I would like to "dump" the tensorboard histograms and plot them via matplotlib. I would have more scientific paper appealing plots.
I managed to hack the way through the Summary file using the tf.train.summary_iterator and dump the histogram that I wanted to dump( tensorflow.core.framework.summary_pb2.HistogramProto object).
By doing that and implementing what the java-script code does with the data (https://github.com/tensorflow/tensorboard/blob/c2fe054231fe77f3a5b05dbc519f713d2e738d1c/tensorboard/plugins/histogram/tf_histogram_dashboard/histogramCore.ts#L104), I managed to get something similar (same trends) with the tensorboard plots, but not the exact same plot.
Can I have some light on this?
Thanks
In order to plot a tensorboard histogram with matplotlib I am doing the following:
event_acc = EventAccumulator(path, size_guidance={
'histograms': STEP_COUNT,
})
event_acc.Reload()
tags = event_acc.Tags()
result = {}
for hist in tags['histograms']:
histograms = event_acc.Histograms(hist)
result[hist] = np.array([np.repeat(np.array(h.histogram_value.bucket_limit), np.array(h.histogram_value.bucket).astype(np.int)) for h in histograms])
return result
h.histogram_value.bucket_limit gives me the value and h.histogram_value.bucket the count of this value. So when i repeat the values accordingly (np.repeat(...)), I get a huge array of expected size. This array can now be plotted with the default matplotlib logic.
The best solution is loading all events and reconstructing all the histogram (as the answer of #khuesmann) but not using EventAccumulator but EventFileLoader. This will give you a histogram per wall time and step as the ones Tensorboard plots. It can be extended to return a list of actions by timestep and wall time.
Don't forget to check which tag will you use.
from tensorboard.backend.event_processing.event_file_loader import EventFileLoader
# Just in case, PATH_OF_FILE is the path of the file, not the folder
loader = EventFileLoader(PATH_Of_FILE)
# Where to store values
wtimes,steps,actions = [],[],[]
for event in loader.Load():
wtime = event.wall_time
step = event.step
if len(event.summary.value) > 0:
summary = event.summary.value[0]
if summary.tag == HISTOGRAM_TAG:
wtimes += [wtime]*int(summary.histo.num)
steps += [step] *int(summary.histo.num)
for num,val in zip(summary.histo.bucket,summary.histo.bucket_limit):
actions += [val] *int(num)
bear in mind that tensorflow approximates the actions and treats the actions as continuous variables, so even if you have discrete actions (e.g. 0,1,3) you will end up actions as 0.2,0.4,0.9,1.4 ... in that case round the values will do it.
A good solution is the one from #khuesmann, but this only allows you to retrieve the accumulated histogram, not the histogram per step -- which is the one actually being showed in tensorboard.
If you want the distribution and so far, what I have understood is that Tensorboard usually compresses the histogram to decrease the memory used to store the data -- imagine storing a 2D histogram over 4 million steps, the memory can increase fast quickly. These compress histograms are accessible by doing this:
from tensorboard.backend.event_processing.event_accumulator import EventAccumulator
n2n = EventAccumulator(PATH)
n2n.Reload()
# Check the tags under histograms and choose the one you want
n2n.Tags()
# This will give you the list used by tensorboard
# of the compress histograms by timestep and wall time
n2n.CompressedHistograms(HISTOGRAM_TAG)
The only problem is that it compresses the histogram to five percentiles (in Basic points they are 0, 668, 1587, 3085, 5000, 6915, 8413, 9332, 10000) which corresponds to (-Inf, -1.5, -1, -0.5, 0, 0.5, 1, 1.5, Inf) in standard deviations. Check the code here.
I haven't read much, but it wouldn't be hard to reconstruct the temporal histograms that tensorboard shows. If I find a way to do it, I will post it here.
The simplest way is to parse the events with tbparse and plot the histograms with seaborn kde_ridgeplot.
