I'm trying to train a RandomForestRegressor using DecisionTree.jl
and RandomizedSearchCV (contained in ScikitLearn.jl) in Julia. Primary datasets like x_train and y_train etc. are provided in my google drive as well, So you can test it on your machine. The code is as follows:
using CSV
using DataFrames
using ScikitLearn: fit!, predict
using ScikitLearn.GridSearch: RandomizedSearchCV
using DecisionTree
x = CSV.read("x.csv", DataFrames.DataFrame)
x_test = CSV.read("x_test.csv", DataFrames.DataFrame)
y_train = CSV.read("y_train.csv", DataFrames.DataFrame)
mod = RandomForestRegressor()
param_dist = Dict("n_trees"=>[50 , 100, 200, 300],
"max_depth"=> [3, 5, 6 ,8 , 9 ,10])
model = RandomizedSearchCV(mod, param_dist, n_iter=10, cv=5)
fit!(model, Matrix(x), Matrix(DataFrames.dropmissing(y_train)))
predict(x_test)
This throws a MethodError like this:
ERROR: MethodError: no method matching fit!(::RandomForestRegressor, ::Matrix{Float64}, ::Matrix{Float64})
Closest candidates are:
fit!(::ScikitLearn.Models.FixedConstant, ::Any, ::Any) at C:\Users\Shayan\.julia\packages\ScikitLearn\ssekP\src\models\constant_model.jl:26
fit!(::ScikitLearn.Models.ConstantRegressor, ::Any, ::Any) at C:\Users\Shayan\.julia\packages\ScikitLearn\ssekP\src\models\constant_model.jl:10
fit!(::ScikitLearn.Models.LinearRegression, ::AbstractArray{XT}, ::AbstractArray{yT}) where {XT, yT} at C:\Users\Shayan\.julia\packages\ScikitLearn\ssekP\src\models\linear_regression.jl:27
...
Stacktrace:
[1] _fit!(self::RandomizedSearchCV, X::Matrix{Float64}, y::Matrix{Float64}, parameter_iterable::Vector{Any})
# ScikitLearn.Skcore C:\Users\Shayan\.julia\packages\ScikitLearn\ssekP\src\grid_search.jl:332
[2] fit!(self::RandomizedSearchCV, X::Matrix{Float64}, y::Matrix{Float64})
# ScikitLearn.Skcore C:\Users\Shayan\.julia\packages\ScikitLearn\ssekP\src\grid_search.jl:748
[3] top-level scope
# c:\Users\Shayan\Desktop\AUT\Thesis\test.jl:17
If you're curious about the shape of the data:
julia> size(x)
(1550, 71)
julia> size(y_train)
(1550, 10)
How can I solve this problem?
PS: Also I tried:
julia> fit!(model, Matrix{Any}(x), Matrix{Any}(DataFrames.dropmissing(y_train)))
ERROR: MethodError: no method matching fit!(::RandomForestRegressor, ::Matrix{Any}, ::Matrix{Any})
Closest candidates are:
fit!(::ScikitLearn.Models.FixedConstant, ::Any, ::Any) at C:\Users\Shayan\.julia\packages\ScikitLearn\ssekP\src\models\constant_model.jl:26
fit!(::ScikitLearn.Models.ConstantRegressor, ::Any, ::Any) at C:\Users\Shayan\.julia\packages\ScikitLearn\ssekP\src\models\constant_model.jl:10
fit!(::ScikitLearn.Models.LinearRegression, ::AbstractArray{XT}, ::AbstractArray{yT}) where {XT, yT} at C:\Users\Shayan\.julia\packages\ScikitLearn\ssekP\src\models\linear_regression.jl:27
...
Stacktrace:
[1] _fit!(self::RandomizedSearchCV, X::Matrix{Any}, y::Matrix{Any}, parameter_iterable::Vector{Any})
# ScikitLearn.Skcore C:\Users\Shayan\.julia\packages\ScikitLearn\ssekP\src\grid_search.jl:332
[2] fit!(self::RandomizedSearchCV, X::Matrix{Any}, y::Matrix{Any})
# ScikitLearn.Skcore C:\Users\Shayan\.julia\packages\ScikitLearn\ssekP\src\grid_search.jl:748
[3] top-level scope
# c:\Users\Shayan\Desktop\AUT\Thesis\MyWork\Thesis.jl:327
Looking at Random Forest Regression example docs in DecisionTree.jl, the example doesn't follow the fit!() / predict() design pattern. The error confirms that fit!() doesn't support RandomForestRegression. Alternatively, you might look at RandomForest.jl package which does follow fit!() / predict() pattern.
