mean"),
df_min(col(column)).alias("min"),
df_max(col(column)).alias("max"),
df_sum(col(column)).alias("sum"),
count(col(column) == 0.0).alias('n_zeros')
).toPandas()
stats_df["variance"] = df.select(column).na.drop().agg(variance_custom(col(column),
stats_df["mean"].iloc[0],
current_result["count"])).toPandas().iloc[0][0]
stats_df["std"] = np.sqrt(stats_df["variance"])
stats_df["skewness"] = df.select(column).na.drop().agg(skewness_custom(col(column),
stats_df["mean"].iloc[0],
current_result["count"])).toPandas().iloc[0][0]
stats_df["kurtosis"] = df.select(column).na.drop().agg(kurtosis_custom(col(column),
stats_df["mean"].iloc[0],
current_result["count"])).toPandas().iloc[0][0]
for x in [0.05, 0.25, 0.5, 0.75, 0.95]:
stats_df[pretty_name(x)] = (df.select(column)
.na.drop()
.selectExpr("percentile_approx(`{col}`,CAST({n} AS DOUBLE))"
.format(col=column, n=x)).toPandas().iloc[:,0]
)
stats = stats_df.iloc[0].copy()
stats.name = column
stats["range"] = stats["max"] - stats["min"]
stats["iqr"] = stats[pretty_name(0.75)] - stats[pretty_name(0.25)]
stats["cv"] = stats["std"] / float(stats["mean"])
stats["mad"] = (df.select(column)
.na.drop()
.select(df_abs(col(column)-stats["mean"]).alias("delta"))
.agg(df_sum(col("delta"))).toPandas().iloc[0,0] / float(current_result["count"]))
stats["type"] = "NUM"
stats['p_zeros'] = stats['n_zeros'] / float(nrows)
# Large histogram
imgdata = BytesIO()
hist_data = create_hist_data(df, column, stats["min"], stats["max"], bins)
figure = plt.figure(figsize=(6, 4))
plot = plt.subplot()
plt.bar(hist_data["left_edge"],
hist_data["count"],
width=hist_data["width"],
facecolor='#337ab7')
plot.set_ylabel("Frequency")
plot.figure.subplots_adjust(left=0.15, right=0.95, top=0.9, bottom=0.1, wspace=0, hspace=0)
plot.figure.savefig(imgdata)
imgdata.seek(0)
stats['histogram'] = 'data:image/png;base64,' + quote(base64.b64encode(imgdata.getvalue()))
#TODO Think about writing this to disk instead of caching them in strings
plt.close(plot.figure)
stats['mini_histogram'] = mini_histogram(hist_data)
return stats
def describe_float_1d(df, column, current_result, nrows):
if spark_version == "1.6+":
stats_df = df.select(column).na.drop().agg(mean(col(column)).alias("mean"),
df_min(col(column)).alias("min"),
df_max(col(column)).alias("max"),
variance(col(column)).alias("variance"),
kurtosis(col(column)).alias("kurtosis"),
stddev(col(column)).alias("std"),
skewness(col(column)).alias("skewness"),
df_sum(col(column)).alias("sum")
).toPandas()
else:
stats_df = df.select(column).na.drop().agg(mean(col(column)).alias("mean"),
df_min(col(column)).alias("min"),
df_max(col(column)).alias("max"),
df_sum(col(column)).alias("sum")
).toPandas()
stats_df["variance"] = df.select(column).na.drop().agg(variance_custom(col(column),
stats_df["mean"].ix[0],
current_result["count"])).toPandas().ix[0][0]
stats_df["std"] = np.sqrt(stats_df["variance"])
stats_df["skewness"] = df.select(column).na.drop().agg(skewness_custom(col(column),
stats_df["mean"].ix[0],
current_result["count"])).toPandas().ix[0][0]
stats_df["kurtosis"] = df.select(column).na.drop().agg(kurtosis_custom(col(column),
stats_df["mean"].ix[0],
current_result["count"])).toPandas().ix[0][0]
for x in np.array([0.05, 0.25, 0.5, 0.75, 0.95]):
stats_df[pretty_name(x)] = (df.select(column)
.na.drop()
.selectExpr("percentile_approx(`{col}`,CAST({n} AS DOUBLE))"
.format(col=column, n=x)).toPandas().ix[:,0]
)
stats = stats_df.ix[0].copy()
stats.name = column
stats["range"] = stats["max"] - stats["min"]
stats["iqr"] = stats[pretty_name(0.75)] - stats[pretty_name(0.25)]
stats["cv"] = stats["std"] / float(stats["mean"])
stats["mad"] = (df.select(column)
.na.drop()
.select(df_abs(col(column)-stats["mean"]).alias("delta"))
.agg(df_sum(col("delta"))).toPandas().ix[0,0] / float(current_result["count"]))
stats["type"] = "NUM"
stats['n_zeros'] = df.select(column).where(col(column)==0.0).count()
stats['p_zeros'] = stats['n_zeros'] / float(nrows)
# Large histogram
imgdata = BytesIO()
hist_data = create_hist_data(df, column, stats["min"], stats["max"], bins)
figure = plt.figure(figsize=(6, 4))
plot = plt.subplot()
plt.bar(hist_data["left_edge"],
hist_data["count"],
width=hist_data["width"],
facecolor='#337ab7')
plot.set_ylabel("Frequency")
plot.figure.subplots_adjust(left=0.15, right=0.95, top=0.9, bottom=0.1, wspace=0, hspace=0)
plot.figure.savefig(imgdata)
imgdata.seek(0)
stats['histogram'] = 'data:image/png;base64,' + quote(base64.b64encode(imgdata.getvalue()))
#TODO Think about writing this to disk instead of caching them in strings
plt.close(plot.figure)
stats['mini_histogram'] = mini_histogram(hist_data)
return stats