This tutorial generates the stacked distribution plot with around 30 lines of Python code:
Tensorboard preview:
Parse by tbparse & plotted by seaborn:
You can open an issue if you encountered any question during parsing. (I'm the author of tbparse)
Related
The Problem
Good evening.
I am learning about the Central Limit Theorem. As practice, I ran simulations in an attempt to find the mean of a fair die (I know, a toy problem).
I took 4000 samples, and in each sample I rolled a die 50 times (screenshot of the code at the bottom). For each of these 4000 samples I computed the mean. Then, I plotted these 4000 sample means in a histogram (with bin size 0.03) using matplotlib.
Here is the result:
Question
Why aren't the sample means normally distributed given that the conditions for CLT (sample size >= 30) were respected?
Specifically, why does the histogram look like two normal distributions superimposed on top of each other? More intriguingly, why does the "outer" distribution look "discrete" with empty spaces occurring at regular intervals?
It almost seems like the result is off in a systematic way.
All help is greatly appreciated. I am very lost.
Supplementary Code
The code I used to generate the 4000 sample means.
"""
Take multiple samples of dice rolls. For
each sample, compute the sample mean.
With the sample means, plot a histogram.
By the Central Limit Theorem, the sample
means should be normally distributed.
"""
sample_means = []
num_samples = 4000
for i in range(num_samples):
# Large enough for CLT to hold
num_rolls = 50
sample = []
for j in range(num_rolls):
observation = random.randint(1, 6)
sample.append(observation)
sample_mean = sum(sample) / len(sample)
sample_means.append(sample_mean)
When num_rolls equals 50, each possible mean will be a fraction with denominator 50. So, in reality, you are looking at a discrete distribution.
To create a histogram of a discrete distribution, the bin boundaries are best placed nicely in-between the values. Using a step size of 0.03, some bin boundaries will coincide with the values, putting the double of values into one bin compared to its neighbor. Moreover, due to subtle floating point rounding problems, the result can become unpredictable when values and boundaries coincide.
Here is some code to illustrate what is going on:
from matplotlib import pyplot as plt
import numpy as np
import random
sample_means = []
num_samples = 4000
for i in range(num_samples):
num_rolls = 50
sample = []
for j in range(num_rolls):
observation = random.randint(1, 6)
sample.append(observation)
sample_mean = sum(sample) / len(sample)
sample_means.append(sample_mean)
fig, axs = plt.subplots(2, 2, figsize=(14, 8))
random_y = np.random.rand(len(sample_means))
for (ax0, ax1), step in zip(axs, [0.03, 0.02]):
bins = np.arange(3.01, 4, step)
ax0.hist(sample_means, bins=bins)
ax0.set_title(f'step={step}')
ax0.vlines(bins, 0, ax0.get_ylim()[1], ls=':', color='r') # show the bin boundaries in red
ax1.scatter(sample_means, random_y, s=1) # show the sample means with a random y
ax1.vlines(bins, 0, 1, ls=':', color='r') # show the bin boundaries in red
ax1.set_xticks(np.arange(3, 4, 0.02))
ax1.set_xlim(3.0, 3.3) # zoom in to region to better see the ins
ax1.set_title('bin boundaries between values' if step == 0.02 else 'chaotic bin boundaries')
plt.show()
PS: Note that the code would run much, much faster if instead of Python lists, the code would work completely with numpy.
I want to do a scatterplot according x and y variables, and the points size depend of a numeric variable and the color of every point depend of a categorical variable.
First, I was trying this with plt.scatter:
Graph 1
After, I tried this using lmplot but the point size is different in relation to the first graph.
I think the two graphs should be equals. Why not?
The point size is different in every graph.
Graph 2
Your question is no so much descriptive but i guess you want to control the size of the marker. Here is more documentation
Here is the start point for you.