As stated here, DecisionTree.jl doesn't support Multi-output RF yet. So I gave up on using DecisionTree.jl, And ScikitLearn.jl is adequate in my case:
using ScikitLearn: #sk_import, fit!, predict
#sk_import ensemble: RandomForestRegressor
using ScikitLearn.GridSearch: RandomizedSearchCV
using CSV
using DataFrames
x = CSV.read("x.csv", DataFrames.DataFrame)
x_test = CSV.read("x_test.csv", DataFrames.DataFrame)
y_train = CSV.read("y_train.csv", DataFrames.DataFrame)
x_test = reshape(x_test, 1,length(x_test))
mod = RandomForestRegressor()
param_dist = Dict("n_estimators"=>[50 , 100, 200, 300],
"max_depth"=> [3, 5, 6 ,8 , 9 ,10])
model = RandomizedSearchCV(mod, param_dist, n_iter=10, cv=5)
fit!(model, Matrix(x), Matrix(DataFrames.dropmissing(y_train)))
predict(model, x_test)
This works fine for me, But it's super slow! Much slower than Python. I'll add the benchmarking with the same data sets across these two languages.
Benchmarking
Here I report the result of benchmarking with the same action, the same values, and the same data. All the data and code files are available in my Google Drive. So feel free to test it by yourself. First, I start with Julia.
Julia
using CSV
using DataFrames
using ScikitLearn: #sk_import, fit!, predict
#sk_import ensemble: RandomForestRegressor
using ScikitLearn.GridSearch: RandomizedSearchCV
using BenchmarkTools
x = CSV.read("x.csv", DataFrames.DataFrame)
y_train = CSV.read("y_train.csv", DataFrames.DataFrame)
mod = RandomForestRegressor(max_leaf_nodes=2)
param_dist = Dict("n_estimators"=>[50 , 100, 200, 300],
"max_depth"=> [3, 5, 6 ,8 , 9 ,10])
model = RandomizedSearchCV(mod, param_dist, n_iter=10, cv=5, n_jobs=1)
#btime fit!(model, Matrix(x), Matrix(DataFrames.dropmissing(y_train)))
# 52.123 s (6965 allocations: 44.34 MiB)
Python
>>> import cProfile, pstats
>>> import pandas as pd
>>> from sklearn.ensemble import RandomForestRegressor
>>> from sklearn.model_selection import RandomizedSearchCV
>>> x = pd.read_csv("x.csv")
>>> y_train = pd.read_csv("y_train.csv")
>>> mod = RandomForestRegressor(max_leaf_nodes=2)
>>> parameters = {
'n_estimators': [50 , 100, 200, 300],
'max_depth': [3, 5, 6 ,8 , 9 ,10]}
>>> model = RandomizedSearchCV(mod, param_distributions=parameters, cv=5, n_iter=10, n_jobs=1)
>>> pr = cProfile.Profile()
>>> pr.enable()
>>> model.fit(x , y_train)
>>> pr.disable()
>>> stats = pstats.Stats(pr).strip_dirs().sort_stats("cumtime")
>>> stats.print_stats(5)
12097437 function calls (11936452 primitive calls) in 73.452 seconds
Ordered by: cumulative time
List reduced from 736 to 5 due to restriction <5>
ncalls tottime percall cumtime percall filename:lineno(function)
1 0.000 0.000 73.445 73.445 _search.py:738(fit)
102/2 0.027 0.000 73.370 36.685 parallel.py:960(__call__)
12252/152 0.171 0.000 73.364 0.483 parallel.py:798(dispatch_one_batch)
12150/150 0.058 0.000 73.324 0.489 parallel.py:761(_dispatch)
12150/150 0.025 0.000 73.323 0.489 _parallel_backends.py:206(apply_async)
So I conclude that Julia performs better than Python in this specific problem in case of speed.
Related
When trying to train/evaluate a support vector machine in scikit-learn, I am experiencing some unexpected behaviour and I am wondering whether I am doing something wrong or that this is a possible bug.
In a very specific subset of circumstances, nested cross-validation using GridSearchCV and SVM, provides inflated predictive results, even with randomly generated data.
For instance, see this code:
from sklearn import svm
from sklearn.linear_model import LogisticRegression
import numpy as np
from sklearn.model_selection import GridSearchCV, StratifiedKFold, LeaveOneOut
from sklearn.metrics import roc_auc_score, brier_score_loss
from tqdm import tqdm
import pandas as pd
N = 20
N_FEATURES = 50
param_grid = {'C': [1e-5, 1e-3, 1, 1e3, 1e5]}
scores = []
for z in tqdm(range(100)):
X = np.random.uniform(size=(N, N_FEATURES))
y = np.random.binomial(1, 0.5, size=N)
if z < 10:
y = np.array([0, 1] * int(N/2))
y = np.random.permutation(y)
for skf_outer in [StratifiedKFold(n_splits=5), LeaveOneOut()]:
for skf_inner in [5, LeaveOneOut()]:
for model in [svm.SVC(probability=True), LogisticRegression()]:
y_pred, y_real = [], []
for train_index, test_index in skf_outer.split(X, y):
X_train, X_test = X[train_index], X[test_index, :]
y_train, y_test = y[train_index], y[test_index]
clf = GridSearchCV(
model, param_grid, cv=skf_inner, n_jobs=-1, scoring='neg_brier_score'
)
clf.fit(X_train, y_train)
predictions = clf.predict_proba(X_test)[:, 1]
y_pred.extend(predictions)
y_real.extend(y_test)
scores.append([str(skf_outer), str(skf_inner), str(model), np.mean(y), brier_score_loss(np.array(y_real), np.array(y_pred)), roc_auc_score(np.array(y_real), np.array(y_pred))])
df_scores = pd.DataFrame(scores)
df_scores.columns = ['skf_outer', 'skf_inner', 'model', 'y_label', 'brier', 'auc']
df_scores['y_0.5'] = df_scores['y_label'] == 0.5
df_scores = df_scores.groupby(['skf_outer', 'skf_inner', 'model', 'y_0.5']).mean()
print(df_scores)
In the following circumstances:
Both in the inner- and outerloop of the CV, LeaveOneOut() is used
The SVM is used
The y labels are balanced (i.e. the mean of y is 0.5)
The predictions are much better than expected by random chance (AUC>0.9, sometimes even 1, Brier of 0.15 or lower). I can replicate this generating more samples, more features etc - the issue stays the same. Swapping the SVM for LogisticRegression (as shown in the analysis above), leads to expected results (AUC 0.5, Brier of 0.25). And for the other scenario's (no LOO-CV in either inner or outer loop, or a different distribution of y labels), the results are as expected.
Can anyone replicate this? Am I missing something obvious?
I've replicated this with an older version of sklearn (0.24.0) and the newest one (1.2.0).
I am trying to make a NLP multi-class sentiment classifier where it takes in sentences as input and classifies them into three classes (negative, neutral and positive). However, when training the model, I run into the error where my logits (None, 3) are not the same size as my labels (None, 1) and the model can't begin training.
My model is a multi-class classifier and not a multi-label classifier since it is only predicting one label per object. I made sure that my last layer had an output of 3 and had the activation = 'softmax'. This should be correct from what I have searched online so I think that the problem lies with my labels.
Currently, my labels have a dimension of (None, 1) since I mapped each class to a unique integer and passed this as my test and train y values (which are in the form of one dimensional numpy array.
Right now I am confused if I have change the dimensions of this array to match the output dimensions and how to go about doing it.