A numeric variable can also be assigned to size to apply a semantic mapping to the areas of the points:
import seaborn as sns
tips = sns.load_dataset("tips")
sns.scatterplot(data=tips, x="total_bill", y="tip", hue="size", size="size")
For seaborn scatterplot:
df = sns.load_dataset("anscombe")
sp = sns.scatterplot(x="x", y="y", hue="dataset", data=df)
And to change the size of the points you use the s parameter.
sp = sns.scatterplot(x="x", y="y", hue="dataset", data=df, s=100)
I am a recent migrant from Matlab to Python and have recently worked with Numpy and Matplotlib. I recoded one of my scripts from Matlab, which employs Matlab's contourf-function, into Python using matplotlib's corresponding contourf-function. I managed to replicate the output in Python, apart that the contourf-plots are not exacly the same, for a reason that is unknown to me. As I run the contourf-function in matplotlib, I get this otherwise nice figure but it has these sharp edges on the contour-levels on top and bottom, which should not be there (see Figure 1 below, matplotlib-output). Now, when I export the arrays I used in Python to Matlab (i.e. the exactly same data set that was used to generate the matplotlib-contourf-plot) and use Matlab's contourf-function, I get a slightly different output, without those sharp contour-level edges (see Figure 2 below, Matlab-output). I used the same number of levels in both figures. In figure 3 I have made a scatterplot of the same data, which shows that there are no such sharp edges in the data as shown in the contourf-plot (I added contour-lines just for reference). Example dataset can be downloaded through Dropbox-link given below. The data set contains three txt-files: X, Y, Z. Each of them are an 500x500 arrays, which can be directly used with contourf(), i.e. plt.contourf(X,Y,Z,...). The code that used was
plt.contourf(X,Y,Z,10, cmap=plt.cm.jet)
plt.contour(X,Y,Z,10,colors='black', linewidths=0.5)
plt.axis('equal')
plt.axis('off')
Does anyone have an idea why this happens? I would appreciate any insight on this!
Cheers,
Jussi
Below are the details of my setup:
Python 3.7.0
IPython 6.5.0
matplotlib 2.2.3
Matplotlib output
Matlab output
Matplotlib-scatter
Link to data set
The confusing thing about the matlab plot is that its colorbar shows much more levels than there are actually in the plot. Hence you don't see the actual intervals that are contoured.
You would achieve the same result in matplotlib by choosing 12 instead of 11 levels.
import numpy as np
import matplotlib.pyplot as plt
X, Y, Z = [np.loadtxt("data/roundcontourdata/{}.txt".format(i)) for i in list("XYZ")]
levels = np.linspace(Z.min(), Z.max(), 12)
cntr = plt.contourf(X,Y,Z,levels, cmap=plt.cm.jet)
plt.contour(X,Y,Z,levels,colors='black', linewidths=0.5)
plt.colorbar(cntr)
plt.axis('equal')
plt.axis('off')
plt.show()
So in conclusion, both plots are correct and show the same data. Just the levels being automatically chosen are different. This can be circumvented by choosing custom levels depending on the desired visual appearance.
BLUF: I want to track a specific Std Dev, e.g. 1.0 to 1.25, by color coding it and making a separate KDF or other probability density graph.
What I want to do with this is be able to pick out other Std Dev ranges and get back new graphs that I can turn around and use to predict outcomes in that specific Std Dev.
Data: https://www.dropbox.com/s/y78pynq9onyw9iu/Data.csv?dl=0
What I have so far is normalized data that looks like a shotgun blast:
Code used to produce it:
data = pd.read_csv("Data.csv")
sns.jointplot(data.x,data.y, space=0.2, size=10, ratio=2, kind="reg");
What I want to achieve here looks like what I have marked up below:
I kind of know how to do this in RStudio using RidgePlot-type functions, but I'm at a loss here in Python, even while using Seaborn. Any/All help appreciated!