import os
import sys
import tensorflow as tf
import numpy as np
import pandas as pd
from tensorflow import keras
from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.preprocessing.sequence import pad_sequences
from keras.optimizers import SGD
device_name = tf.test.gpu_device_name()
if len(device_name) > 0:
print("Found GPU at: {}".format(device_name))
else:
device_name = "/device:CPU:0"
print("No GPU, using {}.".format(device_name))
# Load dataset into a dataframe
train_data_path = "/content/drive/MyDrive/ML Datasets/tweet_sentiment_analysis/train.csv"
test_data_path = "/content/drive/MyDrive/ML Datasets/tweet_sentiment_analysis/test.csv"
train_df = pd.read_csv(train_data_path, encoding='unicode_escape')
test_df = pd.read_csv(test_data_path, encoding='unicode_escape').dropna()
sentiment_types = ('neutral', 'negative', 'positive')
train_df['sentiment'] = train_df['sentiment'].astype('category')
test_df['sentiment'] = test_df['sentiment'].astype('category')
train_df['sentiment_cat'] = train_df['sentiment'].cat.codes
test_df['sentiment_cat'] = test_df['sentiment'].cat.codes
train_y = np.array(train_df['sentiment_cat'])
test_y = np.array(test_df['sentiment_cat'])
# Function to convert df into a list of strings
def convert_to_list(df, x):
selected_text_list = []
labels = []
for index, row in df.iterrows():
selected_text_list.append(str(row[x]))
labels.append(str(row['sentiment']))
return np.array(selected_text_list), np.array(labels)
train_sentences, train_labels = convert_to_list(train_df, 'selected_text')
test_sentences, test_labels = convert_to_list(test_df, 'text')
# Instantiate tokenizer and create word_index
tokenizer = Tokenizer(num_words=1000, oov_token='<oov>')
tokenizer.fit_on_texts(train_sentences)
word_index = tokenizer.word_index
# Convert sentences into a sequence
train_sequence = tokenizer.texts_to_sequences(train_sentences)
test_sequence = tokenizer.texts_to_sequences(test_sentences)
# Padding sequences
pad_test_seq = pad_sequences(test_sequence, padding='post')
max_len = pad_test_seq[0].size
pad_train_seq = pad_sequences(train_sequence, padding='post', maxlen=max_len)
model = tf.keras.Sequential([
tf.keras.layers.Embedding(10000, 64, input_length=max_len),
tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(64, return_sequences=True)),
tf.keras.layers.GlobalAveragePooling1D(),
tf.keras.layers.Dense(64, activation='relu'),
tf.keras.layers.Dense(3, activation='softmax')
])
with tf.device(device_name):
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
num_epochs = 10
with tf.device(device_name):
history = model.fit(pad_train_seq, train_y, epochs=num_epochs, validation_data=(pad_test_seq, test_y), verbose=2)
Here is the error:
ValueError Traceback (most recent call last)
<ipython-input-28-62f3c6445887> in <module>
2
3 with tf.device(device_name):
----> 4 history = model.fit(pad_train_seq, train_y, epochs=num_epochs, validation_data=(pad_test_seq, test_y), verbose=2)
1 frames
/usr/local/lib/python3.8/dist-packages/keras/engine/training.py in tf__train_function(iterator)
13 try:
14 do_return = True
---> 15 retval_ = ag__.converted_call(ag__.ld(step_function), (ag__.ld(self), ag__.ld(iterator)), None, fscope)
16 except:
17 do_return = False
ValueError: in user code:
File "/usr/local/lib/python3.8/dist-packages/keras/engine/training.py", line 1051, in train_function *
return step_function(self, iterator)
File "/usr/local/lib/python3.8/dist-packages/keras/engine/training.py", line 1040, in step_function **
outputs = model.distribute_strategy.run(run_step, args=(data,))
File "/usr/local/lib/python3.8/dist-packages/keras/engine/training.py", line 1030, in run_step **
outputs = model.train_step(data)
File "/usr/local/lib/python3.8/dist-packages/keras/engine/training.py", line 890, in train_step
loss = self.compute_loss(x, y, y_pred, sample_weight)
File "/usr/local/lib/python3.8/dist-packages/keras/engine/training.py", line 948, in compute_loss
return self.compiled_loss(
File "/usr/local/lib/python3.8/dist-packages/keras/engine/compile_utils.py", line 201, in __call__
loss_value = loss_obj(y_t, y_p, sample_weight=sw)
File "/usr/local/lib/python3.8/dist-packages/keras/losses.py", line 139, in __call__
losses = call_fn(y_true, y_pred)
File "/usr/local/lib/python3.8/dist-packages/keras/losses.py", line 243, in call **
return ag_fn(y_true, y_pred, **self._fn_kwargs)
File "/usr/local/lib/python3.8/dist-packages/keras/losses.py", line 1930, in binary_crossentropy
backend.binary_crossentropy(y_true, y_pred, from_logits=from_logits),
File "/usr/local/lib/python3.8/dist-packages/keras/backend.py", line 5283, in binary_crossentropy
return tf.nn.sigmoid_cross_entropy_with_logits(labels=target, logits=output)
ValueError: `logits` and `labels` must have the same shape, received ((None, 3) vs (None, 1)).
my logits (None, 3) are not the same size as my labels (None, 1)
I made sure that my last layer had an output of 3 and had the activation = 'softmax'
my labels have a dimension of (None, 1) since I mapped each class to a unique integer
The key concept you are missing is that you need to one-hot encode your labels (after assigning integers to them - see below).
So your model, after the softmax, is spitting out three values: how probable each of your labels is. E.g. it might say A is 0.6, B is 0.1, and C is 0.3. If the correct answer is C, then it needs to see that correct answer as 0, 0, 1. It can then say that its prediction for A is 0.6 - 0 = +0.6 wrong, B is 0.1 - 0 = +0.1 wrong, and C is 0.3 - 1 = -0.7 wrong.
Theoretically you can go from a string label directly to a one-hot encoding. But it seems Tensorflow needs the labels to first be encoded as integers, and then that is one-hot encoded.
https://www.tensorflow.org/api_docs/python/tf/keras/layers/CategoryEncoding#examples says to use:
tf.keras.layers.CategoryEncoding(num_tokens=3, output_mode="one_hot")
Also see https://stackoverflow.com/a/69791457/841830 (the higher-voted answer there is from 2019, so applies to TensorFlow v1 I think). And searching for "tensorflow one-hot encoding" will bring up plenty of tutorials and examples.
The issue here was indeed due to the shape of my labels not being the same as logits. Logits were of shape (3) since they contained a float for the probability of each of the three classes that I wanted to predict. Labels were originally of shape (1) since it only contained one int.
To solve this, I used one-hot encoding which turned all labels into a shape of (3) and this solved the problem. Used the keras.utils.to_categorical() function to do so.
sentiment_types = ('negative', 'neutral', 'positive')
train_df['sentiment'] = train_df['sentiment'].astype('category')
test_df['sentiment'] = test_df['sentiment'].astype('category')
# Turning labels from strings to int
train_sentiment_cat = train_df['sentiment'].cat.codes
test_sentiment_cat = test_df['sentiment'].cat.codes
# One-hot encoding
train_y = to_categorical(train_sentiment_cat)
test_y = to_categorical(test_sentiment_cat)
I have some dataset which contains different paramteres and data.head() looks like this
Applied some preprocessing and performed Feature ranking -
dataset = pd.read_csv("ML.csv",header = 0)
#Get dataset breif
print(dataset.shape)
print(dataset.isnull().sum())
#print(dataset.head())
#Data Pre-processing
data = dataset.drop('organization_id',1)
data = data.drop('status',1)
data = data.drop('city',1)
#Find median for features having NaN
median_zip, median_role_id, median_specialty_id, median_latitude, median_longitude = data['zip'].median(),data['role_id'].median(),data['specialty_id'].median(),data['latitude'].median(),data['longitude'].median()
data['zip'].fillna(median_zip, inplace=True)
data['role_id'].fillna(median_role_id, inplace=True)
data['specialty_id'].fillna(median_specialty_id, inplace=True)
data['latitude'].fillna(median_latitude, inplace=True)
data['longitude'].fillna(median_longitude, inplace=True)
#Fill YearOFExp with 0
data['years_of_experience'].fillna(0, inplace=True)
target = dataset.location_id
#Perform Recursive Feature Extraction
svm = LinearSVC()
rfe = RFE(svm, 1)
rfe = rfe.fit(data, target) #IT give convergence Warning - Normally when an optimization algorithm does not converge, it is usually because the problem is not well-conditioned, perhaps due to a poor scaling of the decision variables.
names = list(data)
print("Features sorted by their score:")
print(sorted(zip(map(lambda x: round(x, 4), rfe.ranking_), names)))
Output
Features sorted by their score:
[(1, 'location_id'), (2, 'department_id'), (3, 'latitude'), (4, 'specialty_id'), (5, 'longitude'), (6, 'zip'), (7, 'shift_id'), (8, 'user_id'), (9, 'role_id'), (10, 'open_positions'), (11, 'years_of_experience')]
From this I understand that which parameters have more importance.
Is above processing correct to understand the feature important. How can I use above information for better model training?