The following code might point you in the right directly, you can tweak the appearance of the plot as you please from there.
tips = sns.load_dataset("tips")
g = sns.jointplot(x="total_bill", y="tip", data=tips)
top_lim = 4
bottom_lim = 2
temp = tips.loc[(tips.tip>=bottom_lim)&(tips.tip<top_lim)]
g.ax_joint.axhline(top_lim, c='k', lw=2)
g.ax_joint.axhline(bottom_lim, c='k', lw=2)
# we have to create a secondary y-axis to the joint-plot, otherwise the
# kde might be very small compared to the scale of the original y-axis
ax_joint_2 = g.ax_joint.twinx()
sns.kdeplot(temp.total_bill, shade=True, color='red', ax=ax_joint_2, legend=False)
ax_joint_2.spines['right'].set_visible(False)
ax_joint_2.spines['top'].set_visible(False)
ax_joint_2.yaxis.set_visible(False)
I'm new to Python and MatPlotlib.
This is my first posting to Stackoverflow - I've been unable to find the answer elsewhere and would be grateful for your help.
I'm using Windows XP, with Enthought Canopy v1.1.1 (32 bit).
I want to plot a dotted-style linear regression line through a scatter plot of data, where both x and y arrays contain random floating point data.
The dots in the resulting dotted line are not distributed evenly along the regression line, and are "smeared together" in the middle of the red line, making it look messy (see upper plot resulting from attached minimal example code).
This does not seem to occur if the items in the array of x values are evenly distributed (lower plot).
I'm therefore guessing that this is an issue with how MatplotLib renders dotted lines, or with how Canopy interfaces Python with Matplotlib.
Please could you tell me a workaround which will make the dots on the dotted line type appear evenly distributed; even if both x and y data are non-evenly distributed; whilst still using Canopy and Matplotlib?
(As a general point, I'm always keen to improve my coding skills - if any code in my example can be written more neatly or concisely, I'd be grateful for your expertise).
Many thanks in anticipation
Dave
(UK)
import matplotlib.pyplot as plt
import numpy as np
from scipy import stats
#generate data
x1=10 * np.random.random_sample((40))
x2=np.linspace(0,10,40)
y=5 * np.random.random_sample((40))
slope, intercept, r_value, p_value, std_err = stats.linregress(x1,y)
line = (slope*x1)+intercept
plt.figure(1)
plt.subplot(211)
plt.scatter(x1,y,color='blue', marker='o')
plt.plot(x1,line,'r:',label="Regression Line")
plt.legend(loc='upper right')
slope, intercept, r_value, p_value, std_err = stats.linregress(x2,y)
line = (slope*x2)+intercept
plt.subplot(212)
plt.scatter(x2,y,color='blue', marker='o')
plt.plot(x2,line,'r:',label="Regression Line")
plt.legend(loc='upper right')
plt.show()
Welcome to SO.
You have already identified the problem yourself, but seem a bit surprised that a random x-array results in the line be 'cluttered'. But you draw a dotted line repeatedly over the same location, so it seems like the normal behavior to me that it gets smeared at places where there are multiple dotted lines on top of each other.
If you don't want that, you can sort your array and use that to calculate the regression line and plot it. Since its a linear regression, just using the min and max values would also work.
x1_sorted = np.sort(x1)
line = (slope * x1_sorted) + intercept
or
x1_extremes = np.array([x1.min(),x1.max()])
line = (slope * x1_extremes) + intercept
The last should be faster if x1 becomes very large.
With regard to your last comment. In your example you use whats called the 'state-machine' environment for plotting. It means that specified commands are applied to the active figure and the active axes (subplots).
You can also consider the OO approach where you get figure and axes objects. This means you can access any figure or axes at any time, not just the active one. Its useful when passing an axes to a function for example.
In your example both would work equally well and it would be more a matter of taste.
A small example:
# create a figure with 2 subplots (2 rows, 1 column)
fig, axs = plt.subplots(2,1)
# plot in the first subplots
axs[0].scatter(x1,y,color='blue', marker='o')
axs[0].plot(x1,line,'r:',label="Regression Line")
# plot in the second
axs[1].plot()
etc...