When I to model training it gives very high accuracy. How come it gives so high accuracy?
from sklearn.linear_model import LinearRegression
from sklearn.model_selection import train_test_split
dataset = pd.read_csv("prod_data_for_ML.csv",header = 0)
#Data Pre-processing
data = dataset.drop('location_id',1)
data = data.drop('status',1)
data = data.drop('city',1)
#Find median for features having NaN
median_zip, median_role_id, median_specialty_id, median_latitude, median_longitude = data['zip'].median(),data['role_id'].median(),data['specialty_id'].median(),data['latitude'].median(),data['longitude'].median()
data['zip'].fillna(median_zip, inplace=True)
data['role_id'].fillna(median_role_id, inplace=True)
data['specialty_id'].fillna(median_specialty_id, inplace=True)
data['latitude'].fillna(median_latitude, inplace=True)
data['longitude'].fillna(median_longitude, inplace=True)
#Fill YearOFExp with 0
data['years_of_experience'].fillna(0, inplace=True)
#Start training
labels = dataset.location_id
train1 = data
algo = LinearRegression()
x_train , x_test , y_train , y_test = train_test_split(train1 , labels , test_size = 0.20,random_state =1)
# x_train.to_csv("x_train.csv", sep=',', encoding='utf-8')
# x_test.to_csv("x_test.csv", sep=',', encoding='utf-8')
algo.fit(x_train,y_train)
algo.score(x_test,y_test)
output
0.981150074104111
from sklearn import ensemble
clf = ensemble.GradientBoostingRegressor(n_estimators = 400, max_depth = 5, min_samples_split = 2,
learning_rate = 0.1, loss = 'ls')
clf.fit(x_train, y_train)
clf.score(x_test,y_test)
Output -
0.99
Am I doing anything wrong? What ithe s correct way to build model for this sort of situati?n.
I know there is some way that I can get Precision, recall, f1 for each paramteres. Can anyone give me reference link to perform this?
So this question is about GANs.
I am trying to do a trivial example for my own proof of concept; namely, generate images of hand written digits (MNIST). While most will approach this via deep convolutional gans (dgGANs), I am just trying to achieve this via the 1D array (i.e. instead of 28x28 gray-scale pixel values, a 28*28 1d array).
This git repo features a "vanilla" gans which treats the MNIST dataset as a 1d array of 784 values. Their output values look pretty acceptable so I wanted to do something similar.
Import statements
from __future__ import print_function
import matplotlib as mpl
from matplotlib import pyplot as plt
import mxnet as mx
from mxnet import nd, gluon, autograd
from mxnet.gluon import nn, utils
import numpy as np
import os
from math import floor
from random import random
import time
from datetime import datetime
import logging
ctx = mx.gpu()
np.random.seed(3)
Hyper parameters
batch_size = 100
epochs = 100
generator_learning_rate = 0.001
discriminator_learning_rate = 0.001
beta1 = 0.5
latent_z_size = 100
Load data
mnist = mx.test_utils.get_mnist()
# convert imgs to arrays
flattened_training_data = mnist["test_data"].reshape(10000, 28*28)
define models
G = nn.Sequential()
with G.name_scope():
G.add(nn.Dense(300, activation="relu"))
G.add(nn.Dense(28 * 28, activation="tanh"))
D = nn.Sequential()
with D.name_scope():
D.add(nn.Dense(128, activation="relu"))
D.add(nn.Dense(64, activation="relu"))
D.add(nn.Dense(32, activation="relu"))
D.add(nn.Dense(2, activation="tanh"))
loss = gluon.loss.SoftmaxCrossEntropyLoss()
init stuff
G.initialize(mx.init.Normal(0.02), ctx=ctx)
D.initialize(mx.init.Normal(0.02), ctx=ctx)
trainer_G = gluon.Trainer(G.collect_params(), 'adam', {"learning_rate": generator_learning_rate, "beta1": beta1})
trainer_D = gluon.Trainer(D.collect_params(), 'adam', {"learning_rate": discriminator_learning_rate, "beta1": beta1})
metric = mx.metric.Accuracy()
dynamic plot (for juptyer notebook)
import matplotlib.pyplot as plt
import time
def dynamic_line_plt(ax, y_data, colors=['r', 'b', 'g'], labels=['Line1', 'Line2', 'Line3']):
x_data = []
y_max = 0
y_min = 0
x_min = 0
x_max = 0
for y in y_data:
x_data.append(list(range(len(y))))
if max(y) > y_max:
y_max = max(y)
if min(y) < y_min:
y_min = min(y)
if len(y) > x_max:
x_max = len(y)
ax.set_ylim(y_min, y_max)
ax.set_xlim(x_min, x_max)
if ax.lines:
for i, line in enumerate(ax.lines):
line.set_xdata(x_data[i])
line.set_ydata(y_data[i])
else:
for i in range(len(y_data)):
l = ax.plot(x_data[i], y_data[i], colors[i], label=labels[i])
ax.legend()
fig.canvas.draw()
train
stamp = datetime.now().strftime('%Y_%m_%d-%H_%M')
logging.basicConfig(level=logging.DEBUG)
# arrays to store data for plotting
loss_D = nd.array([0], ctx=ctx)
loss_G = nd.array([0], ctx=ctx)
acc_d = nd.array([0], ctx=ctx)
labels = ['Discriminator Loss', 'Generator Loss', 'Discriminator Acc.']
%matplotlib notebook
fig, ax = plt.subplots(1, 1)
ax.set_xlabel('Time')
ax.set_ylabel('Loss')
dynamic_line_plt(ax, [loss_D.asnumpy(), loss_G.asnumpy(), acc_d.asnumpy()], labels=labels)
for epoch in range(epochs):
tic = time.time()
data_iter.reset()
for i, batch in enumerate(data_iter):
####################################
# Update Disriminator: maximize log(D(x)) + log(1-D(G(z)))
####################################
# extract batch of real data
data = batch.data[0].as_in_context(ctx)
# add noise
# Produce our noisey input to the generator
latent_z = mx.nd.random_normal(0,1,shape=(batch_size, latent_z_size), ctx=ctx)
# soft and noisy labels
# real_label = mx.nd.ones((batch_size, ), ctx=ctx) * nd.random_uniform(.7, 1.2, shape=(1)).asscalar()
# fake_label = mx.nd.ones((batch_size, ), ctx=ctx) * nd.random_uniform(0, .3, shape=(1)).asscalar()
# real_label = nd.random_uniform(.7, 1.2, shape=(batch_size), ctx=ctx)
# fake_label = nd.random_uniform(0, .3, shape=(batch_size), ctx=ctx)
real_label = mx.nd.ones((batch_size, ), ctx=ctx)
fake_label = mx.nd.zeros((batch_size, ), ctx=ctx)
with autograd.record():
# train with real data
real_output = D(data)
errD_real = loss(real_output, real_label)
# train with fake data
fake = G(latent_z)
fake_output = D(fake.detach())
errD_fake = loss(fake_output, fake_label)
errD = errD_real + errD_fake
errD.backward()
trainer_D.step(batch_size)
metric.update([real_label, ], [real_output,])
metric.update([fake_label, ], [fake_output,])
####################################
# Update Generator: maximize log(D(G(z)))
####################################
with autograd.record():
output = D(fake)
errG = loss(output, real_label)
errG.backward()
trainer_G.step(batch_size)
####
# Plot Loss
####
# append new data to arrays
loss_D = nd.concat(loss_D, nd.mean(errD), dim=0)
loss_G = nd.concat(loss_G, nd.mean(errG), dim=0)
name, acc = metric.get()
acc_d = nd.concat(acc_d, nd.array([acc], ctx=ctx), dim=0)
# plot array
dynamic_line_plt(ax, [loss_D.asnumpy(), loss_G.asnumpy(), acc_d.asnumpy()], labels=labels)
name, acc = metric.get()
metric.reset()
logging.info('Binary training acc at epoch %d: %s=%f' % (epoch, name, acc))
logging.info('time: %f' % (time.time() - tic))
output
img = G(mx.nd.random_normal(0,1,shape=(100, latent_z_size), ctx=ctx))[0].reshape((28, 28))
plt.imshow(img.asnumpy(),cmap='gray')
plt.show()
Now this doesn't get nearly as good as the repo's example from above. Although fairly similar.
Thus I was wondering if you could take a look and figure out why:
the colors are inverted
why the results are sub par
I have been fiddling around with this trying a lot of various things to improve the results (I will list this in a second), but for the MNIST dataset this really shouldn't be needed.
Things I have tried (and I have also tried a host of combinations):
increasing the generator network
increasing the discriminator network
using soft labeling
using noisy labeling
batch norm after every layer in the generator
batch norm of the data
normalizing all values between -1 and 1
leaky relus in the generator
drop out layers in the generator
increased learning rate of discriminator compared to generator
decreased learning rate of i compared to generator
Please let me know if you have any ideas.
1) If you look into original dataset:
training_set = mnist["train_data"].reshape(60000, 28, 28)
plt.imshow(training_set[10,:,:], cmap='gray')
you will notice that the digits are white on a black background. So, technically speaking, your results are not inversed - they match the pattern of original images you used as a real data.
If you want to invert colors for visualization purposes, you can easily do that by changing the pallete to reversed one by adding '_r' (it works for all color palletes):
plt.imshow(img.asnumpy(), cmap='gray_r')
You also can play with ranges of colors by changing vmin and vmax parameters. They control how big the difference between colors should be. By default it is calculated automatically based on provided set.
2) "Why the results are sub par" - I think this is exactly the reason why the community started to use dcGANs. To me the results in the git repo you provided are quite noisy. Surely, they are different from what you receive, and you can achieve the same quality just by changing your activation functions from tanh to sigmoid as in the example on github:
G = nn.Sequential()
with G.name_scope():
G.add(nn.Dense(300, activation="relu"))
G.add(nn.Dense(28 * 28, activation="sigmoid"))
D = nn.Sequential()
with D.name_scope():
D.add(nn.Dense(128, activation="relu"))
D.add(nn.Dense(64, activation="relu"))
D.add(nn.Dense(32, activation="relu"))
D.add(nn.Dense(2, activation="sigmoid"))
Sigmoid never goes below zero and it works better in this scenario. Here is a sample picture I get if I train updated model for 30 epochs (the rest of the hyperparameters are same).
If you decide to explore dcGAN to get even better results, take a look here - https://mxnet.incubator.apache.org/tutorials/unsupervised_learning/gan.html It is a well explained tutorial on how to build dcGAN with Mxnet and Gluon. By using dcGAN you will get way better results than that.
Im new to Tensorflow&ML and following this example:
https://www.tensorflow.org/get_started/tflearn
It works very well until change hidden_units parameter here:
classifier = tf.contrib.learn.DNNClassifier(feature_columns=feature_columns,
hidden_units=[10, 20, 10],
n_classes=3,
model_dir="/tmp/iris_model")
When i try anything, for example hidden_units = [20, 40, 20] or hidden_units = [20] it throws an error.
I tried to find out on my own but unsuccessfully so far and thought someone here can help.
The question is how to chose a number of hidden layers for DNN Classifier and why two my examples above do not work?
Here is a full code:
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import urllib
import tensorflow as tf
import numpy as np
IRIS_TRAINING = "iris_training.csv"
IRIS_TRAINING_URL = "http://download.tensorflow.org/data/iris_training.csv"
IRIS_TEST = "iris_test.csv"
IRIS_TEST_URL = "http://download.tensorflow.org/data/iris_test.csv"
if not os.path.exists(IRIS_TRAINING):
raw = urllib.request.urlopen(IRIS_TRAINING_URL).read()
with open(IRIS_TRAINING,'wb') as f:
f.write(raw)
if not os.path.exists(IRIS_TEST):
raw = urllib.request.urlopen(IRIS_TEST_URL).read()
with open(IRIS_TEST,'wb') as f:
f.write(raw)
# Load datasets.
training_set = tf.contrib.learn.datasets.base.load_csv_with_header(
filename=IRIS_TRAINING,
target_dtype=np.int,
features_dtype=np.float32)
test_set = tf.contrib.learn.datasets.base.load_csv_with_header(
filename=IRIS_TEST,
target_dtype=np.int,
features_dtype=np.float32)
# Specify that all features have real-value data
feature_columns = [tf.contrib.layers.real_valued_column("", dimension=4)]
# Build 3 layer DNN with 10, 20, 10 units respectively.
classifier = tf.contrib.learn.DNNClassifier(feature_columns=feature_columns,
hidden_units=[10, 20, 10],
n_classes=3,
model_dir="/tmp/iris_model")
# Define the training inputs
def get_train_inputs():
x = tf.constant(training_set.data)
y = tf.constant(training_set.target)
return x, y
# Fit model.
classifier.fit(input_fn=get_train_inputs, steps=2000)
# Define the test inputs
def get_test_inputs():
x = tf.constant(test_set.data)
y = tf.constant(test_set.target)
return x, y
# Evaluate accuracy.
accuracy_score = classifier.evaluate(input_fn=get_test_inputs,
steps=1)["accuracy"]
print("\nTest Accuracy: {0:f}\n".format(accuracy_score))
Found it,
if model_dir is not specified than moel works just fine with new hidden_units